<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Leif Ulstrup on Medium]]></title>
        <description><![CDATA[Stories by Leif Ulstrup on Medium]]></description>
        <link>https://medium.com/@lulstrup?source=rss-ee9ed3dae4c2------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*nX1Yl6Khm8dQv-h6.jpg</url>
            <title>Stories by Leif Ulstrup on Medium</title>
            <link>https://medium.com/@lulstrup?source=rss-ee9ed3dae4c2------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 11:39:46 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@lulstrup/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[From Redlines to Rubrics: How to Get Great Work Out of AI Agents Without Lowering Your Standards]]></title>
            <link>https://lulstrup.medium.com/from-redlines-to-rubrics-how-to-get-good-work-out-of-ai-agents-without-lowering-your-standards-52da354c4945?source=rss-ee9ed3dae4c2------2</link>
            <guid isPermaLink="false">https://medium.com/p/52da354c4945</guid>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[professional-development]]></category>
            <category><![CDATA[skills-development]]></category>
            <category><![CDATA[management]]></category>
            <dc:creator><![CDATA[Leif Ulstrup]]></dc:creator>
            <pubDate>Thu, 25 Jun 2026 14:17:39 GMT</pubDate>
            <atom:updated>2026-10-20T22:18:54.768Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3laveD_-kI-Nz-wtJEPIaQ.png" /></figure><p><em>The shift in one view: </em><strong><em>from marking up drafts in red ink to writing down the standard the agent has to meet.</em></strong></p><p><strong>The executives most worried about AI slop are the best equipped to prevent it.</strong><em> </em>The skill they need is one they already half-possess — they’ve just never written it down.</p><ul><li><em>Have you ever read an AI-generated document that looked polished and trusted it less the more you read?</em></li><li><em>Have you ever watched two respected colleagues review the same draft and mark up completely different things?</em></li><li><em>Have you ever wondered whether leaning on these tools will dull the judgment that got you where you are?</em></li></ul><p>If so, this article is for you — and your skepticism is not the obstacle. It’s the qualification.</p><p>In <a href="https://lulstrup.medium.com/how-i-work-with-ai-agents-and-why-the-method-compounds-360c75ed0f76">my last article</a>, I described the working method I currently use for most of my AI agent projects. Since then, an observation from my own projects has kept getting confirmed: the piece that determines whether agentic AI delivers quality work is the written standard the agent must meet before the work is “done.” Chat never forced this on us — you read each response, and you steered in real time. The moment you exit chat and hand a full agent an hour of autonomous work, that informal steering disappears, and <strong>the question “how will I know it’s good?” has to be answered before the work starts, not after</strong>.</p><p>That answer matters most to the people who haven’t started using Agentic AI tools yet — the senior professionals watching colleagues paste AI output into documents and wondering whether the whole thing is a quality disaster in slow motion.</p><p>Some of it is. Here is how to make sure yours isn’t.</p><h3><strong>Your worries are correct — and that’s the point</strong></h3><p>The two fears I hear most from senior people are specific and well-founded.</p><p>First, hallucinations: the tools state falsehoods with the same confident tone they state facts. Second, slop: fluent, generic output that reads fine and says little — and is now flooding inboxes everywhere.</p><p><strong>Notice what both failures have in common. They survive only where nobody applies a real standard</strong>. A fabricated citation dies the moment someone checks sources. Generic filler dies the moment someone asks, “What does this paragraph actually claim?” <strong>Hallucinations and slop are quality-control problems before they are anything else, and quality control is the thing experienced professionals know how to do.</strong></p><p>That’s why your hesitation is worth more than your colleagues’ enthusiasm. The person who can tell good work from plausible-looking work is exactly who these tools need in charge. The challenge is that the way you’ve always enforced quality doesn’t transfer — and that’s worth understanding before you start.</p><h3><strong>The red pen doesn’t scale</strong></h3><p>Think about how quality actually gets assured in most knowledge work. You assign a deliverable. Someone drafts it. You read it and mark it up — these days in tracked changes and margin comments rather than red ink — and hand it back. A few rounds later, the work is done, not because it passed an explicit test, but because you stopped finding things to fix or ran out of steam.</p><p>The standard lives in your head. It always has. New employees learn it the way we all did — by submitting work and getting it back marked up. The red pen mostly gave way years ago to tracked changes and a column of comments in Word or Google Docs, but <strong>the ritual is unchanged: you slowly build a model of what this boss, in this organization, considers acceptable.</strong> We still call it redlining even though the pencil is long gone. The philosopher <a href="https://en.wikipedia.org/wiki/Michael_Polanyi">Michael Polanyi</a> explained why no one could simply hand you that standard in writing:</p><p><strong><em>“We can know more than we can tell.”</em></strong></p><p>You recognize strong work the way you recognize a friend’s face — instantly, and with little ability to list the features behind the judgment without more thought about the why.</p><p>That system works with people because people remember. An AI agent — Claude, Microsoft Copilot, ChatGPT, Google’s Gemini agents — starts each project knowing nothing about your standards unless you write them down. Redline an agent’s draft, and the next version often fixes what you marked while degrading something you didn’t. <strong>You end up in an endless editing loop with a partner who has infinite stamina and no memory.</strong></p><p>So the habit that protected quality for your whole career — iterate until it feels right — is precisely the habit that breaks. What replaces it is older than the technology.</p><h3><strong>Drucker saw this coming in 1954</strong></h3><p>When <a href="https://en.wikipedia.org/wiki/Peter_Drucker">Peter Drucker</a> introduced Management by Objectives, <strong>the core trade was to give up control over methods in exchange for clarity about results</strong>. Set the objective, define how success will be measured, and let the person figure out how.</p><p>Delegating to an AI agent is the same trade. You won’t supervise each step of an hour-long run — that defeats the purpose. The agent owns the methods. <strong>Everything now rides on the two things you still control: the objective and the acceptance criteria.</strong></p><p>The research on MBO carried a warning worth hearing. When Robert Rodgers and John Hunter reviewed seventy studies of MBO programs in 1991, they found productivity gains averaging 56% where senior leaders were genuinely committed — setting real objectives, using them to make decisions, and holding people to them — and just 6% where the program was treated as paperwork: goals filled in on a form and filed away.</p><p><strong><em>The framework was never the variable. The thinking was.</em></strong></p><p>The same will be true here. Everyone has access to the same models; <strong>the advantage goes to whoever does the harder work of judging the output.</strong></p><h3><strong>What teachers know that managers don’t</strong></h3><p>One profession already does, routinely, what managers of AI agents now must learn: educators. A teacher facing 80 essays cannot grade based on stamina and tacit feel. So education developed rubrics — written criteria, defined levels, and descriptions of what each level looks like. Three of their hard-won rules transfer directly:</p><p><strong>Write criteria about quality, not tasks.</strong> “The argument is supported by evidence” is a quality criterion. “Includes three sources” is a task, and an AI can satisfy the task while missing the quality you actually wanted.</p><p><strong>Score each criterion separately.</strong> A single overall grade tells the agent nothing. Separate scores tell it exactly which dimension to fix on the next pass.</p><p><strong>Expect your first rubric to be wrong.</strong> Teachers revise rubrics every term because students find the gaps. Agents find them faster.</p><p>Teachers also learned that <strong>examples beat criteria</strong>. Students calibrate fastest on annotated samples of excellent work. So do agents: two or three of your best past deliverables, with a note on what makes each one good, will lift output quality more than a page of instructions.</p><p>Writing your standards down comes with a catch, though — one I see constantly in my CliftonStrengths coaching practice, with more than 200 leaders, professionals, and MBA students since 2023. When someone reviews a draft, their redlines track their natural talents. The Analytical leader attacks the evidence chain; the one leading with Communication rewrites the narrative; the one leading with Responsibility checks every stakeholder commitment. <strong>Each is grading against a rubric they’ve never written down — and each would write a different one.</strong> That’s why <strong>the first rubric you draft deserves suspicion</strong>: the criteria that come easily tend to be the ones you already check instinctively, and the ones you need most are often the ones your profile makes invisible. The real work is deliberately adding the criteria your instincts skip.</p><h3><strong>Start as a critic, not a delegator</strong></h3><p>Here’s the on-ramp I recommend to every hesitant senior professional, because it requires trusting the AI with nothing: <strong>don’t ask it to create. Ask it to critique.</strong> You keep the pen; the AI becomes the most patient, least defensive reviewer you’ve ever had.</p><p><strong>Five ways to start building this capability</strong> this week, in rising order of trust:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*n7jtZRwFF31Y9b4se2hLUA.png" /></figure><p><strong>1. Have it stress-test your own draft.</strong> Paste in something you wrote and ask: <em>“Argue the opposite of my recommendation, as a skeptical board member would. What’s my weakest claim? What objection have I failed to anticipate?”</em> or <em>“Reverse outline this for me. What’s the central claim for each paragraph? What evidence supports each claim? Which claims are the weakest and why? Be specific.”</em></p><p>You’ll keep maybe half of what comes back. The half you keep would have cost you a bruising meeting to learn.</p><p><strong>2. Have it build your rubric with you — research first, then interview.</strong> Most people never think through how to judge the quality of their own work, and this is where the latest tools are quietly powerful: they can handle<strong> both halves of the job</strong>. Tell it: <em>“Research how experts evaluate quality in [board briefings / market analyses / proposals]. Then interview me, one question at a time, until we’ve refined that into the five to nine criteria that separate my best work from my average work.”</em> The research half surfaces dimensions you’d never list on your own — correctness of every factual claim, completeness against what the audience needs, whether the central premise is actually backed by logical and factual support rather than just asserted. The interview half tailors those to your standards. (Five to nine is deliberate — Atul Gawande’s research on surgical checklists found that short lists of “killer items” outperform exhaustive ones.) The result is your tacit judgment, externalized in an hour — a document most professionals have never possessed about their own work.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_F5erldKMFNBIn8663BQzA.png" /></figure><p><strong>3. Make sourcing non-negotiable.</strong> Whenever AI drafts or researches anything for you, require: <em>“List every factual claim with its source. Flag any claim you cannot verify with a primary source.”</em> This single instruction converts the hallucination problem from a hidden risk into a visible checklist. Anything flagged, you check or cut. The natural next step is to automate the checking itself: a reusable agent skill — a Claude Code or Cowork SKILL.md, for instance — can confirm that each external link actually resolves and that the source genuinely supports the claim, not merely that some page exists at the URL. How to build skills like that is a subject for a future article.</p><p><strong>4. When you do delegate, hand over the objective and the rubric — and require written self-grading.</strong> <em>“Draft this, then score your draft against each criterion with one sentence of reasoning. Revise whatever scores lowest. Repeat until every criterion meets the bar, then show me the scores.”</em> The work that reaches you is better, and the scorecard tells you where to aim your scarce attention.</p><p><strong>5. Keep the rubric and improve it.</strong> Every time you catch something the rubric missed, you have found a missing criterion. Add it. Unlike every junior analyst you ever marked up, the agent inherits the lesson permanently — <strong>as long as the rubric travels with the next task.</strong> Ask the agent to create a Markdown (.md) file with the updated rubric so you can take it with you to the next project or even convert it into an agentic skill (SKILL.md). <strong>This is the step that turns a one-off effort into a compounding asset.</strong></p><p>Two cautions from the field. First, agents grade their own work generously — for anything that matters, have a separate session, a different model, or you do the final grading; magazine fact-checkers were never the writers, for good reason. Second, watch for gaming: an agent optimizing your rubric will exploit its gaps, padding length where length scores points. When that happens, tighten the criterion. <strong>The rubric is a living document, not a contract.</strong></p><h3><strong>About losing your edge</strong></h3><p>The fear underneath all of this deserves a direct answer: if the AI does the work, do I atrophy?</p><p>It depends entirely on which work you hand over. Your edge was never typing speed or formatting stamina. It is judgment: knowing what the consumer of your deliverable actually needs, which argument will move this audience, and which number is wrong before you’ve checked it. The redline era spent that judgment one markup at a time, and most of it evaporated with turnover and forgetting.</p><p>Writing your standards down does the opposite. Drafting a rubric forces you to articulate what you’ve meant by “good” for twenty years, and in my experience, you come away understanding your own judgment better than you did going in. Using AI as your critic sharpens you further, because defending your work against a tireless skeptic is its own kind of practice. <strong>The professionals who will lose their edge are the ones who paste in prompts and accept whatever comes back</strong>. The ones who bring a standard will find their edge is suddenly worth more, because it’s the scarcest input in the whole system.</p><p>The accountability never moves, either. The agent doesn’t sign the work. You do.</p><p>You can delegate the task to humans or AI, but you are still personally responsible.</p><h3><strong>Start here</strong></h3><p>You already know what good looks like. The question is whether you can write it down.</p><p><em>Pick one deliverable you produce often. Draft the five criteria you’d grade it against — or better, have an AI interview you until the criteria emerge. What did you find in the rubric you’d never written?</em></p><p>I welcome feedback and your thoughts — and if you spot errors in my reasoning or sources, please note them. I am eager to fix them. And if you already keep a rubric you rely on — for a board briefing, a proposal, anything you produce often — <strong>share it in the comments</strong>. I’d like to see which criteria others decided were worth writing down, and I suspect we’d all learn more from a dozen real rubrics than from any single template. <strong>What should I dig into next?</strong></p><p>— — — — — — — —</p><p><em>Leif Ulstrup is the founder and CEO of Primehook Technology and teaches in the MBA program at the Kogod School of Business at American University. He is a certified Gallup CliftonStrengths coach. His work focuses on practical AI integration in industry and government.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=52da354c4945" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How I Work With AI Agents — and Why the Method Compounds]]></title>
