<?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 Dylan Anderson on Medium]]></title>
        <description><![CDATA[Stories by Dylan Anderson on Medium]]></description>
        <link>https://medium.com/@dylansjanderson?source=rss-e0f73d4872fc------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*LozUUGP-GEkay2pC9_ehXQ.jpeg</url>
            <title>Stories by Dylan Anderson on Medium</title>
            <link>https://medium.com/@dylansjanderson?source=rss-e0f73d4872fc------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 06:57:43 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@dylansjanderson/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[The Agentic CDP]]></title>
            <link>https://medium.com/data-science-collective/the-agentic-cdp-3f93f55506cf?source=rss-e0f73d4872fc------2</link>
            <guid isPermaLink="false">https://medium.com/p/3f93f55506cf</guid>
            <category><![CDATA[customer-data-platform]]></category>
            <category><![CDATA[data-platforms]]></category>
            <category><![CDATA[customer-experience]]></category>
            <category><![CDATA[marketing-technology]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Dylan Anderson]]></dc:creator>
            <pubDate>Thu, 01 Oct 2026 12:23:18 GMT</pubDate>
            <atom:updated>2026-10-01T12:23:18.824Z</atom:updated>
            <content:encoded><![CDATA[<h4>AI marketing is moving into the data platform. Here’s what that means for data teams</h4><h3><strong>TL;DR</strong></h3><ul><li><strong>Agentic AI is still maturing for non-technical teams</strong>; marketing will likely be the first to change as agents are <strong>impacting the customer journey from both sides</strong> (for research and purchasing, and to understand and engage those same customers).</li><li>CDPs are a great technology for marketing teams, but they are <strong>middleware tools that usually sit outside the governed data estate</strong> and were <strong>not built with Agentic AI in mind</strong></li><li>An Agentic CDP is <strong>completely embedded in the governed data platform, agent-first, and built with marketers in mind</strong>. That combination creates “golden context” about your customers that agents are able to use to better understand, segment, and cater to customers</li><li><strong>Use cases build on each other</strong> (from questions about governed data to campaign design to next-best action and continuous engagement), but buying the platform doesn’t move you up the maturity ladder; it just <strong>means the ladder is there to climb</strong>. So the foundations need to be there</li></ul><p>If you believe LinkedIn, Substack, Google, influencers, etc., you would think Agentic AI is everywhere.</p><p>I would disagree.</p><p>In 2026, it has made incredible leaps, and <a href="https://www.elitebrains.com/blog/aI-generated-code-statistics-2025">AI now writes a serious share of new code</a>, solution development and other technical work. <strong>But it still hasn’t properly cracked business domains.</strong> I work with these people daily. Most are still <strong>using AI as a chatbot</strong>, while some have an email organizer or daily note routine. Maybe the odd person has built a few skills that are already months outdated.</p><blockquote>The biggest reason for the gap is that <strong>Agentic AI wasn’t built for marketers, finance directors, or operations professionals;</strong> it was first built in the terminal (a term most business folk don’t know). Tech-savvy people could connect AI to their IDEs, spin up tasks against files scattered across their machine, and run agents on pre-scoped requirements. <strong>This is intuitive for a data or tech person, but not for everyone else.</strong></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/687/1*0n0qJ73w8dBJkDuleIVUrw.png" /></figure><blockquote>Well, that reality is evolving, and it starts with the industry that brought data to the forefront 30-odd years ago — <strong>marketing</strong>!</blockquote><p>This type of <strong>innovation is taking the form of Agentic CDPs.</strong> I partnered with <a href="https://www.databricks.com/?scid=701Vp00001CpGGDIA3&amp;utm_medium=third-party&amp;utm_source=dylan-anderson">Databricks</a> to create this article for their new product <a href="https://www.databricks.com/product/customerlake-cdp?scid=701Vp00001CpGGDIA3&amp;utm_medium=third-party&amp;utm_source=dylan-anderson">CustomerLake</a>. As you know, I’m selective with sponsored partners, but I was excited about this one: I love Databricks and the ease of their tools, and this is a genuinely new approach to bridging the gap between data/AI and the business.</p><p><em>This article &amp; newsletter is free (thanks to sponsors like </em><a href="https://www.databricks.com/?scid=701Vp00001CpGGDIA3&amp;utm_medium=third-party&amp;utm_source=dylan-anderson"><em>Databricks</em></a><em>), which means passing it on is the only way it grows. If it was useful to you, it’s probably useful to someone on your team (especially the marketing function).</em></p><h3>Agents Haven’t Made a Big Business Impact…Yet</h3><p>My wife is a VP of Marketing at a decently sized scale-up. Her company is very AI-focused and using Claude all over the place. But despite my help and her company’s AI goals, <strong>a huge gap remains between the value she gets from Agentic AI and what I get as a technical person.</strong></p><p>And the numbers back that up. McKinsey’s <a href="https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai">State of AI</a> says 23% of organizations are using an agentic system, with another 39% experimenting. But no more than 10% are scaling agents in any given business function, because that scaling remains confined to IT, data, service operations and software engineering.</p><blockquote><strong>My wife fits in that 23% — using an agentic system but not scaling agents.