            <link>https://lulstrup.medium.com/how-i-work-with-ai-agents-and-why-the-method-compounds-360c75ed0f76?source=rss-ee9ed3dae4c2------2</link>
            <guid isPermaLink="false">https://medium.com/p/360c75ed0f76</guid>
            <category><![CDATA[personal-productivity]]></category>
            <category><![CDATA[agentic-workflow]]></category>
            <category><![CDATA[continuous-improvement]]></category>
            <dc:creator><![CDATA[Leif Ulstrup]]></dc:creator>
            <pubDate>Fri, 01 May 2026 22:25:18 GMT</pubDate>
            <atom:updated>2026-06-10T12:49:31.783Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*16T9QbRiqgyGYDTspRuRsg.png" /></figure><p><em>Five files, four artifacts, and a partnership that makes both you and your work better over time.</em></p><p>If you’ve used Claude or ChatGPT for real work — not a one-off email, not a brainstorm, but a deliverable that someone else is going to read or run — you’ve probably had this experience. The first hour is electric. The second hour is fine. And somewhere in the third hour, the thing the agent ships isn’t quite right, and you can’t tell why. You go back, you re-prompt, the next version is closer, then it’s worse, then you give up and finish it by hand.</p><p>That experience isn’t about model capability. The frontier models in 2026 are extraordinary. The problem is that almost everyone is using them like a slightly more capable autocomplete — without the working memory, the quality standard, or the learning loop that any senior professional would build for a real apprentice doing serious work.</p><p>I’ve been refining a different way of working with agents — across slide decks, research memos, code projects, a capstone lecture, internal tools, and client deliverables — and the pattern is reproducible. It is also unglamorous. There are no tricks, no super-prompts, no “10x this with one prompt” claims. There is <strong>a folder, four artifacts, and a loop</strong>.</p><p><strong><em>The interesting part is what happens when you run the loop project after project</em>: </strong>the system you build around the agent gets better, your thinking gets sharper, and your output quality and speed both improve. The work compounds — and so does the worker.</p><p>Here is the method.</p><h3><strong>A note for LLM Chat -only readers</strong></h3><p>Before the method, a definition. Most readers in 2026 have used <a href="https://www.microsoft.com/en-us/microsoft-copilot/for-individuals/do-more-with-ai/general-ai/understanding-ai-agents-vs-chatbots">an LLM in chat mode</a> — type a prompt, get a response. Some have used “thinking mode” or “reasoning mode,” in which the model thinks more before answering. Both of those are <em>single-response</em> products: the model answers, then stops.</p><p>When this piece says ‘<strong>agent</strong>,’ I mean something else. An agent — Claude in <a href="https://support.claude.com/en/articles/13345190-get-started-with-claude-cowork">Cowork mode</a>, <a href="https://www.claude.com/product/claude-code">Claude Code</a>, <a href="https://openai.com/codex">OpenAI Codex</a>, Cursor, Copilot agents, Perplexity Computer — runs in a <em>loop</em>. You give it a goal and a working environment (a folder, a codebase, a browser). It reads files, <a href="https://docs.claude.com/en/docs/agents-and-tools/tool-use/overview">uses tools</a>, runs commands, edits files, fetches web pages, runs tests, evaluates whether it’s done, and if not, decides what to do next. A single agentic run might last minutes or hours.</p><p><strong>The model is not just answering — it is</strong> <em>acting</em>. <a href="https://www.anthropic.com/research/building-effective-agents">Anthropic has a clear technical explainer</a> if you want the deeper read.</p><p><strong><em>That distinction is the entire premise of what follows</em>.</strong> If you have only used chat, the experience of working with an agent is qualitatively different. The method below assumes the second.</p><h3><strong>The premise — a brilliant intern with no judgment</strong></h3><p>Treat the agent the way a senior professional treats a high-potential apprentice — or more precisely, <strong>a brilliant intern who has read everything but done almost nothing.</strong> Capable beyond expectation in raw intellectual horsepower. Naive in practical judgment. Will produce confident output that is sometimes correct, sometimes subtly wrong, and rarely able to tell the difference without help.</p><p>You stay in control. Always. You scrutinize the work, supervise the direction, set the quality bar, and curate the source material the agent draws from. The agent doesn’t replace your thinking — and the moment you let it try, the work goes sideways.</p><p>But — and this is the part most “use AI more” advice misses — you also don’t shrink your thinking to fit what the agent can do alone. You expand it. You ask the agent to do the research you’d never have time for. To explain the foundational concept you’ve cited a hundred times but couldn’t actually defend in a meeting. To challenge your reasoning. To surface ideas you wouldn’t have arrived at on your own. The agent becomes a sparring partner for your judgment, not a substitute for it.</p><p>Done right, this is mutual discovery. The agent gets better at <em>your</em> work over time. <em>You</em> get better at <em>your</em> work over time. Both can be true.</p><h3><strong>The five-file folder pattern</strong></h3><p>Every project I run with an agent starts with the same folder structure:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Nwvqmrh-rqi9YZIfGnwXMw.png" /><figcaption>The five-file folder template</figcaption></figure><p>Five files, one research folder, one <a href="https://docs.claude.com/en/docs/skills">Claude Skills</a> folder. Every project. It scales from a 30-minute task to a 30-day project. The discipline is the same.</p><p>The agent reads this folder at the start of every session. The folder is its working memory.</p><p><strong>Recommended but optional: track the folder with </strong><a href="https://git-scm.com/"><strong>git</strong></a>. Every iteration becomes a checkpoint you can return to. If a revision goes sideways, you revert to the previous good state. The agent runs the git commands for you (git init, git add ., git commit -m “…”) — you don’t need to learn the command line first. Prefer to click rather than type? Use <a href="https://desktop.github.com/">GitHub Desktop</a>. New to git entirely? See the <a href="https://github.com/leifulstrup/agentic-knowledge-work-method/blob/main/guides/05-git-and-version-control.md">Git for Non-Technical Knowledge Workers</a> guide in the repo, or skip this and add it to your second project. The method works without it. The method is <em>better</em> with it.</p><h3><strong>The four artifacts that do the work</strong></h3><p><strong>LOG.md — the action log</strong></p><p>A running, timestamped record of what we tried, what worked, what didn’t, and why we changed direction. The agent appends to this after every meaningful step. Reading the log a week later, I can reconstruct the project’s actual path — including the wrong turns. That last part matters more than it sounds. Most “lessons learned” docs filter out the wrong turns and lose the texture that makes the lessons meaningful.</p><p><strong>Research/ — the curated context</strong></p><p>A folder for source materials, not just a list of URLs. Real research papers. Verified quotes with primary-source citations. Domain-specific best practices that the agent should respect. A few of my own go-to sources I trust for this type of work dropped in alongside whatever the agent gathered. The folder is curated, not exhaustive. The agent uses these sources as ground truth — not its training data — when there’s a question of fact.</p><p><strong>RUBRIC.md — the quality gate (with iterative self-grading)</strong></p><p>A project-specific evaluation rubric that the agent uses to self-check before declaring work “done.” Two parts: big-picture first-principles tests (“does this answer the actual question we set out to answer?”) and tactical-quality tests (“are slides readable when rendered at 1280×720? does every cited stat have a primary source on the slide?”). The rubric grows during the project. When I see a failure mode, I don’t want the model to repeat that mistake, and it goes into the evaluation rubric. The agent inherits the lesson permanently.</p><p>The discipline that makes this work: <strong>I tell the agent explicitly to grade itself against the rubric, in writing, after every meaningful revision</strong> — and to keep iterating until the score crosses a threshold I set (typically 3.5 on a 0–4 scale, with no zeros allowed). The grading goes into a <a href="https://www.markdownguide.org/">Markdown</a> file in the project — RUBRIC_SCORES.md or similar — with the score for each criterion, what fell short, and the specific edit being made on the next pass. The next iteration starts from that document. Three or four loops in, you can read the score progression and see exactly where the work was weak and how it improved.</p><p>For final-pass language correctness — typos, comma splices, modifier placement, voice-preserving copy edits — I still hand the work to <a href="https://www.grammarly.com/">Grammarly</a>. Not because the agent can’t catch grammar issues, but because Grammarly is built on rules-based linguistic analysis with calibrated confidence and tends to make targeted edits that preserve voice; LLMs tend to <em>rewrite</em> when asked to <em>correct</em>. Use the right tool for the right pass.</p><p>Complex projects often have <em>multiple rubrics</em> — one for the work product itself, one for the research-source integrity, one for accessibility, one for security or compliance. Projects with software applications also get a tests/ folder with unit and integration tests the agent runs after every change; failing tests are themselves a kind of rubric, and the LOG records which tests broke, which got fixed, and which exposed an underlying design issue. The principle is the same regardless of project type: explicit, iterative, written-down self-evaluation against criteria you control.</p><p><strong>LESSONS_LEARNED.md — the deeper reflection</strong></p><p>This is where the project’s <em>insight</em> lives, separate from the action log. Lessons get captured <em>during</em> the project, not at the end. A lesson written mid-project — when I just discovered something — is richer than a lesson written in retrospect, when memory has flattened the surprise. Each entry is two or three short paragraphs: what happened, why it matters, what to do differently.</p><h3><strong>How the work actually divides</strong></h3><p>This is the part that’s worth slowing down on, because it’s where most people get the AI-collaboration frame wrong.</p><p>The questions worth asking before any handoff: Is this task one where my judgment is the value, or where the routine is the cost? Does the work need my voice on it, or just my approval of someone else’s? Will I learn something by doing this myself that I’d miss if I handed it off?</p><p>The intent is to keep your judgment cycles for the work that requires them, and to push everything else to the agent.</p><p><strong>What you ask the agent to do.</strong></p><ul><li>Routine craft that doesn’t require judgment: grammar checks, formatting consistency, font and color choices, slide layout, code style, citation formatting, table-of-contents generation, and filename hygiene. These tasks are important — work that fails on craft fails — but they don’t reward your scarce attention. Hand them off entirely.</li><li>Research the agent can do at scale that you can’t: pull current headlines from named publications across the last 60 days. Find primary sources for a quote you’ve heard a thousand times. Surface the underlying empirical literature for a framework you’re about to use. Build out the option set when you ask, “What are the three to five reasonable approaches here?”</li><li><strong>Explain things you’ve cited but don’t fully own.</strong> This is the most underused move. When the agent surfaces a concept — Markowitz portfolio theory, no-arbitrage pricing, conservation of momentum, Pareto optimality, the EPOCH framework — and you realize you know the <em>name</em> but couldn’t actually defend the <em>idea</em> in front of a sharp audience, ask the agent to teach you. Five minutes of patient explanation from a model that has read the relevant primary sources is genuinely better than what most of us got in our original coursework on the topic.</li><li>Challenge your thinking. Ask the agent to argue the opposite of your draft. To find the strongest counter-argument. To name the audience misreading you haven’t anticipated. The agent isn’t a yes-man unless you let it be.</li></ul><p>⠀<strong>What stays with you?</strong></p><ul><li>The intent. What is this project actually trying to accomplish, and for whom?</li><li>The judgment. Which of the agent’s three options is right <em>for this room</em>? Which counter-argument actually has weight? Which “lesson” is durable vs. project-specific?</li><li>The taste. Does this slide land? Is this paragraph honest? Does this code feel right when read by a senior?</li><li>The accountability. If the work is shipped and is wrong, you are accountable. The agent doesn’t sign anything.</li></ul><p>⠀The framing that holds: <strong>offload the routine, sharpen the cognitive.</strong> The agent removes the drag of craft, so your cognitive bandwidth goes to the parts of the work that actually require human judgment. Done well, you finish a project not just with the deliverable but with a deeper understanding of the topic than you had when you started. The agent didn’t think for you. It thought <em>with</em> you, faster.</p><h3><strong>The loop</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7J45eAoqNPR7DteTbG0yJA.png" /><figcaption>The Compounding Loop</figcaption></figure><p>Five stages. Stages 1–3 take 60–90 minutes upfront. Stage 4 is the bulk of the project, and iterations are <em>assumed</em>, not exceptional — a typical project runs 5–30 rounds of revisions before the rubric passes. Each round in Stage 4 has its own micro-loop: the agent produces a revision, grades it against the rubric in RUBRIC_SCORES.md, identifies the lowest-scoring criteria, proposes specific fixes, and runs again. The loop continues until the score exceeds the user-set threshold. <strong>Stage 5 is short and is what makes the method compound</strong>.</p><h3><strong>The hidden gift — meta-cognition becomes nearly free</strong></h3><p>There is a benefit to this method that isn’t obvious until you’ve done it once or twice, and it’s the part I most want you to notice.</p><p>Before agents, doing the work, keeping a meaningful log of how you did it, and distilling lessons mid-flight were prohibitively expensive. Most of us didn’t keep work logs because writing them slowed work to the point of frustration. Lessons-learned documents were written at the end, in retrospect, when the surprise’s texture had already faded.</p><p>With an agent, the log writes itself. Or rather: the agent writes it for you, in real time, after every meaningful step, in a format you can read a week or a year later and reconstruct exactly what happened. The cost of meta-cognition collapses. <strong>Continuous improvement</strong> — the kind of practice that distinguishes craftspeople from amateurs in any field — <strong>becomes a routine byproduct of the work, not a separate effort.</strong></p><p>Read your own <a href="http://log.md/">LOG.md</a> and LESSONS_LEARNED.md from a project six months later. You will find patterns you couldn’t have seen at the time — the kinds of mistakes you keep making, the kinds of insights you keep almost-having. That visibility is the gift. The agent did the bookkeeping. You can do the thinking on top.</p><h3><strong>The compounding stage</strong></h3><p>At the end of every project, I ask the agent to do one specific job: read LOG.md and LESSONS_LEARNED.md, and propose new Claude Skills — either for this project’s .claude/skills/ folder or for my system-wide skills library — that would have made the work I just did faster, more consistent, or higher quality.</p><p>The output is always specific. On the guest-lecture project I describe below, the agent surfaced 24 candidate skills. The strongest five included a <em>headline-wall builder</em> that automates a slide we built by hand three times before the pattern was visible, a <em>strengths-aware voice tailor</em> that scores a deck against a speaker’s <a href="https://www.gallup.com/cliftonstrengths/en/253715/34-cliftonstrengths-themes.aspx">CliftonStrengths</a> profile, and a <em>speaker-notes weave checker</em> that catches stale forward and backward references when slides are moved.</p><p>I implement two or three of those skills before the next similar project starts. The next project takes less time, produces fewer rounds of rework, and lands more on target. The skills library grows. So does the agent’s effective fluency in <em>my</em> work — and so does my own ability to direct it.</p><p>This is the part most “how to use AI” content misses entirely. <strong>The compounding is the point.</strong></p><h3><strong>A real example, in numbers</strong></h3><p>The most recent run of this method was a 25-minute guest lecture I prepared for the <a href="https://kogod.american.edu/learn-more/analytics">AU Kogod MS Analytics</a> students — a talk on AI and the future of analyst work. Folder structure: the five files plus a research-notes/ directory with nine source-traced documents. Iteration count: ~30. Final deck: 16 <a href="https://marp.app/">Marp</a>-rendered slides plus rendered HTML, PDF, PowerPoint, and a separate slide-by-slide rehearsal file. Verified quote pack: seven quotes with primary citations. New skill candidates mined at project end: 24.</p><p>One specific moment from that project that the method captured but a one-shot prompt would have lost: a flag emoji on one slide that rendered as broken character boxes, caught by visual review on the rendered output but invisible in the source — <em>yielding a permanent rule about render-then-review discipline</em>. The kind of catch that came not from the agent and not from me alone, but from the artifacts capturing the work clearly enough that one of us could see what the other had missed.</p><p>That is what mutual discovery looks like, in practice.</p><h3><strong>How to start</strong></h3><p>Pick one upcoming project this week. A presentation. A research memo. A code refactor. Anything with a real deliverable.</p><ol><li><strong>Make the folder.</strong> Five files, one Research subfolder. Start each file with the smallest plausible content — even just a one-line intent in <a href="http://readme.md/">README.md</a> and an empty <a href="http://log.md/">LOG.md</a> is enough to begin.</li><li><strong>Give the agent the intent — but ask it to interrogate you about it.</strong> Don’t dive in. Ask the agent to ask you clarifying questions until you both know what success looks like. Capture the answers in README.md.</li><li><strong>Run the loop. Stay engaged.</strong> Iterate. Ask the agent to log every step. Ask it to reflect when something interesting happens. <strong>And explicitly ask it to grade its own work against the rubric on every revision</strong> — write the score in a Markdown file (RUBRIC_SCORES.md), and keep iterating until the score crosses a threshold you set. <em>And ask it to teach you.</em> When a concept surfaces that you can’t fully defend, stop and have the agent explain it to you. When it produces an option you wouldn’t have considered, ask it to make the strongest case for that option, even if you’re inclined to disagree.</li><li><strong>At the end, ask it to mine for skills.</strong> Read the log and lessons together; propose what should become a reusable skill. Implement two or three. Watch the next project go faster.</li><li><strong>Optional but highly recommended for your second project: track everything with git.</strong> Ask the agent to set up version control on day one and commit a checkpoint after each iteration. You’ll thank yourself the first time a revision goes sideways, and you want the previous good state back.</li></ol><p>⠀That’s the method. There is no proprietary prompt; the discipline is the artifact. The compounding is real, but it requires the discipline to ship — and the willingness to stay engaged with the work rather than hand it over. Do that, and the system you’ve built starts feeling different by the third project. So does your own thinking.</p><p>The repo with templates, starter Claude Skills, and a fuller guide is here: <a href="https://github.com/leifulstrup/agentic-knowledge-work-method"><strong>github.com/leifulstrup/agentic-knowledge-work-method</strong></a></p><h3><strong>A small recursion</strong></h3><p>I should tell you how this article was written, because the answer itself is the strongest evidence that the method works.</p><p>I drafted it using the same five-file folder I use for client projects. The intent went into README.md. The agent helped me curate a small Research/ folder — four notes, including one on LinkedIn engagement patterns in 2026, one on Medium audience behavior, one on why Grammarly’s correctness check still outperforms LLMs, and one explaining the chat-versus-agent distinction for readers who haven’t seen the difference. We built a RUBRIC.md specific to <em>publishing on Medium and LinkedIn</em> — covering hook quality, technical-term accessibility, Medium length sweet spots, LinkedIn algorithm signals, and method integrity. The agent scored every revision against the rubric and kept iterating until the score crossed the threshold I set. We caught and fixed gaps that I would never have surfaced while reading the draft on my own — including the realization that the original draft assumed readers knew what an agent was, which is precisely the gap the explainer near the top of this piece is meant to close.</p><p>By the end of writing this article, the method had refined itself in three concrete ways: a new template for the project README, a new candidate, Claude Skill, for “publication rubric scoring,” and a sharper version of the chat-vs-agent explainer that will get reused on every future piece I write for this audience. The whole project lives in git, with a checkpoint after every iteration — five rounds, each one inspectable and revertible. The article didn’t just describe the method; it also explained the rationale. The act of writing it <em>grew</em> the method. <strong>That is what compounding looks like when the system around the agent gets better with every project — even when the project</strong> <em>is</em> <strong>writing about the method itself.</strong></p><p>_________</p><p><strong><em>Please share your experience and recommend upgrades to improve my approach.</em></strong></p><p><em>Leif Ulstrup is the founder and CEO of Primehook Technology and teaches in the MBA program at the Kogod School of Business at American University. His work focuses on practical AI integration in industry and government.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=360c75ed0f76" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What if your greatest untapped resource is your natural talent?]]></title>
            <link>https://lulstrup.medium.com/what-if-your-greatest-untapped-resource-is-your-natural-talent-c72b3a41a96b?source=rss-ee9ed3dae4c2------2</link>
            <guid isPermaLink="false">https://medium.com/p/c72b3a41a96b</guid>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[cliftonstrengths]]></category>
            <category><![CDATA[professional-development]]></category>
            <category><![CDATA[management]]></category>
            <dc:creator><![CDATA[Leif Ulstrup]]></dc:creator>
            <pubDate>Thu, 03 Oct 2024 19:07:52 GMT</pubDate>
            <atom:updated>2024-10-04T13:56:15.153Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/782/1*GSHeakv1iUGpq6U12Msd2w.jpeg" /></figure><p>Since becoming a <a href="https://www.gallup.com/cliftonstrengths/en/253850/cliftonstrengths-for-individuals.aspx">Gallup CliftonStrengths</a> <a href="https://www.gallup.com/cliftonstrengths/en/253958/become-coach.aspx">coach</a> in 2023, I’ve been fortunate to work with over 200 individuals—senior leaders, professionals, and MBA students—who have uncovered transformative insights about themselves and their teams.</p><p>One of the most rewarding aspects of this journey has been watching experienced leaders discover that their top strengths were underutilized or contributing to career satisfaction or dissatisfaction in ways they hadn’t realized. It’s a powerful moment when someone recognizes how their natural talents can unlock potential they didn’t know they had.</p><p>A recent example was a senior executive who had always excelled but felt something was missing. Through CliftonStrengths, they realized that one of their top talents, <em>Context</em>, was sidelined in their role. By consciously leveraging this strength, they redefined their approach and found more fulfillment and effectiveness in their work.</p><p>But <strong>here’s the challenge</strong>: CliftonStrengths <strong>isn’t just a one-time exercise</strong>. The true value comes when individuals continuously commit to understanding and applying these insights. It’s about recognizing patterns, evolving, and refining those talents into true strengths that drive impact.</p><p>As a coach, I’ve seen firsthand how consistent practice leads to personal growth, stronger teams, and more dynamic leaders. Working with others has also deepened my understanding of how my strengths (<em>Learner, Achiever, Responsibility, Individualization, Self-Assurance</em>) can create meaningful change.</p><p><strong>My advice for anyone looking to leverage their strengths:</strong> Be mindful of how your talents shape your perceptions, decisions, and interactions. The more aware you are of your strengths — and the strengths of those around you — the better equipped you’ll be to collaborate, lead, and achieve great things.</p><p><strong><em>What are your top natural talents, and are you using them daily?</em></strong></p><p><a href="https://www.gallup.com/cliftonstrengths/en/253676/how-cliftonstrengths-works.aspx">Gallup CliftonStrengths Overview</a></p><p><a href="https://www.gallup.com/topic/cliftonstrengths.aspx">Interviews with Leaders aware of their CliftonStrengths and those around them</a></p><p>#CliftonStrengths #Gallup #Leadership #Coaching #PersonalDevelopment #StrengthsFinder</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c72b3a41a96b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A GPT to Assist with USAspending.gov — an experiment]]></title>
            <link>https://lulstrup.medium.com/a-gpt-to-assist-with-usaspending-gov-an-experiment-00dc7b8f352d?source=rss-ee9ed3dae4c2------2</link>
            <guid isPermaLink="false">https://medium.com/p/00dc7b8f352d</guid>
            <category><![CDATA[open-data]]></category>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[market-research]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[federal-government]]></category>
            <dc:creator><![CDATA[Leif Ulstrup]]></dc:creator>
            <pubDate>Fri, 10 Nov 2023 23:38:02 GMT</pubDate>
            <atom:updated>2023-11-16T03:41:49.905Z</atom:updated>
            <content:encoded><![CDATA[<h3>A GPT to Assist with USAspending.gov — an experiment</h3><h3><strong>Embracing AI — revisiting my earlier blog posts on working with USAspending.gov and the API</strong></h3><p>It’s been awhile since I last posted a blog on using USAspending.gov. My series on the topic starts here: <a href="https://lulstrup.medium.com/gaining-an-analytics-edge-using-federal-spending-open-data-7cc69a063c68">https://lulstrup.medium.com/gaining-an-analytics-edge-using-federal-spending-open-data-7cc69a063c68</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/847/1*SAIuuHyH4XCX-MMqrnUpRQ.jpeg" /><figcaption>OpenAI GPT Store Screen Shot</figcaption></figure><p>And ends here: <a href="https://lulstrup.medium.com/blog-11-the-kipps-curve-undertanding-federal-agency-buying-preferences-74f897308baf">https://lulstrup.medium.com/blog-11-the-kipps-curve-undertanding-federal-agency-buying-preferences-74f897308baf</a></p><p>I’m <strong><em>experimenting</em></strong> with the new <strong>OpenAI GPT</strong> feature.</p><p>I’ve assembled a prompt and some documents to make it easier to write code to access the USAspending.gov API’s.</p><p>Here is the <strong>link to this AI assist GPT tool</strong>: <a href="https://chat.openai.com/g/g-VUKAHV22R-usaspending-gov-api-expert-experimental"><em>https://chat.openai.com/g/g-VUKAHV22R-usaspending-gov-api-expert-experimental</em></a></p><p>It does not include any custom functions nor any special error checking or factual validation at this point. It’s just an experiment. It does produce working Python code that uses the API interface.</p><p><strong><em>TODO: add GPT Actions via OpenAI yaml specification to enable calls directly to the USAspending.gov API to enabled testing of the interface and exploration of data. Two completed as of 11/15/2023. Try:</em></strong></p><ul><li>Call the api.usaspending.gov API with the pscAutocompletePost operation</li><li>Call the api.usaspending.gov API with the fundingAgencyAutocompletePost operation</li></ul><p>It will also create training aids and answer questions about using USAspending.gov.</p><p><strong>I welcome any comments</strong> on how well it works and what issues arose when you used it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=00dc7b8f352d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Love open-source software? What about open-source golf lessons?]]></title>
            <link>https://lulstrup.medium.com/love-open-source-software-what-about-open-source-golf-lessons-9c73a676d6f8?source=rss-ee9ed3dae4c2------2</link>
            <guid isPermaLink="false">https://medium.com/p/9c73a676d6f8</guid>
            <category><![CDATA[golf-tips]]></category>
            <category><![CDATA[recreation-and-sports]]></category>
            <category><![CDATA[personal-development]]></category>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[learning]]></category>
            <dc:creator><![CDATA[Leif Ulstrup]]></dc:creator>
            <pubDate>Fri, 30 Sep 2022 20:19:44 GMT</pubDate>