</strong></blockquote><p>This divide comes down to two things:</p><ol><li>A lack of <strong>technical understanding</strong> to work effectively with Agentic AI</li><li>A lack of<strong> products built directly into how these professionals work</strong>, therefore embedded into the workflow (I don’t count bolt-on AI features in this btw)</li></ol><p>The other component is that companies are obsessed with productivity as the main value driver for AI. I’ve written about the <a href="https://thedataecosystem.substack.com/p/issue-59-ai-strategy">four opportunity areas of AI</a>: productivity gains are the lowest-hanging fruit, knowledge &amp; context is the next rung up, and AI-enabled knowledge is what eventually creates growth.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1011/0*xWuaTVAaxTD0KIgB.jpeg" /></figure><blockquote>The point is that <strong>very little AI investment is pointed at the company’s value proposition or growth; exactly the areas that the business owns</strong>. Productivity only goes so far in marketing, finance or operations, and it’s incredibly hard to measure there anyway.</blockquote><h3>…But Agentic Is Coming</h3><p>Despite all that, we are seeing <strong>a shift in the industry where agents are changing how business teams work</strong>. And that is no more apparent than in marketing.</p><p>It makes sense. Marketing is where <strong>data science gained its popularity</strong>: profiling customers, targeting segments, proving campaign value. Direct-mail segmentation in the seventies, loyalty programs in the 90s, then CRM, web analytics, the MarTech stack, and CDPs. Each of these involved data and reshaped how marketing teams worked.</p><p>What makes this wave different is that AI (and agents specifically) are making an impact from both directions at once:</p><ul><li><strong>Customers</strong> are using agents to research, compare, and increasingly transact</li><li><strong>Companies</strong> are deploying agents to understand and engage those same customers</li><li><strong>Marketing infrastructure</strong> now has to support both faster buyer behaviour and agent-operated internal workflows</li></ul><figure><img alt="The Customer Journey with AI Agents" src="https://cdn-images-1.medium.com/max/1024/0*Az-42HNzRf7MFCYO.png" /><figcaption>Agents are being used on both sides of the customer journey, something marketing teams need to now plan for and anticipate</figcaption></figure><p>And the average buyer is probably using AI more than most marketing teams realize. According to CI&amp;T research across 1,040 US consumers, <a href="https://retailmediabreakfastclub.com/ai-shopping-hits-74-adoption/">74% now use AI tools in their shopping journey</a> at least occasionally. And it isn’t only chat interfaces; agents increasingly browse via APIs, which lets them research options, compare pricing, and purchase across more channels and far faster than any person clicking around. I mean <a href="https://openai.com/index/expanding-access-to-ai-with-chatgpt-ads/">ChatGPT Ads hit $1 billion in annual run rate in 200 days</a>, so obviously some marketing teams are starting to realize it…</p><p>This changes what marketing teams respond to. When a customer’s first pass at your category is an agent summarizing options rather than a person browsing them, the time to act shrinks, and you have to be far more attuned to your competitors.</p><p>So if marketing has been the entryway for every data wave, Agentic AI will be next, especially because marketers already use customer data daily for segmentation, personalization and campaigns.</p><p>And the Customer Data Platform is a natural place for this evolution, given its role in connecting customer data to marketing action.</p><h3>Explaining the CDP</h3><p>A Customer Data Platform <a href="https://www.databricks.com/discover/enterprise-data-platform/cdp">ingests and organizes customer data from the many places it exists across an organization</a> — the website, app, CRM, transactional systems, social media, support tickets, etc. It resolves that into a <strong>unified profile per person, letting marketers build audiences and push them into campaigns</strong>. In plenty of companies, joining online browsing behaviour with in-store purchases and loyalty membership is effectively impossible. <strong>A CDP brings those sources together</strong>, builds a segment (say “lapsed high-value customer”), and streams it into a campaign without a request to the data team.</p><p>I’ve worked with a lot of marketing and sales teams, and doing this well was extremely difficult. Doing it as a data team, without a platform, was even harder. Data teams got caught up handling identity resolution across fragmented sources, directing data into activation tools, or just providing the right data when marketing needed it, usually on short notice.</p><p>So the CDP was a game changer for a lot of organizations. It provided:</p><ul><li>Unified customer profiles</li><li>Data-driven segmentation of customers</li><li>Reverse ETL into activation tools and platforms</li><li>Accessible UI for research and insights</li></ul><p>All of that was great; however, CDPs were never the perfect fix. They are <strong>middleware</strong> — sitting between the data platform, the MarTech stack, and the channels — and usually owned by marketing, so they <strong>lived outside the governed data estate</strong>. <a href="https://thedataecosystem.substack.com/p/issue-35-business-value-reverse-etl">Reverse ETL</a> helped bridge the gap, but it synced a copy rather than live data.</p><p>Essentially, CDPs were built for marketing and activation, not for reasoning and data clarity. So they often lack the <a href="https://thedataecosystem.substack.com/p/issue-14-data-modelling">modelled data</a> that keeps things consistent over time: defined grain, conformed dimensions, modelled relationships, lineage, and agreed business definitions.