            <atom:updated>2025-05-04T17:18:48.615Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Are you intrigued to learn to play golf but fear you will embarrass yourself or that you are too old to learn?</em></p><p>The pandemic has <a href="https://www.forbes.com/sites/erikmatuszewski/2021/12/28/golf-continued-to-boom-in-2021-from-play-to-purchases/?sh=631b638c3c33">boosted interest in golf</a>. I regularly meet people in business who would like to take it up again or start playing golf but aren’t sure they have the skills or the time to become competent enough to enjoy the sport and don’t even start. Others fear the many unknowns around the game, from the etiquette norms to being paired with a much better player and holding them up. I know many people in business who miss great opportunities to play in scramble format charity events for fear of the sport. They miss significant opportunities for fun and networking.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TzVxyz-wQz_9A-qjjPg5iw.png" /><figcaption>Credit — OpenAI DALL-E “golfer watching YouTube video golf lessons in style of picasso”</figcaption></figure><p>My Dad picked up golf in his forties and fell in love with the game. My brother and I joined him for rounds, but at the time had other hobbies as teenagers that we found more dynamic and exciting (tennis, skateboarding, Boogie boarding). Our Dad loved playing and often would zoom over to our local golf course to walk nine holes after Sunday football games ended. I didn’t understand his enthusiasm for the game then, but I do now — context matters.</p><p>In my early forties, I realized why he was so enthusiastic about the game and began playing regularly with my friends. We started our golf quest slowly by walking nine holes at a local course early on the weekends before our kids and wives were up. Gradually we started tackling eighteen holes.</p><p>As I played more, I was paired with others. I soon realized most people <a href="https://golf.com/news/how-your-handicap-index-compares/">are not super amazing golfers.</a> Even the better players struggle with errant shots, as you will, though they don’t tend to compound their problems as many novices do.</p><p>I love to learn and have watched many sources of online instruction and paid for professional coaching. I can recommend a few open-source YouTube channels that I’ve found valuable.</p><p>Before you begin watching those, I recommend reading and framing your approach to learning golf with this: “The Inner Game of Golf” by W. Timothy Gallwey — https://theinnergame.com/inner-game-books/inner-game-of-golf/</p><p>The YouTube video sources in order, they are:</p><p><strong>Danny Maude </strong>— <a href="https://www.youtube.com/c/DanielMaude">https://www.youtube.com/c/DanielMaude</a></p><p><em>Danny Maude publishes a new video every Saturday morning. Sign up for the alerts.</em></p><ul><li><a href="https://twitter.com/dannymaude">@dannymaude</a></li><li><a href="https://www.dannymaude.com/">https://www.dannymaude.com/</a></li></ul><p><strong>Andrew Emery</strong> -</p><ul><li><a href="https://www.youtube.com/@AndrewEmeryGolf">https://www.youtube.com/@AndrewEmeryGolf</a></li><li><a href="https://andrewemerygolf.com/">https://andrewemerygolf.com/</a></li></ul><p><strong>Me and My Golf</strong> — <a href="https://www.youtube.com/c/meandmygolf">https://www.youtube.com/c/meandmygolf</a></p><p><strong>AlexElliottGolf</strong>-</p><ul><li><a href="https://www.youtube.com/@AlexElliottGolf">https://www.youtube.com/@AlexElliottGolf</a></li></ul><p><strong>Scratch Golf Academy </strong>— <a href="https://www.youtube.com/c/Scratchgolfacademyonline">https://www.youtube.com/c/Scratchgolfacademyonline</a></p><p><strong>Shawn Clement’s Wisdom in Golf Lessons </strong>— <a href="https://www.youtube.com/c/ShawnClementWisdominGolf">https://www.youtube.com/c/ShawnClementWisdominGolf</a></p><ul><li><a href="https://twitter.com/shawnclement">https://twitter.com/shawnclement</a></li><li><a href="https://wisdomingolf.com/">https://wisdomingolf.com/</a></li></ul><p><em>Shawn Clement publishes one to two videos per week</em></p><p><strong>Erika Larkin’s “True Swing” — </strong><a href="https://www.youtube.com/channel/UC8mb9jkIteXUVnpuqnmNREg">https://www.youtube.com/channel/UC8mb9jkIteXUVnpuqnmNREg</a></p><ul><li><a href="https://www.youtube.com/watch?v=csOv3AoEwH4">https://www.youtube.com/watch?v=csOv3AoEwH4</a></li><li><a href="https://www.youtube.com/watch?v=nBBpriIPuzQ">https://www.youtube.com/watch?v=nBBpriIPuzQ</a></li><li><a href="https://www.youtube.com/watch?v=nBBpriIPuzQ">https://www.youtube.com/watch?v=nBBpriIPuzQ</a></li><li>IG: <a href="https://www.instagram.com/erikalarkingolf/">https://www.instagram.com/erikalarkingolf/</a></li><li><a href="https://www.erikalarkin.com/a-true-swing-book">https://www.erikalarkin.com/a-true-swing-book</a></li></ul><p><strong>Padraig Harrington</strong></p><ul><li><a href="https://www.youtube.com/c/PadraigHarringtonOfficial">https://www.youtube.com/c/PadraigHarringtonOfficial</a></li><li><strong>Introduction to Achieving a Single Figure Handicap</strong> | Paddy’s Golf Tips | Padraig Harrington</li><li><a href="https://youtu.be/2qQwo_eVAkI?si=k9GwzNd9CDtZLz2B">https://youtu.be/2qQwo_eVAkI?si=k9GwzNd9CDtZLz2B</a></li></ul><p><strong>DrKwonGolf</strong></p><ul><li><a href="https://www.youtube.com/channel/UCwCWfCapZNK_ulayYNTBJxQ">https://www.youtube.com/channel/UCwCWfCapZNK_ulayYNTBJxQ</a></li><li>Rope Swing Part I on Dr Kown Golf — <a href="https://youtu.be/0hyrcov8lz0">https://youtu.be/0hyrcov8lz0</a></li><li>Rope Swing Part II on Dr. Kown Golf — <a href="https://youtu.be/4GdZJ-acILg">https://youtu.be/4GdZJ-acILg</a></li><li><em>Dr. Kwon’s videos are a bit different from those above. He is a biomechanics professor. His perspective and the novel techniques he uses to help people develop dynamic motions may change your mind about how you think a golf swing works. His research informs some of the instruction that Maude and Clement uses. Maude and Clement rely on your intuitive, athletic skills in teaching golf which is less complex than other approaches.</em></li></ul><p><strong>Why did I have a grounder or blop shot with my Driver?</strong></p><ul><li><a href="https://www.youtube.com/c/AliTaylorGolf">AliTaylorGolf</a> has an easy fix — <a href="https://www.youtube.com/watch?v=x_DNY8c6KDw">https://www.youtube.com/watch?v=x_DNY8c6KDw</a></li></ul><p>Here is another great video by <a href="https://www.youtube.com/c/AliTaylorGolf">AliTaylorGolf</a> describing how to think about using the low point of the club to achieve maximum performance with various clubs — <a href="https://www.youtube.com/watch?v=WFGpxL76OmA">https://www.youtube.com/watch?v=WFGpxL76OmA</a></p><p><strong>Etiquette tips:</strong></p><ul><li><a href="https://www.pga.com/story/golf-for-beginners-golf-etiquette-rules-and-glossary">https://www.pga.com/story/golf-for-beginners-golf-etiquette-rules-and-glossary</a></li><li><a href="https://www.golfdigest.com/story/arniesrules">https://www.golfdigest.com/story/arniesrules</a> (note that #1 is “Don’t be the slowest player”)</li></ul><p>Like fly-fishing, golf is a great sport to enjoy the outdoors, forget the stresses of work, spend time with family and friends, meet new people, and continue learning and growing. It enables one to remove distractions from their mind and focus on the task at hand.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*egUZAjKHdeIWQ6L6B48_pg.png" /><figcaption>Credit OpenAI DALL-E</figcaption></figure><p>One of the great things about golf is that you are playing against yourself and on a quest to learn and improve. Start with a few lessons with the pro at your local golf course and rent or buy some used clubs (see <a href="https://www.2ndswing.com)./">https://www.2ndswing.com).</a> Even the PGA pros have full-time coaches, so you don’t feel bad if you’ve not mastered the sport in a few weeks, months, or years.</p><p>Once you’ve started to play some, you will find the YouTube channels above valuable. Find a driving range where you can practice, keep some clubs in your trunk, and test new ideas after watching the latest Maude or Clement video.</p><p>Our Dad’s main advice when playing golf was not to hold others up. Be prepared to take your shot when it is your turn. That advice has been invaluable to enjoying the game — especially early on. He also loved walking the course, which was excellent advice and one I follow today unless the golf course insists we use carts.</p><p>Enjoy the journey and satisfaction of developing new skills at any age.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8Caglj2cr78wE3AoX_zy8A.png" /><figcaption>Credit OpenAI DALL-E</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9c73a676d6f8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Professional Services First Principles — Rediscovering David Maister]]></title>
            <link>https://lulstrup.medium.com/professional-services-first-principles-rediscovering-david-maister-e2d6e097a90d?source=rss-ee9ed3dae4c2------2</link>
            <guid isPermaLink="false">https://medium.com/p/e2d6e097a90d</guid>
            <dc:creator><![CDATA[Leif Ulstrup]]></dc:creator>
            <pubDate>Thu, 08 Jul 2021 17:43:59 GMT</pubDate>
            <atom:updated>2021-07-08T17:59:52.632Z</atom:updated>
            <content:encoded><![CDATA[<h3>Professional Services First Principles — Rediscovering David Maister</h3><p><strong>Audience:</strong> Leaders in professional services firms.</p><p><strong>Premise:</strong> Professional Services firms have more in common than differences, and understanding those commonalities can help leaders adopt best practices and differentiate.</p><p><em>Have you ever wondered how similar or different the challenges you face as a leader of a professional services organization are from your peers at competitors?</em></p><p><em>Have you ever struggled with where you should focus on gaining a sustainable competitive advantage and aligning the activities of your professionals around those actions?</em></p><p><em>Have you ever found your organization’s professionals debating which metrics to use to assess the health of your business?</em></p><p>Former Harvard Business School (HBS) Professor <a href="https://en.wikipedia.org/wiki/David_Maister">David Maister</a> published a book in 1993 that helped me answer questions like those above when I was new to professional services. That book is <a href="https://www.amazon.com/Managing-Professional-Service-David-Maister/dp/0684834316/ref=sr_1_1?dchild=1&amp;keywords=David+H.+Maister&amp;qid=1625765953&amp;s=audible&amp;sr=1-1-catcorr">“Managing the Professional Services Firm.”</a></p><p>My degree is in mechanical engineering, and most of my work was software development and managing projects with capable technical talent. What I had learned up to that point about managing a professional services business was through informal coaching, word of mouth, trial and error, self-study, and osmosis. As an engineer, I looked for business <a href="https://en.wikipedia.org/wiki/First_principle">“first principles”</a> to guide me but could not find them. I had the aha moment as I began reading David Maister’s recently published book (updated several times since).</p><p>At the time, I assumed a general manager role responsible for a profit center with a portfolio of projects. I was working with clients that needed a combination of management, technology, and organizational change consulting. The practice that I was building needed a wide variety of skills, more strategic thinking, and quickly adapting to market changes as our projects became more complex.</p><p>I wanted to learn as much as possible as quickly as possible about how the best firms in professional services operate. Maister’s book was there to help.</p><p>I am a voracious reader of business and technology-related books and articles. Former colleagues and friends know how annoying I can be in sending them articles and books. It is rare to find business books that contain truths that transcend the context of the time they are published. Many popular business books encapsulate the current fad or tell the story of a then-popular business success story. Often, those firms have lost their luster a few years after the book appears.</p><p>Maister’s work is different. He set out to identify and explain the patterns common to all professional services businesses (e.g., law firms, architecture firms, management consulting firms, medical practices, etc.). He describes his insights by focusing on <a href="https://en.wikipedia.org/wiki/First_principle">“first principles.”</a> He also documents best practices he observed and developed for the challenges that all professional services firms face as they grow and compete in the market.</p><p>More recently, I transitioned from leading and managing profit centers to advising senior business leaders on strategy, operations, and emerging technology. I also have an opportunity once a year to teach a course to first-year MBAs on the role of information and information technology (IT) in shaping business strategy, customer and employee experience, innovation, and operations. I am always searching for case studies and research that illustrate enduring principles and describe cause and effect relationships in models that the students can use in their careers. Those are very hard to find for most topics. The research and writing of Drucker, Maister, Porter, Schein, Kotter, and Christensen are examples of management “first principles” thinkers that I’ve found valuable in my role as a profit center leader. Deep thinkers and clear communicators who have uncovered enduring ideas.</p><p>In working with clients who are the leaders of professional services firms, I find myself drawing upon the work of David Maister. These include his other books published after “Managing the Professional Services Firm” and his excellent website (<a href="https://davidmaister.com/">https://davidmaister.com</a>), which provides a broad array of ideas from his experience as a business advisor. Maister’s work is specific to the unique context of professional services and is as enduring as those <a href="https://en.wikipedia.org/wiki/First_principle">“first principles”</a> so clearly illustrated in the writing of Peter Drucker.</p><p>Whether you are new to professional services or an experienced executive, Maister’s work in books and his website has gems worth revisiting from time to time. It’s not a typical business book that you might read from beginning to end. Instead, each chapter explores an important topic and provides both ways to analyze your current situation and formulate a plan of action to address the challenges you face.</p><p>Much like the ancient Greek philosophers or writers like Shakespeare who uncovered profound truths about human nature that are as applicable today, Maister’s work is as valuable today as when it was published.</p><p>I encourage you to explore his <a href="https://davidmaister.com/">website</a> and his books.</p><p>One of my other favorites by David Maister is <a href="https://www.amazon.com/True-Professionalism-Courage-People-Clients-ebook/dp/B000FC0VV6/ref=tmm_kin_swatch_0?_encoding=UTF8&amp;qid=1625765914&amp;sr=8-13">“True Professionalism”.</a></p><p>Are there other books on professional services you think have enduring value and should be in my Kindle library? If so, please share those in the comments below.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e2d6e097a90d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Blog #11 The  “Kipps Curve” — Undertanding Federal Agency Buying Preferences]]></title>
            <link>https://lulstrup.medium.com/blog-11-the-kipps-curve-undertanding-federal-agency-buying-preferences-74f897308baf?source=rss-ee9ed3dae4c2------2</link>
            <guid isPermaLink="false">https://medium.com/p/74f897308baf</guid>
            <category><![CDATA[open-data]]></category>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[strategy]]></category>
            <category><![CDATA[federal-government]]></category>
            <dc:creator><![CDATA[Leif Ulstrup]]></dc:creator>
            <pubDate>Wed, 03 Mar 2021 02:26:54 GMT</pubDate>
            <atom:updated>2021-03-03T23:02:42.109Z</atom:updated>
            <content:encoded><![CDATA[<h3>Blog #11 The “Kipps Curve” — Understanding Federal Agency Buying Preferences</h3><p><strong>TL;DR </strong>— Uncover competitive insights from information and analytics using Federal contracting #opendata. Analyze &gt; 100GB’s of US Federal spending open data with a few lines of Python code.</p><p><strong><em>Previous Blog Post</em></strong> in this series can be found at this <a href="https://lulstrup.medium.com/blog-10-federal-contracting-market-share-trends-and-patterns-eaaf4af9c4d6">link.</a></p><h3>Introduction</h3><p><a href="https://www.usaspending.gov/">USAspending.gov </a>is a valuable resource to inform business strategy and investment decisions for firms focused on the Federal contracting market. In my previous posts, I’ve highlighted how to wrangle the tens of gigabytes of Federal obligation transactions to mine valuable insights about the market.</p><p>This post will focus on understanding Department buying patterns for IT-related services. One of the key insights from analyzing Department and Agency-level spending from this perspective is to highlight whether there are biases toward firms of a certain scale.</p><h3>The “Kipps Curve”</h3><p>This analysis is inspired by investment banker and Federal M&amp;A guru <a href="https://kippsdesanto.com/bob-kipps/"><strong>Bob Kipps</strong></a> of <a href="https://kippsdesanto.com/">Kipps-DeSanto</a> (now a Capital One company). In a recent conversation, Bob noted how many segments of the Federal contracting market are heavily weighted toward smaller businesses and the largest businesses; thus, making it increasingly difficult for companies in the middle scale that neither have the advantages of the set-aside designators nor the scale or brand names of the largest firms they compete against.</p><p>The analysis I present here is an effort to understand what that looks like in practice using <a href="https://www.usaspending.gov/">USAspending.gov</a> data to review the current state of the market in the government fiscal year 2020 (GFY2020) that ended September 30, 2020.</p><p>For simplicity, I focus my attention on the product or service codes where the dominate type of work has been designated as IT and Telecom Services (also known as the <strong>“D” PSC_Cat </strong>codes). That was a <strong>$49B</strong> market in GFY2020 and has been growing faster than the overall Federal contracting market.</p><p>It is especially hard for successful businesses that graduate from the protections of set-aside programs when they enter “into the wild” of full and open competition against the largest firms in the market. The analysis below makes it even starker how big a challenge there is for firms between $200m/year to $1B/year to compete and scale.</p><h3>Analysis</h3><p><strong>Source: </strong><a href="https://www.usaspending.gov/download_center/award_data_archive">https://www.usaspending.gov/download_center/award_data_archive</a></p><p><strong>GFY 2020</strong>, Archive download 02/07/2021</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ox9KbpF9kucsN1YAnqPasg.png" /></figure><h3>Big Picture Context</h3><p>In GFY220, the Federal Government obligated ~<strong>$666B</strong> in funds across <strong>~85k</strong> contractors (at the parent firm level if multi-division or multiple independent units).</p><p>The curves that follow show the cumulative percentage of the total obligations by the estimated prime-contractor federal total revenue.</p><p>The x-axis plots the size of the prime contractor as the sum of all obligations they were issued across all work as a prime contractor. They are ordered from smallest to largest. The y-axis is the cumulative percent of spending for a specific product or service category work (PSC_Cat) as designated by the contracting agency. Each dot on the chart represents a contractor and plots them by ordered size along the x-axis and by their cumulative % of the total along the y-axis. The largest Federal contractor on the right completes the trail of dots to reach 100% of all obligations in that category. This first figure displays all Federal obligations for all goods and services in blue (~85k contractor-dots). The orange dots represent obligations tagged as PSC_Cat “D” for IT and Telecom Services (~6k contractor-dots).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/714/1*5kWPbbgGPHc3A8mr3mVNQA.png" /><figcaption>Figure 1. PSC_Cat D IT and Telecom Services Spending vs Overall Obligations.</figcaption></figure><h4><em>Why do the orange and blue lines diverge around the 35% of market share mark?</em></h4><p>One can see that the IT and Telecom Services segment follows a pattern more skewed to the smaller firms in the market though that is relative since the ticks on the x-axis are $5B/year!</p><p>The upper right dot is an overlay of a blue and orange dot representing the largest contractor. If you look to the left of the blue dot representing the second largest contractor, you can see that the largest contractor has greater than 10% market share (100% vs ~89% =&gt; ~11%). Each jump in altitude from one dot to the next represents the incremental market share of that firm. One can see from the blue dots that the largest firm has much larger market share than #2. In the case of the orange dots that represent PSC_Cat “D” IT and Telecom spending, one can see that the largest contractor has relatively little market share but there is a large jump from #3 to #2.</p><h4><strong>Let’s look at a much smaller portion of the graph above.</strong></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/709/1*7DbwbAL23foN0hPyUKOcTQ.png" /><figcaption>Figure 2. IT and Telecom Services vs Overall Market — Smallest firms to $200M/year in size</figcaption></figure><p>Even with this magnification of the chart in Figure 1, one can see that IT and Telecom Services spending is skewed toward the smaller firms vs the overall Federal market.</p><p>This next chart magnifies the segment below $10B. 30% of the IT and Telecom Services market is spent on firms $100M/year or smaller. And, ~38% of the spending for IT and Telecom is spent on firm less than $200M/year in size. ~50% of the spending goes to firms greater than $1B/year in size. A small number of firms between $1B-$1.5B in size capture about 12% of the market while firms between $200M-$1B capture about 12% also.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/714/1*7wwXsO_lTBq_8zuZ9AjxrA.png" /><figcaption>Figure 3. IT and Telecom Services vs Overall Market — Smallest firms to $10B/year in size</figcaption></figure><h3>Advantages of Scale</h3><p>The two charts highlight bifurcation of the market between the firms &lt;$1B/year and those &gt;$1B/year in scale. The bar chart plots the average obligations per prime contractor in each bin vs the contractor size bins on the x-axis.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/726/1*nVA5tTUMRo0H47u_knPSLA.png" /><figcaption>Figure 4. IT and Telecom Services — Average Obligations Per Contractor by Contractor Size Bin</figcaption></figure><p>The number above the bars represent the number of companies with prime contract obligations in that size bin. Note the large jump in average obligations per contractor from the $200m-$1B bin to the &gt;$1B bins.</p><h4>3 Times the Spending and 1/3 the Number of Competitors</h4><p>104 contractors are competing for 12% of the market in the $200M-$1B segment while 33 contractors are competing for 30% of the market in the $1B-$5B segment. The average IT and Telecom obligations per contractor in the $5B-$10B segment is almost two times the previous category with only a 1/4 the number of competitors.</p><p>Note the scaling of the reduced number of competitors at each stage. The $25-$100M segment has 1/10 the competitors of the &lt;$25M segment. The $100M-$1B segment has about 40% of the previous segment. The $1B-$5B segment has 33/(95+104) ~ 16% of the competition.</p><p>The next diagram displays the same information for all obligations of all types of products and services across all types of work (~$666B in GFY2020).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/739/1*sB2CDUdAokMtcNxhBZdRUw.png" /><figcaption>Figure 5. All Products and Services — Average Obligations Per Contractor by Contractor Size Bin</figcaption></figure><p>The next tables describe the distribution of PSC_Cat “D” IT and Telecom spending across Departments that spend more than $50M/year on that type of work. The top line <strong>“*ALL”</strong> computes the distribution of PSC_Cat “D” obligations across all Federal Departments and every spending level. It provides a <strong>reference point</strong> to compare the other Department percentages.</p><h4>Market Share by Contractor Size by Department</h4><p><em>(filtered with &gt; $50M/year in IT and Telecom Obligations)</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/766/1*7iC8CVlf9XZW5e7fLZ3omw.png" /></figure><h4>Competitor Count By Contractor Size by Department</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/749/1*cinTUQF67tTaEzrW9mrwvw.png" /></figure><h4>GFY2020 IT and Telecom Service (PSC_Cat “D”) Total Obligations</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/501/1*sKWdjiQWcMxS6dckE1iNvA.png" /></figure><h3>Conclusion</h3><p>The graphs and tables show that the IT and Telecom Services (PSC_Cat “D”) sub-market within the overall Federal contracting market skews toward firms below $200M/year and prime contractors greater than $1B/year in size as Bob Kipps noted.</p><p>The Department level spending distribution table highlights different preferences by Departments in their choices and willingness to select firms of different size to meet their IT needs. For instance, the SBA spends ~94% of its ~$257M/year in IT on prime contractors less than $100M/year in size (77 companies). Which makes sense given their mission. In contrast, the VA spend ~40% of its ~$4.8B/year in IT obligations on firms between $1B-$5B/year (16 companies).</p><p>A logical next step is to take this analysis to the Agency level inside the largest Departments to better understand the preferences at that level and the competitive intensity of those sub-segments of the market.</p><p><strong><em>What are the implications for the firms in that $200M/year-$1B/year range and those $100M/year and below firms hoping to thrive as they graduate “into the wild” of full and open competition?</em></strong></p><p>I welcome feedback and your thooughts. Also, please note any errors that I’ve made. I am eager to fix them. <strong>What should I look at next?</strong></p><p><strong><em>Previous Blog Post</em></strong> in this series can be found at this <a href="https://lulstrup.medium.com/blog-9-spotting-emerging-i-t-adoption-trends-in-the-federal-market-b257255ea299">link</a>.</p><p><strong>Appendix:</strong></p><p><a href="https://opensource.org/licenses/MIT"><em>MIT Open Source License</em></a><em> Copyright 2020 Leif C Ulstrup</em></p><p><em>When working with USAspending.gov data, see this note embedded on the USAspending.gov site about attribution of D&amp;B data and “D&amp;B Open Data” that is embedded in the download data and USAspending.gov website reports such as the one below — </em><a href="https://www.usaspending.gov/db_info"><em>https://www.usaspending.gov/db_info</em></a></p><p><strong>Steps Followed</strong></p><p>Step 1: Download the GFY2020 Archive (<a href="https://lulstrup.medium.com/3-automated-download-of-usaspending-gov-gfy-archives-e6c71b3eb7d1">see earlier blog on that topic</a>)</p><p>Step 2: Read the CSV files into Python Pandas via Dask and apply data wrangling cleanup (<a href="https://lulstrup.medium.com/4-usaspending-gov-analytics-loading-data-into-dask-and-pandas-3e83af441901">see earlier blog on this topic</a>)</p><p>Step 3: Filter the transactions to identify parent companies with prime obligations in product_or_services_code (“PSC_Cat”) that start with “D” for IT and Telecom services</p><p>Step 4: Compute the Federal-wide total prime obligations by parent firm identified in the previous step</p><p>Step 5: Narrow the analysis to a specific Department/Agency/PSC_Cat</p><p>Step 6: Compute total obligations for that Department/Agency/PSC_Cat by Contractor</p><p>Step 7: Sort Contractors by Total Federal Prime Obligations</p><p>Step 8: Analyze and Plot</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=74f897308baf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Blog #10 Federal Contracting Market Share Trends and Patterns]]></title>
            <link>https://lulstrup.medium.com/blog-10-federal-contracting-market-share-trends-and-patterns-eaaf4af9c4d6?source=rss-ee9ed3dae4c2------2</link>
            <guid isPermaLink="false">https://medium.com/p/eaaf4af9c4d6</guid>
            <category><![CDATA[python-pandas]]></category>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[strategy]]></category>
            <category><![CDATA[federal]]></category>
            <category><![CDATA[open-data]]></category>
            <dc:creator><![CDATA[Leif Ulstrup]]></dc:creator>
            <pubDate>Tue, 08 Dec 2020 17:54:46 GMT</pubDate>
            <atom:updated>2021-01-01T14:12:34.407Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>TL;DR </strong>— Uncover competitive insights from information and analytics using Federal contracting #opendata. Analyze &gt; 100GB’s of US Federal spending open data with a few lines of Python code.</p><p><strong><em>Previous Blog Post</em></strong> in this series can be found at this <a href="https://lulstrup.medium.com/blog-9-spotting-emerging-i-t-adoption-trends-in-the-federal-market-b257255ea299">link</a>.</p><p>Python code and the jupyter notebook for this analysis is on <a href="https://github.com/leifulstrup/USAspending_Medium_Blog_Analytics_Series/blob/master/Blog_10y_Market_Share_Analytics_read_parquet_github.ipynb">Github here</a>.</p><h3>Introduction</h3><p>Several of the most popular and influential <a href="https://hbr.org/">HBR</a> articles on <a href="https://hbr.org/2007/09/demystifying-strategy-the-what">business strategy</a> explore the relationship between market share, market growth, and profitability. In this tenth blog in my series, I show you how you can use Federal spending open data (USAspending.gov) to help guide your analysis of the market and shape your business strategy.</p><p>For most companies competing in the Federal contracting market (“Federal market”), the ~$600B/year in annual spending seems like an almost infinite amount relative to their annual revenue and a market with unlimited growth potential. With the exception of weapons systems platforms, most of the aggregated Federal market is highly-fragmented, relative to most commercial markets. However, at the Agency level, market share is more concentrated and it is challenging to dislodge incumbents that have solid past performance records. Competition for that work at re-compete time (~ every 5 years) can be intense. Incumbents have huge advantages in the Federal market. Also, most services contracts are long term (typically, one year with four option years) and <a href="https://www.marketconnectionsinc.com/wp-content/uploads/2018/10/Gov-Con_White-Paper_FINAL.pdf">incumbent recompete win rates over 75% are not uncommon</a>.</p><p>Capturing market share for your firm’s services is often a long game and it can be expensive to displace incumbents in terms of bid and proposal to capture the work and managing profitability after the win.</p><p>Many Federal contractors thrive on a steady proposal machine fueled by a healthy pipeline of qualified opportunities.</p><p><strong><em>Is it possible to make better capture investment decisions using insights about analytical features in the market the reveal competitive intensity and Agency-level buyer preferences?</em></strong></p><p><strong><em>Are there ways to use Federal #OpenData and analytical techniques to guide your planning efforts to yield higher ROI with a more focused strategy?</em></strong></p><h3>Natural Limits to Market Share?</h3><p>For some of the largest segments of the Federal contracting market, there seem to be boundaries on market share by product or service category that reflect market competition, customer preferences, contracting practices, and other factors. Exploring those forces at the Federal, Department, and Agency for the segment of products or services your firm concentrates on can help in shaping your investment plans for new opportunities and retention of incumbent work.</p><p>Spending by Federal designated <a href="https://www.acquisition.gov/psc-manual">product_or_service code</a> for GFY2019 (<strong><em>~$590B</em></strong> total prime obligations):</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/943/1*YZfLiegkPp7a4f5ZwVVy5w.png" /></figure><ul><li><a href="https://www.acquisition.gov/psc-manual">PSC_Cat = product_or_service_code</a> which is the dominant type of work within the prime contract.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/651/1*IZTLQNKhc6OykSggVON0yA.png" /></figure><p>Selecting a subset of these codes (A, B, D, M, Q, R, S) that are focused more on services than products, you can see the GFY10-GFY19 market share trends for the #1 company each year in that PSC_Cat.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/843/1*-uOmonbp7Y318DJDxi6EHQ.png" /><figcaption>Market share for #1 Company in each PSC_Cat: GFY2010–2019.</figcaption></figure><p><strong><em>Growing market concentration at the top:</em></strong> The top market share firm in PSC_Cat Q Medical Services has jumped significantly (50% increase— from 20% to 30%) between GFY17 and GFY19 over the ~20% baseline earlier in the decade. PSC_Cat B Special Studies has doubled from ~5% to ~10%. PSC_Cat M Operation of Government-Owned Facility has increased from ~12% to %16. Those categories represent ~<strong><em>$54B</em></strong>/year and <strong><em>~9%</em></strong> of total Federal obligations.</p><p><strong>Stable or declining market concentration:</strong> The #1 contractor in PSC_Cat A Research and Development has dropped from ~18% market share to ~13% over the decade while PSC Cat D IT and Telecom remained relatively stable at ~6% (with a significant fluctuation between GFY13-GFY17 of 2x) at the start and end of the decade. PSC Cat R Support Services has seen a slight decline from ~4% to ~3.8%. For GFY 2019 PSC_Cat A + D + R =&gt; <strong><em>~$177B</em></strong> which is <strong><em>~31%</em></strong> of total prime contracting.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/719/1*2Fw4vLoP0X9ftDHPehgLDA.png" /><figcaption>Clustering of Market Share for the #1 Contractor each GFY by $ Value and Share Fraction — select PSC_Cat</figcaption></figure><p>Note how the PSC_Cat market share values are clustered by type of work. It is possible to capture 20%-30% of the spending in Medical Services (PSC_Cat Q), a &gt; $20B/year market, but the #1 contractors in other major spending categories are only able to capture about 5%-10% of the spending in their work segment.</p><h4>Exploring PSC_Cat D (IT &amp; Telecom), R (Support Svcs), and A (R&amp;D)</h4><p>Let’s dive into the large PSC_Cat D, R, and A categories and explore those further.</p><p><em>Top Contractor (“Top_1”) in a GFY and PSC_Cat</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/532/1*oCpww8qXQKrmgqsMoW1G4g.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/692/1*jpb__9s9ulYUEsNBPqyikQ.png" /><figcaption>Tracking the #1 Contractor in each of these PSC_Categories from GFY-GFY20YTD — Note the clustering on the left and trends on the right.</figcaption></figure><p><em>Top_5 Contractors Total Obligations in a GFY and PSC_Cat</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/534/1*xSpsyzkyn-khYPLCtukYYw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/707/1*_Eu8dOqFD9m72E79HYP4iQ.png" /><figcaption>Tracking the Market Share Total for the Top 5 Contractors in each of these PSC_Categories from GFY-GFY20YTD</figcaption></figure><p>The charts on the left show a clear clustering of market share values by PSC_Cat. The charts on the right show trends in those values over time. They have either become stable or declined over the last decade. For reference, overall spending from GFY10-GFY19 followed this pattern:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/598/1*CJWn7vdNFX9r_uo_YHf_vQ.png" /></figure><p>Although the last decade has seen significant restructuring and M&amp;A of the contractors in the PSC_Cat D and R segments of work with major growth in PSC_Cat D (IT &amp; Telecom) spending, the market share for the #1 and top 5 contractors in any one year has remained relatively stable. Note that being #1 in GFY2019 for PSC_Cat D is worth $1B/year in additional revenue but only 0.4 basis points increase in market share vs GFY10. The PSC_Cat R #1 contractor in GFY19 saw a decline of ~$0.5B in revenue vs the #1 in GFY10.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/862/1*xuZTe08O3yMiPExhOSE2RA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/596/1*-FvE1hAAsZ9QMS3RRLBftA.png" /><figcaption>Federal-wide Obligations relative to GFY2010 for PSC_Cat D and R from GFY10-GFY19</figcaption></figure><p>Note that the #1 and top 1–5 contractors may be different each year though in most cases it does not change rapidly with the exception of M&amp;A activity and strategic sales of divisions. This next graph shows how much change there has been the PSC_Cat D leadership of Top 10 firms from GFY10–19 through M&amp;A and organic growth:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/937/1*YFdvSx4daXptGwbbg8NCDg.png" /></figure><p><em>Link to interactive version of the chart above: </em><a href="https://leifulstrup.github.io/Interactive_Medium_HTML/"><em>https://leifulstrup.github.io/Interactive_Medium_HTML/</em></a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/514/1*5_7Usa-eYPnMjRtAmbutbg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/478/1*hTIyNKnwe-Q-NUqzMNi9Ww.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/691/1*csNvv-7yx_t8olmQTOHZBQ.png" /></figure><h3>Both Fragmented and Concentrated</h3><p>At the Federal-wide market level, the #1 firm in the IT and Telecommunication Services segment did not exceed 8% market share between GFY2010 (<em>GFY: Government Fiscal Year — Oct 1 through Sep 30</em>)and GFY2019. That was a ~$45B market in GFY2019. For the second-largest category of services spending, “Support Services (Professional, Management, Admin)”, ~$85B/year, the #1 competitor between GFY10–19 reached a high of ~4.4% in GFY10.</p><p>$85B x 4.4% ($3.7B) is a big number for sure, but the market is fragmented as seen as a whole.</p><p>The picture of market share fragmentation can change quickly when one drills down to view spending at the Department (e.g., US Department of Agriculture), Agency (e.g., Defense Information Systems Agency of DoD), or Office level. At the level of an Agency or Administration, it is not unusual for the top contractor to have a much higher percentage of spending in that category — a 20%-40% market share range for #1 is not atypical as seen in the histograms of #1 contractor market share by Agency (from GFY10-GFY19) below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/605/1*zTBsmg0uOQtFs_3Ccayoaw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/593/1*wGfgtK02JNl_gDQXlDdqcQ.png" /><figcaption>Left: ≥ $50M/year in IT and Telecom Services Spending, Right: ≥$10M/year and &lt;$50M/year (Red Line: Median Market Share of #1 Contractor in PSC_Cat D) [GFY10-GFY19 with multiple years for each agency]</figcaption></figure><p>Starting from an existing prime contract with a Federal customer in an Office, Agency, or Department, one can explore the relative attractiveness of capturing more local market share for like-work (e.g., IT Services) at the same Agency, like-work at other Agencies in the same Department, or venturing into other types of work (e.g., PSC_Cat R) in the same or a different agency. The relative odds of success and return on investment (ROI) depend on the competitive intensity that surrounds a firm’s existing work and the target of opportunity. That ROI is also affected by the relative growth rate of spending for work in that category and the contracting type preferences of the customer (e.g., time and materials, fixed price, cost-plus, etc.).</p><h4>Department Level Analysis: DoD</h4><p>Let’s explore the data at the Department level with the largest contracting Department in the Federal Government — the Department of Defense (DoD). Here is a snippet of the Python pandas code (see this <a href="https://github.com/leifulstrup/USAspending_Medium_Blog_Analytics_Series">Github repo</a> and <a href="https://github.com/leifulstrup/USAspending_Medium_Blog_Analytics_Series/blob/master/Blog_10y_Market_Share_Analytics_read_parquet_github.ipynb">jupyter notebook for blog #10</a>):</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*D0Kt-DjRR2VAe7pGc3E4cw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/542/1*PDjt-6k2dao-wgehfOhjbg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/705/1*uSK2JAOoMKl-KitOl1dqLA.png" /><figcaption>DOD Top Contractor (#1) Market Share for PSC_Cat A, D, and R (*GFY20 is YTD and incomplete)</figcaption></figure><p>PSC Cat D #1 contractor market share at DoD reached a high of ~12% just before the overall market bottom in GFY15 and has now settled into ~6%. PSC_Cat R has remained relatively stable fluctuating around 6%. PSC_Cat A (R&amp;D) fluctuates significantly relative to the others. PSC_Cat A was a $48B market in GFY19. PSC_Cat R includes a wide range of services and many smaller contracts while PSC_Cat A includes very large and small contracts. A deeper dive is warranted to understand these segments at the next levels of granularity (specific lower-level codes with the PSC_Cat, NAICS of the contractor, named-programs, and contracts).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2uEmqLYxeE9G7Dh_rGODCg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/552/1*e4kYaQN6Gz5JaV1mZA2baQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/709/1*qqVug3fhU7urjRvpzfHRhA.png" /><figcaption>DOD Top 5 Contractors’ Market Share for PSC_Cat A, D, and R (*GFY20 is YTD and incomplete)</figcaption></figure><p>Though the top 5 contractors&#39; total market share is relatively stable for R and A. It is interesting to note that for PSC_Cat A, the #1 contractor share is growing, yet the Top 5 is relatively stable. GFY20 final figures will provide a complete picture of that seeming trend.</p><p>Another interesting ratio to examine is the relative power of being #1 in your segment. For instance, the ratio of the Top 5 market share/#1 contractor market share is a clear indicator of the value in reaching that #1 spot in your segment (<em>lower is better</em>).</p><p><em>For GFY19, DoD Top 5/#1:</em></p><p><strong>PSC_Cat D ~25%/6% =&gt; ~4</strong></p><p><strong>PSC_Cat R ~19%/5% =&gt; ~4</strong></p><p><strong>PSC_Cat A ~ 36%/16% =&gt; ~2</strong></p><h4><strong>Agency/Administration Level:</strong> DoD Defense Information Systems Agency (DISA) and Social Security Administration (SSA)</h4><p><strong><em>DoD DISA</em></strong></p><p><em>PSC_Cat A is excluded since DISA nor SSA has much spending in the Research and Development category.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/539/1*3ihO8q1UfYnbOiZDqHAmSw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/705/1*3aojPLkzqvNBiV-W5dT3qA.png" /><figcaption>Top Contractor Market Share for DoD DISA for PSC_Cat D and R work</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/531/1*4VFZf-on5DeAVxQQWA23QQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/699/1*q3GmLddUzbI9O2vhY-zp5w.png" /><figcaption>Top 5 Contractors at DISA for PSC_Cat D and R work — Market Share Trends</figcaption></figure><h4>Social Security Administration (SSA)</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/536/1*d51Bz-sOtyqIzAEce-8JQg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/697/1*ZMiWbKnM5dXxstkRFoEliw.png" /><figcaption>Top Contractor at SSA Market Share for PSC_Cat D and R work (*GFY20 is YTD incomplete)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/529/1*G3gryqCKBtpHQpr7kx51EQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/704/1*8M9jkXaqrJ7xxwpb2cAPGw.png" /><figcaption>Top 5 Contractors Market Share for PSC-Cat D and R work at SSA (*GFY20 is YTD and incomplete)</figcaption></figure><p>It is interesting to compare PSC_Cat D market share patterns between DISA and SSA. For GFY19, DISA’s #1 contractor had ~18% of a ~$3B market (~$0.5B) while SSA’s #1 contractor had ~35% of a $0.7B market (~$0.25B). For the top 5, DISA had ~33% of $3B (~$1B) and SSA ~70% of $0.7B (~$0.5B).</p><p>Five (5) contractors at SSA capture 70% of PSC_Cat D spending while it is less than half that for the top 5 competitors at DISA. Not only is the DISA market ~5x larger, but it also is not nearly as concentrated. <strong><em>Which segment of the market is more likely to generate the higher profit margins and volume of profit for a particular market share rank?</em></strong></p><p><strong>DISA Top 5/#1 = ~1.9 and SSA Top 5/#1 = ~2</strong></p><p>The lower that ratio, the more value captured by being #1 in that segment and, thus, the importance of retaining that #1 position.<strong><em> How might this be a predictor of competitor behavior when new entrants appear and others compete for this work at contract refresh points?</em></strong></p><p>Understanding the relative differences in the market dynamics at the Agency level is an important context for opportunity investment decisions. Knowing this context will help you within an Agency that you already serve and want to capture more market share. It is especially useful when you are venturing into new Agency territory and have less experience and insight on buyer behavior and competitive forces.</p><h3>Conclusion</h3><p><strong><em>What can be learned from Federal contracting market share analysis that can inform business strategy?</em></strong></p><p>Although the analysis above is a high-level view of the overall market, it demonstrates that the choice of market share analytics (not just information such as obligation growth trends) to assess market segment attractiveness can be extracted from the USAspending.gov open data. <em>Potential insights</em>:</p><p>(A) Understanding Agency-level market share dynamics reveal whether there is a major advantage to being #1 or in the top 5 at an Agency.</p><p>(B) Different PSC_Cat segments of the market can have very different implications in terms of future market share and volume of prime obligation capture (revenue and profit potential) potential.</p><p>(C ) The stability of contractor market share rankings (not explored above but in the data source) over time can reveal customer preferences. Their preference for a broad or narrow contractor base and how open they are to new entrants.</p><p><strong><em>What do you see in this series of analytics that I missed?</em></strong></p><h3>Coming Attractions</h3><p>I will explore more analytical and market topics in future blog posts — including more market-share trend analysis and <a href="https://en.wikipedia.org/wiki/Natural_language_processing">NLP</a> techniques using USAspending.gov and beta.SAM.gov award and opportunity textual data and mashups with other open data sources.</p><p><strong><em>Previous Blog Post</em></strong> in this series can be found at this <a href="https://lulstrup.medium.com/blog-9-spotting-emerging-i-t-adoption-trends-in-the-federal-market-b257255ea299">link</a>.</p><p>The Python and pandas code used in the examples above and more can be found here: <a href="https://github.com/leifulstrup/USAspending_Medium_Blog_Analytics_Series/blob/master/Blog_10y_Market_Share_Analytics_read_parquet_github.ipynb">https://github.com/leifulstrup/USAspending_Medium_Blog_Analytics_Series/blob/master/Blog_10y_Market_Share_Analytics_read_parquet_github.ipynb</a></p><p>Blog series Github repo: <a href="https://github.com/leifulstrup/USAspending_Medium_Blog_Analytics_Series">https://github.com/leifulstrup/USAspending_Medium_Blog_Analytics_Series</a></p><p><a href="https://opensource.org/licenses/MIT"><em>MIT Open Source License</em></a><em> Copyright 2020 Leif C Ulstrup</em></p><p><em>When working with USApsneding.gov data, see this note embedded on the USAspending.gov site about attribution of D&amp;B data and “D&amp;B Open Data” that is embedded in the download data and USAspending.gov website reports such as the one below — </em><a href="https://www.usaspending.gov/db_info"><em>https://www.usaspending.gov/db_info</em></a></p><h3>References:</h3><p>(1) <a href="https://hbr.org/2011/12/the-charts-that-changed-the-world">https://hbr.org/2011/12/the-charts-that-changed-the-world</a></p><p>(2) <a href="https://hbr.org/1975/01/market-share-a-key-to-profitability">https://hbr.org/1975/01/market-share-a-key-to-profitability</a></p><p>(3) <a href="https://hbr.org/1985/03/building-strategy-on-the-experience-curve">https://hbr.org/1985/03/building-strategy-on-the-experience-curve</a></p><p>(4) <a href="https://hbr.org/2015/06/navigating-the-dozens-of-different-strategy-options">https://hbr.org/2015/06/navigating-the-dozens-of-different-strategy-options</a></p><p>(5) <a href="https://store.hbr.org/product/the-five-competitive-forces-that-shape-strategy-hbr-bestseller/r0801e?sku=R0801E-PDF-ENG&amp;Ntt=Michael%2520Porter&amp;referral=00269&amp;cm_sp=endeca-_-spotlight-_-link">The Five Competitive Forces That Shape Strategy</a></p><p>(6) <a href="https://www.economist.com/news/2009/09/14/the-experience-curve">https://www.economist.com/news/2009/09/14/the-experience-curve</a></p><p>(7) <a href="https://hbr.org/2007/09/demystifying-strategy-the-what">https://hbr.org/2007/09/demystifying-strategy-the-what</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=eaaf4af9c4d6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Blog #9 — Spotting Emerging I.T. Adoption Trends in the Federal Market]]></title>
            <link>https://lulstrup.medium.com/blog-9-spotting-emerging-i-t-adoption-trends-in-the-federal-market-b257255ea299?source=rss-ee9ed3dae4c2------2</link>
            <guid isPermaLink="false">https://medium.com/p/b257255ea299</guid>
            <category><![CDATA[open-data]]></category>
            <category><![CDATA[federal]]></category>
            <category><![CDATA[altair]]></category>
            <category><![CDATA[python-pandas]]></category>
            <category><![CDATA[analytics]]></category>
            <dc:creator><![CDATA[Leif Ulstrup]]></dc:creator>
            <pubDate>Mon, 23 Nov 2020 17:33:43 GMT</pubDate>
            <atom:updated>2020-11-23T17:33:43.151Z</atom:updated>
            <content:encoded><![CDATA[<h3>Blog #9 — Spotting Emerging I.T. Adoption Trends in the Federal Market</h3><h3>Gaining an Analytics Edge Using Federal Spending Open Data Series</h3><h3>Context</h3><p><strong>Premise of this Blog Series</strong>: Publicly available “<strong>open data</strong>” sources and open-source technology analytics can provide information to reduce uncertainty when making critical business investment decisions.</p><p><strong>Purpose</strong>: Demonstrate to <strong>executives</strong> how easy it is to access and analyze free, open-data sources and motivate them to expect more insight and context from their team when making crucial go/no-go <strong>investment decisions</strong> on bids and new offerings.</p><p><strong>Scope</strong>: US Federal Contracting Market.</p><p><strong>TL;DR </strong>— Analyze GB’s of US Federal spending data with a few lines of Python code. Uncover competitive insights from information and analytics.</p><p><strong><em>Previous Blog Post</em></strong> in this series can be found at this <a href="https://lulstrup.medium.com/8-usaspending-gov-8263a5d0f3f3">link</a>.</p><h3>Introduction</h3><p>Many professional services companies in the Federal market have successfully expanded their business by riding an information technology (I.T.) adoption wave. Whether your company is a system integrator, technology specialists, business consulting, or staff augmentation firms, there are growth and profit opportunities if you can gain a competitive advantage before services around a growing technology are commoditized. Those advantages include:</p><ul><li>a trusted partnership with the technology vendor,</li><li>opportunities to develop mission and function-specific add-ons,</li><li>work plan estimation and risk management insights,</li><li>the attraction of talent,</li><li>marketing attention for early wins with customers,</li><li>valuable past performance, and</li><li>an opportunity to provide thought leadership in influencing prospective customer adoption of the technology.</li></ul><p><strong>How does one spot these growth trends in the Federal market (and declines of older technologies) before the broader market catches on?</strong></p><p>Many larger firms in the federal market subscribe to I.T. industry analysts such as <a href="https://www.gartner.com/en">Gartner</a>, <a href="https://go.forrester.com/">Forrester</a>, and <a href="https://www.idc.com/getdoc.jsp?containerId=IDC_P38441">IDC</a>. Those services provide excellent insights on emerging technologies, the relative strengths of technology companies in a specific segment (e.g., <a href="https://www.gartner.com/en/research/methodologies/magic-quadrants-research">Gartner’s famous Magic Quadran</a>t), and analysis of how commercial organizations are adopting the technologies. They are excellent resources. These services can help develop an inventory of technology types and specific companies to focus your investment strategy.</p><p>The purpose of this blog is to demonstrate ways you can use Federal open data sources and open-source software to look for broad market signals and explore Agency-specific buying patterns around emerging technology.</p><h3>Let’s get started with the Analysis</h3><p>We will use <a href="https://www.usaspending.gov/">USAspending.gov open data</a> for this analysis (we will use <a href="https://beta.sam.gov/reports/awards/standard">beta.SAM.gov opportunity and award archive</a> text with <a href="https://en.wikipedia.org/wiki/Natural_language_processing">NLP</a> tools in a future blog).</p><p>In <a href="https://lulstrup.medium.com/gaining-an-analytics-edge-using-federal-spending-open-data-b91b517f2c04">previous blogs in this series</a>, I’ve explained how to download USAspending.gov archives.</p><p><strong><em>The Python code for this post can be found at this link -— </em></strong><a href="https://github.com/leifulstrup/USAspending_Medium_Blog_Analytics_Series/blob/master/Blog_9b_Simple_Ways_to_Identify_Federal_Technology_Buying_Trends_github.ipynb"><em>https://github.com/leifulstrup/USAspending_Medium_Blog_Analytics_Series/blob/master/Blog_9b_Simple_Ways_to_Identify_Federal_Technology_Buying_Trends_github.ipynb</em></a><em>.</em></p><p>I will restrict my analysis to the Government Fiscal Year (GFY) 2016–2020 period (October 2015-September 2020). When writing this, the GFY2020 data is close to complete but is still missing data for the last ~60 days from DoD. <em>Keep that in mind, when you interpret the charts that follow.</em> By the time you replicate this analysis, you may have the complete GFY2020 files (they should be near complete by mid-January 2021) for a more thorough examination of the trends.</p><p><em>See this note embedded on the USAspending.gov site about attribution of D&amp;B data and “D&amp;B Open Data” that is embedded in the download data and USAspending.gov website reports such as the one below — </em><a href="https://www.usaspending.gov/db_info"><em>https://www.usaspending.gov/db_info</em></a></p><p>My first data processing step is to load USAspending.gov data from GFY2016–2020 into my <a href="https://jupyter.org/">jupyter notebook</a>, where I will use Python and several open-source add-on packages to analyze the data. I will restrict the fields (aka columns) read into memory to a fraction (~14) of the total fields available from each source record (~280). Doing this will speed processing and reduce the memory needed. The GFY2016-GFY2020 data fits into ~3.3GB of memory in a Python pandas dataframe.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/950/1*t4UEPZztka6cJIZCL4QsLA.png" /></figure><p>Next, read in the USAspending.gov data:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LUnRw5X1D1gyDU3Ywf4tHA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YNguIo_Gna0kvSTZObME-Q.png" /></figure><p>It has ~27 millions obligation action records and uses ~3.3GB of RAM.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/581/1*fPaUFQ4MfN_X7kl61ySf5g.png" /></figure><p>The next step is to build a master list of popular I.T. software products and brand names, including emerging technology companies. For ease in this example, I will read a list of top 2020 software products (aka platforms) from a market research website,</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dqGj3NnKYE3D54qmd3LbnQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/368/1*1Iak6AIvN9tozoEOwSkppQ.png" /></figure><p>I supplement that with brands and technologies that I track. Hopefully, many of those will overlap.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/673/1*h8DkPwvA8zsx_-GlWL9cbg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/639/1*FbIvU5jMm2yL6_Ks3C8dqg.png" /></figure><p>Once you have a search terms list, we can use lightweight text searching to find matching text records in the award_description field. This technique is a crude but simple approach, and there is significant room to improve this (future blog discussion).</p><p>This simple approach will include noise since it is not uncommon for technology names to have multiple meanings in different contexts. For instance, the term “zoom” refers to the video conferencing technology and appears in the name of other technologies and can be a verb or adjective describing a feature of different requirements or systems. This approach will help us with a broad sweep to spot trends and require additional work to refine and validate those trends. I’ll demonstrate the use of NLP tools to improve these searches in future blog posts.</p><p>I’ll start this process with a brute-force approach and elementary text searching and add more complexity to refine the analysis.</p><p>I’ve restricted the search to Federal product_or_service _codes that start with the I.T. products category ‘70’, “D” (I.T. and Telecomm services and solutions), and “R” (professional services).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/784/1*20v6P0teCXJvpybtzUZXWg.png" /></figure><p>This subset is ~4.3M records of ~27.4M records in the complete GFY2016-GFY2020 data source. The code below cycles through the list of I.T. brands and product names and assembles a pandas dataframe to look for trends at a maco-level.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5BQuMvgeHKEeBS_z_iskNA.png" /></figure><p>The dataframe include spending associated with that search term and counts of records that cite the search term. These are gross estimates. Also, I split up some products by the same company, such as Microsoft’s Azure cloud platform and Amazon’s AWS or “Amazon Web Services,” to see citations specific to those offerings. For a more rigorous analysis, I need to recombine those under the parent company’s name. Here is an example of that:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZWyFsi8TASEMaXw1m5stnQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*w5ks3kMVMpgOYX2y4EAVFg.png" /></figure><h3>Convert to Pivot Table Format for Easier Trend Spotting</h3><p>The next step is to create pivot table versions of the records to identify trends in mentions (‘count’) and spending (‘federal_action_obligation’) over time and compute some year-over-year (YoY) ratios. Viewing the records in this way makes it easy to sort the data and also plot trends. The spending amounts listed are not one-for-one with the products. It only means those obligation dollars are associated with that term. It might be the case where multiple technologies appear in the text (e.g., Splunk hosted on AWS), and I have allocated the same spending to both, which oversates the amounts. The purpose of this stage of analysis is to paint a broad picture to motivate a detailed analysis.</p><p>The spending analysis gives one a sense of the market’s magnitude, but the count of references to the product or brand may be an even better clue of picking up signals on buying trends. For instance, a sort of the 2019_vs_2019_ratio of counts for various search terms highlights references to technologies growing in mentions relative to a small base for most of them (note: the “999.00” is artificial where it would be infinite otherwise). These might be early signals of emerging interest in these technologies.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GfuisItYHr3SbrPKAS8-4w.png" /><figcaption>Trends in Count of References to the Technology Name</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jYUXWYfPihWP8ZXuSd-Ylg.png" /><figcaption>Trends in Obligations Associated with References to the Technology Name</figcaption></figure><p>This table shows some of the faster growing reference counts for the larger brands (at least 100 references) —<strong><em> note that GFY20 is incomplete</em></strong>:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IUmU2u6_oqLkc4bdOV7qyQ.png" /></figure><p>This table shows GFY2019 vs GFY2018 declines in references to the larger, established technologies (greater than 100 references):</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1mhqQLtjQkMDjPPJX58ozw.png" /></figure><p>The declining references <strong><em>may be erroneous</em></strong> due to the multiple ways that a company or its products are referenced (e.g., Amazon vs Amazon Web Services vs AWS). More thorough analysis is needed to fully understand the trends. Also, a moving-average may make more sense since a large acquisition may not be replicated in the next. These references include services contract references also to these technologies and are NOT exclusively software (SW) purchases. Analysis of beta.SAM.gov award and opportunity text and attachments is needed to fully understand the trends.</p><h3>Exploration of the Growing RPA Segment</h3><p>One of the hot growth areas in enterprise software is <a href="https://en.wikipedia.org/wiki/Robotic_process_automation">Robotic Process Automation (RPA)</a>. You can see firms such as <a href="https://www.uipath.com/rpa/robotic-process-automation">UIPath</a>, <a href="https://www.blueprism.com/">Blue Prism</a>, <a href="https://www.automationanywhere.com/">Automation Anywhere</a>, <a href="https://www.workfusion.com/">WorkFusion</a>, and <a href="https://www.pega.com/">Pegasystems</a> growing in citations.</p><p>It is notable that the fourth member of the Gartner Magic Quadrant for RPA, <a href="https://www.workfusion.com/">Workfusion</a>, is missing from the list even though it is part of the search terms list.</p><p><em>Is this an opportunity for a partnership with a Federal systems integrator to introduce them to the market?</em></p><p><em>Maybe the Federal market is not part of WorkFusion’s priorities like it is for their other Gartner Magic Quadrant competitors.</em></p><p>This chart shows the growing spending associated with these RPA firms:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/570/1*QY7VJLmU93om0dFTVcJwLQ.png" /></figure><p>Here is a table that shows which Agencies are adopting these RPA tools and how much they are spending in conjunction with references to the tools:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/726/1*V-QqJC0bUbgYS_WHpQd4mA.png" /></figure><p>Here is a view of RPA tool adoption by Agency:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/797/1*vbgnxfdaW4GYd3dg0qA2gw.png" /></figure><p>There are other tools in the RPA category. Expanding the list may uncover more growth opportunities for this popular segment of the enterprise I.T. market. You can products and company names of interest using the code from this example as a starting point for your analysis.</p><p>Though this approach only provides a rough estimate of the trends for the most popular information technology brands and products, it gives an analyst some ‘signal’ to explore in-depth.</p><p>Also, there is ambiguity in the names of technology companies/products with other related concepts (e.g., Stripe can refer to the company and many other I.T. associated products such as magnetic stripe readers). One needs to fully use the <a href="https://www.acquisition.gov/psc-manual">Federal product_or_service_code system</a> to narrow the search further and hope that everything is coded correctly.</p><p>In a future blog, I will use NLP tools to explore the beta.SAM.gov opportunity and awards archive to see what else we can learn about how the government describes technologies in award and opportunity announcements to look for deeper pattern in the requirements that specific technologies are satisfying.</p><p><strong><em>The Python code for this post can be found at this link</em></strong><em> — </em><a href="https://github.com/leifulstrup/USAspending_Medium_Blog_Analytics_Series/blob/master/Blog_9b_Simple_Ways_to_Identify_Federal_Technology_Buying_Trends_github.ipynb"><em>https://github.com/leifulstrup/USAspending_Medium_Blog_Analytics_Series/blob/master/Blog_9b_Simple_Ways_to_Identify_Federal_Technology_Buying_Trends_github.ipynb</em></a><em>.</em></p><p><em>See this note embedded on the USAspending.gov site about attribution of D&amp;B data and “D&amp;B Open Data” that is embedded in the download data and USAspending.gov website reports such as the one below — </em><a href="https://www.usaspending.gov/db_info"><em>https://www.usaspending.gov/db_info</em></a></p><h3>Coming Attractions</h3><p>I will explore more analytical and market topics in future blog posts — including more market-share trend analysis and <a href="https://en.wikipedia.org/wiki/Natural_language_processing">NLP</a> techniques using USAspending.gov and beta.SAM.gov award and opportunity textual data and mashups with other open data sources.</p><p>The code listed in the examples above and more can be found here:</p><p><a href="https://github.com/leifulstrup/USAspending_Medium_Blog_Analytics_Series">https://github.com/leifulstrup/USAspending_Medium_Blog_Analytics_Series</a></p><p><a href="https://opensource.org/licenses/MIT"><em>MIT Open Source License</em></a><em> Copyright 2020 Leif C Ulstrup</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b257255ea299" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[#8 USAspending.gov]]></title>
            <link>https://lulstrup.medium.com/8-usaspending-gov-8263a5d0f3f3?source=rss-ee9ed3dae4c2------2</link>
            <guid isPermaLink="false">https://medium.com/p/8263a5d0f3f3</guid>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[management]]></category>
            <category><![CDATA[python-pandas]]></category>
            <category><![CDATA[open-data]]></category>
            <category><![CDATA[federal-government]]></category>
            <dc:creator><![CDATA[Leif Ulstrup]]></dc:creator>
            <pubDate>Thu, 29 Oct 2020 13:39:54 GMT</pubDate>
            <atom:updated>2020-10-30T17:44:22.039Z</atom:updated>
            <content:encoded><![CDATA[<h3>#8 USAspending.gov Analytics-Which Contractors Performed the Best in Capturing Information Technology Spending Increases from GFY16-GFY19?</h3><h3>Gaining an Analytics Edge Using Federal Spending Open Data Series</h3><h3>Context</h3><p><strong>Premise of this Blog Series</strong>: Publicly available “<strong>open data</strong>” sources and open-source technology analytics can provide information to reduce uncertainty when making critical business investment decisions.</p><p><strong>Purpose</strong>: Demonstrate to <strong>executives</strong> how easy it is to access and analyze free, open-data sources and motivate them to expect more insight and context from their team when making crucial go/no-go <strong>investment decisions</strong> on bids and new offerings.</p><p><strong>Scope</strong>: US Federal Contracting Market.</p><p><strong>TL;DR </strong>— Analyze GB’s of US Federal spending data with a few lines of Python code. Uncover competitive insights from information and analytics.</p><h3>Introduction</h3><p>In <a href="https://lulstrup.medium.com/4-usaspending-gov-analytics-loading-data-into-dask-and-pandas-3e83af441901">earlier Medium.com blog posts</a> in this series, I focused on the mechanics of downloading and preparing <a href="https://www.usaspending.gov/">USAspending.gov</a> Federal contractor obligations (spending) data using Python and its ecosystem of open-source packages such as pandas and Dask. In this post, I demonstrate how a business analyst and strategist can use this information to analyze the data in ways that provide context for business investment decisions.</p><p>This blog uses Federal contractor spending from Government Fiscal Year (GFY) 2010 through GFY2019 and covers these topics:</p><p>(1) Relative and absolute changes in expenditures by Federal <a href="https://www.fpds.gov/downloads/top_requests/PSC_Manual_FY2016_Oct1_2015.pdf">“Product and Services Code”</a> categories (PSC_Cat).</p><p>(2) Analysis of contractor spending on work designated by Federal procurement officers as “IT and Telecommunications” using the criteria “predominant product or service that is being purchased.” Exploring the expansion and contraction of contractor spending at the Department, Agency, and contractor levels.</p><p>The <a href="https://jupyter.org/">jupyter notebook</a> and <a href="https://python.org/">Python</a> code that accompanies this analysis is located at this <a href="https://github.com/leifulstrup/USAspending_Medium_Blog_Analytics_Series/blob/master/Blog_08_USAspending_PSC_Category_Growth_Rates_github.ipynb">GitHub link</a> within this <a href="https://github.com/leifulstrup/USAspending_Medium_Blog_Analytics_Series">repository</a>.</p><h3>Background</h3><p>The Federal government codes every contractor obligation with a designator that indicates the predominant type of work in that contract obligation. Per the instructions in the “<a href="https://www.fpds.gov/downloads/top_requests/PSC_Manual_FY2016_Oct1_2015.pdf">Federal Procurement Data System Product and Service Codes Manual</a>”:</p><p><em>“The Product and Service Codes (PSC) Manual provides codes to describe products, services, and research and development (R&amp;D) purchased by the federal government. These codes indicate “WHAT” was bought for each contract action reported in the Federal Procurement Data System (FPDS). “</em></p><p><em>“These product/service codes are used to record the products and services being purchased by the Federal Government. In many cases, a given contract/task order/purchase order will include more than one product and/or service. In such cases, the “Product or Service Code” data element code should be selected based on the predominant product or service that is being purchased.</em></p><p><em>For example, a contract for $1000 of lumber and $500 of pipe would be coded under 5510, “Lumber &amp; Related Wood Materials.”</em></p><p>This coding of obligations with the “Product and Service Codes” (“PSC_Cat”) gives an analyst a good starting point to develop market context using the <a href="https://www.usaspending.gov/">USAspending.gov</a> obligation transaction data. Every obligation transaction is coded with one of the designators from the scheme described here:</p><p><a href="https://www.acquisition.gov/sites/default/files/manual/PSC_Data_March_2020_Edition.xls">https://www.acquisition.gov/sites/default/files/manual/PSC_Data_March_2020_Edition.xls</a></p><p>Based on guidance from this <a href="https://www.acquisition.gov/content/psccategoryalignmentapril2020xlsx">link</a>.</p><p>Here is a sample of the major category codes:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/657/1*G1kq48qS6NcXzZxz2FRNJQ.png" /></figure><p>Here is a detailed breakdown of the detailed codes under the PSC_Cat “D” category for “IT and Telecommunications”:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/613/1*cgrB_I5udJTAsOFCDy2ogg.png" /></figure><p><strong><em>Caveats:</em></strong> When analyzing obligations for a specific product or service, this coding structure creates a broad picture of the spending. To get a detailed and more fine-grained picture, you may need to study the original RFP and Award documents posted on beta.sam.gov (former FBO.gov). Also, note that IT-related spending may be coded in the ‘D’ codes for integrated services and products, while independent hardware and software purchases can usually be found in PSC_Cat ‘70’. This <a href="https://www.acquisition.gov/sites/default/files/manual/PSC_Category_Alignment_April_2020.xlsx">document</a> shows an even more complex hierarchy of the relationship of the detailed PSC_Categories and topics such as “information technology,” “medical”, “human capital”, etc. For Information Technology (IT), it is not unusual to find significant content in programs coded in the ‘R’ family of services — “SUPPORT SVCS (PROF, ADMIN, MGMT).” Depending on the fidelity needed and your analysis’s purpose, you may need to go beyond the depth of the examples I show below.</p><h3>Big Picture Context</h3><p>I usually start my analysis by calculating analytics for the broader outline of the study subject and then dig deeper from multiple angles and levels. Doing this also helps as a way to check whether the results make sense.</p><p>For instance, I usually calculate GFY annual totals and then reference those later in the form of <a href="https://en.wikipedia.org/wiki/Checksum">checksums</a> to confirm that I have not lost something in translation. When I first started working with the USAspending data and the Python pandas tools, I was not familiar with how pandas computes sums when it encounters NaN (blank values). I discovered that it stops counting when it encounters a NaN instead of treating them as 0.0. You will see in my data preparation the uses of the pandas fillna(0.0) and fillna(“UNSPECIFIED”) to address this. It is not unusual in formulating pandas selections, slices, and other translations to mistakenly create the wrong calculation, so liberal use of <a href="https://docs.python.org/3/reference/simple_stmts.html#the-assert-statement">Python assert</a> statements to check values can be valuable as you refine your code.</p><p>For this analysis, I start with a quick calculation and graph plotting the total obligations for each GFY from 2010–2019:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/493/1*k2xrD4XNEkdRWpX5lBVqog.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/386/1*Vii8sgAJI0q_NDQjgjQY0Q.png" /><figcaption>Note: GFY2020 is YTD through September 2020 with 3-month lag for DoD obligations</figcaption></figure><p>Next, I load the reference “Product and Service Code” Excel sheet and use that for subsequent calculations and reports.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8Ya9G-SIIQoEr3UppvHPNQ.png" /></figure><p>A fast and straightforward (the source data has ~46 million records) pandas groupby([‘action_date_fiscal_year’, ‘PSC_Cat’]) and summation of federal obligations reveal the distribution of spending on the major categories of work:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WMN732oRDOPGADhH7s8c6w.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/674/1*JrClBos_AXuDFcut_xc6fg.png" /></figure><p>A translation to a pivot_table structure and sorting with a plot shows the relative spending by major PSC_Category and the fraction of total spending for FY19, FY15 (market bottom), and FY10. One can see that the product PSC_Category “1” (10–19) that includes everything from Aircraft and Ships to Ammunition and Weapons has grown from ~14% of contractor obligations in GFY10 to ~18% in GFY19 while R&amp;D has dropped from ~10.5% to ~8%.</p><p>The next plot shows the top six (6) categories over time with a thick black line that shows the relationship to total contractor spending bottomed in GFY2015 and recovered in GFY2018 to GFY2010 levels (this graph does not consider the inflationary differences between GFY10 and GFY19):</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/730/1*t2v8FK2G_sFKSYUzPuJT0g.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/759/1*KnmQ6frY7_s8yJjtHUOn8Q.png" /></figure><p>Since I am personally most familiar with the work in PSC_Cat codes ‘D’ (IT) and ‘R’ (Contract Services), I narrow the analysis to those categories and plot that relationship:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/759/1*ubyeLBzprwjxom0m7XaQGA.png" /></figure><p>We now have context as we dive further into the obligation transactions.</p><h3>Exploring PSC_Cat “D” — IT and Telecommunications Spending</h3><p>The first step is to restrict my analysis to a restricted view of the data. I select obligations where the df_PSC_Cat_D = df.query(“PSC_Cat == ‘D’”) and restrict to GFY16-GFY19. That takes ~2 seconds to extract that view from the ~46 million records.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/824/1*7QZji9Hcino5HIs27Oyhfg.png" /></figure><p>Next, I use the pandas pivot_table (similar to Excel or Google Sheets Pivot Table functionality) to reorganize the view of the data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*h3nYiMFuu1uWFc1mMNDGZA.png" /></figure><p>I then calculate the three (3) year Compound-Annual-Growth-Rate (CAGR) (GFY2019 vs. GFY2016). Here are the top ten (10) largest Departments:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1017/1*8rajxjc8ltPo9eGYbnf1rw.png" /></figure><p>The GFY16 to GFY19 overall Federal CAGR was ~7.5%, while the growth for PSC_Cat “D” was:</p><ul><li>overall market: 8.9%</li><li>DOD: 10.3%</li><li>Civilian Departments: 7.8%</li></ul><p>The Departments with PSC_Cat D spending greater than $50 million/year in GFY19 and higher than the 8.9% CAGR:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1016/1*c8U5vuxFCAY12PuHdwzwig.png" /></figure><p>The lowest growth Departments:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1023/1*tEz55cdEgyyH4ig83Fc3dg.png" /></figure><p>We can take this analysis to a lower level and inspect where the dollar volume of growth was by Agency inside Departments:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MfVYil3bv4UjCvd3oVP4ng.png" /></figure><p>Outside of DoD, the biggest dollar growth from GFY16 to GFY19 was the VA at #1, the US Census Bureau preparing for the 2020 Decennial, HHS/CMS, and Commerce’s NOAA. The VA PSC_Cat D spending grew from ~$2.7B to ~$4.1B =&gt; ~47% increase in spending. The increased VA spending is larger than the GFY19 annual budgets for every Department except for DoD, HHS, DHS, DOC (due to US Census Bureau Decennial spending), and Treasury!</p><p>Some Agencies inside Departments saw significant reductions in PSC_Cat D spending too:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*I3cUFsN3tB4WqSRwuuHGPg.png" /></figure><h3>Analysis of Contractors that Benefited from the Increased Spending</h3><p>With the context at the Department and Agency level, we can explore which contractors were the most successful in capturing PSC_Cat D market share faster than market growth rates from GFY16-GFY19. The PSC_Cat D market three (3) year CAGR was ~8.9% (~20% faster per year than the overall Federal market grew, which is a big difference with compounding growth). For the ten (10) largest recipients of PSC_Cat D contractor obligations in GFY19, this is how they performed:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/962/1*S7vi1Nnb4bFVADlpPqdbgA.png" /></figure><p>Note that Perspecta emerged from M&amp;A, and Cerner jumped into this list very recently. Understanding the baseline for the Perspecta requires more analysis. Cerner experienced impressive GFY19/GFY18 growth of ~$723M/~$490M =&gt; ~48% YoY.</p><p>How much of the growth was due to M&amp;A, and how much from ‘organic’ expansion of the legacy business? Given the size of these businesses and the amount of M&amp;A over the last several years, answering that requires a deeper dive into the obligation transactions that I won’t do here. It is possible to assess that by looking at the additional fields in the USAspending.gov obligation records.</p><p>Which contractors with at least $50M/year (GFY19) in obligations in PSC_Cat D work had the highest PSC_Cat D CAGR?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/949/1*UbUS7XEV34zZq0FN99FLAg.png" /></figure><p>Which contractors increased their annual prime obligations for PSC_Cat D work the most?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/949/1*s7lM0XTYum6TX0TE1J1cvA.png" /></figure><p>Which contractors with at least $350M/year (GFY19) in PSC_Cat D work obligations had PSC_Cat D CAGR’s that exceeded the 8.9% CAGR benchmark (sorted from high to low)?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/958/1*JASc_voviyHGf2hO-r0S-w.png" /></figure><h3>Exporting Data for Data Sharing with Colleagues</h3><p>Once you get the data into a format and record size you need, you can easily export the data into CSV or Excel format files using the convenient pandas to_csv or to_excel functions. Excel and Google Worksheets have limits on the rows and columns, so you need to winnow the data down to a size that can be imported into those packages first. I frequently use Python pandas to organize and pre-process data and import that smaller dataset into Excel or Google Worksheets.</p><p>The Excel Pivot Table functionality is powerful, easy to use, and fast. It is often an excellent choice for quick analysis and sharing with others. You have lots of options once you get the data into thousands of rows and tens of columns from tens of millions of rows and ~300 columns.</p><p>You can import the data into other popular business intelligence (BI) tools such as Tableau, Microsoft Power BI, Microstrategy, Google Data Studio, etc.</p><p>There are other tools emerging that will host interactive analytics on a cloud service using Python open-source tools that are options also. This enables you to both share the data but also build your analytics tools using the power of the Python open-source ecosystem. Some of these are:</p><ul><li><a href="https://www.streamlit.io/">streamlit.io</a></li><li><a href="https://plotly.com/">plotly.com</a></li><li><a href="https://panel.holoviz.org/">holoviz.org</a> on a platform such as Heroku, AWS, Azure, Google Cloud</li><li><a href="https://github.com/voila-dashboards/voila">voila</a></li><li><a href="https://mybinder.org/">binder</a></li><li><a href="https://colab.research.google.com/">colab</a> — I <a href="https://colab.research.google.com/drive/1AISewRUz3NkmAHa4pWV1KOt8hAVu1yYD?usp=sharing">used Google’s colab</a> for my <a href="https://lulstrup.medium.com/gaining-an-analytics-edge-using-federal-spending-open-data-b91b517f2c04">second blog in the series</a> to demonstrate how to access the <a href="https://api.usaspending.gov/">USAspending.gov API’s</a> directly without having to stage the GB’s of obligation records on your machine or cloud storage directory</li></ul><h3>Closing</h3><p>I hope this example inspires you to use some of these techniques to explore relevant topics to your business and the strategic and tactical business investments you must make. The ~280 columns of the original records include information on contract vehicles used, task order expiration dates, the number of bidders on a contract, set-aside status, “contracting agency” in addition to “funding agency,” and many more fields. Many important market insights can be extracted from this open-data source. You can assess how your business performs relative to other firms and guide your business investment decisions, such as selecting strategic accounts to invest in and significant business development pursuits.</p><p>Previous post: #7 USAspending.gov Analytics — Time Series Forecasting with Facebook’s Open Sou</p><p><strong><em>Previous Blog Post</em></strong> in this series: <a href="https://lulstrup.medium.com/7-usaspending-gov-analytics-time-series-forecasting-with-facebooks-open-source-prophet-package-d940bb5af45b"><em>#7 USAspending.gov Analytics — Time Series Forecasting with Facebook’s Open Source Prophet Package</em></a></p><h3>Coming Attractions</h3><p>I will explore more analytical and market topics in future blog posts — including more market-share trend analysis and <a href="https://en.wikipedia.org/wiki/Natural_language_processing">NLP</a> techniques using USAspending.gov data and mashups with other open data sources.</p><p>The code listed in the examples above and more can be found here:</p><p><a href="https://github.com/leifulstrup/USAspending_Medium_Blog_Analytics_Series">https://github.com/leifulstrup/USAspending_Medium_Blog_Analytics_Series</a></p><p><a href="https://opensource.org/licenses/MIT"><em>MIT Open Source License</em></a><em> Copyright 2020 Leif C Ulstrup</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8263a5d0f3f3" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>