</p><blockquote>That was fine when humans reviewed the data, shaped the segments and crafted campaigns before manually activating.</blockquote><blockquote><strong>It gets much harder with AI, especially if a marketing team wants agents handling personalized sends or campaign measurement</strong> (and you know every exec aspires to that). The agent needs to know the customer definition, the campaign metrics, what counts as a churn event, the purchase cadence, and how all of it links together.</blockquote><p>This all comes back to one growing problem in the AI world: the <strong>lack of contextual data to map these things together.</strong></p><h3>When There’s No Governed Route to the Data</h3><p>A CDP definitely helps because it at least stores this data in the same place. <strong>Its biggest bottleneck is the middleware disconnect</strong>. Now that we are in an AI world, a campaign-building agent may only get a representation of the data and context it needs because the CDP is disconnected from the primary data warehouse and the data team’s governance.</p><p>Worse still are organizations that don’t even have this middleware. Some use AI to reach from their harness (e.g., Claude, ChatGPT, etc.) directly into their data warehouse. This is dangerous because without proper governance and oversight, it <strong>opens up a host of issues for the data team that owns and manages that data</strong>. Especially when business stakeholders prompt AI to pull this or take a look at that, it becomes a security and management nightmare.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/598/0*siSUR9xaO269XkIZ.jpeg" /></figure><blockquote><strong>So many data teams simply don’t give marketing or sales access to the warehouse.</strong></blockquote><p>Again, I’ll reference my wife. As a VP of Marketing and avid AI user, she would love access to her Snowflake data warehouse. But she has no feasible route to it: there’s a technical gap in getting at it directly, no quality checks between the warehouse and her AI harness, and no architectural oversight of how she might use the data. She could request an MCP connection, but that hands a non-technical operator a raw pipe to production customer data. And while these warehouses support governed MCP access, it still takes a lot of data team resources to teach non-technical users and oversee their usage (and no data team has that kind of availability); ideally, a purpose-built application would be better.</p><p>This pattern worries me most when we talk about AI democratization. <strong>Non-technical stakeholders don’t quite know what they are doing with their AI harness, but they are doing it ad hoc nonetheless.</strong> Then the definitions drift, nothing is repeatable, and the whole thing goes stale. Or marketing builds its own thing because working with the data team is too slow, leading to a second version of the truth. That’s the risk with AI, and while CDPs were built to solve that kind of problem, they weren’t built with embedded AI in mind!</p><h3>From CDP to Agentic CDP</h3><p>Technology comes in waves, with each iteration adding another layer to solve for new problems that have come about. Obviously, the most recent wave is AI…and more specifically Agentic AI.</p><p>In that vein, it also makes sense that these tools eventually reach the business domains. And as described above, marketing is a great candidate, with the CDP the most obvious first step.</p><p>An <a href="https://www.databricks.com/blog/what-is-an-agentic-cdp">Agentic CDP</a> does three things:</p><ol><li>It is completely <strong>embedded in the governed data platform</strong>, rather than being copied and synced to a marketing tool, therefore giving it access to the data model, governance, and resources it needs to share the right insights.</li><li>It is <strong>agent-first</strong>, treating Agentic operation as the primary mode of delivery for personalization, segmentation, campaign activation, etc.</li><li>It’s <strong>built with marketers in mind</strong>, ensuring that the UI and UX is not a terminal-based agent swarm, but an accessible tool that non-technical people can use, built for their use cases.</li></ol><p>Most CDPs on the market might say they’re agentic, but in reality, <strong>they were built before ChatGPT existed, and most agentic capabilities are bolted on</strong>, reducing their effectiveness. There is a real difference between adding a chat window to an existing tool and rebuilding the tool so agents are embedded in the workflows.</p><blockquote>What does this enable from a business value perspective? Instead of just static records of customer data, this approach creates what <a href="https://chiefmartec.com/">Scott Brinker</a> originally termed <strong>“golden context”:</strong> the overlap between your company’s goals, your customer’s needs, and your systems’ ability to deliver it in real time.</blockquote><p>Say someone is shopping for running shoes on your site. The customer record tells you who they are: they bought a pair eighteen months ago, are size ten, spent about $140, and opted into email. This is useful information, and a good CDP will give this to you today.</p><p>But the golden context takes it to the next level. Beyond segmentation, golden context can provide size-level inventory data about the customer and show that they should have a contact-frequency limit based on their email interactions and how they purchase. No individual is realistically putting that together for each customer, but an agent embedded in the CDP should. And it might do it with the context that the company’s goal is to increase the rate of 2nd-time purchases, <strong>so the agent prioritizes that in its marketing actions using the golden context.</strong></p><p>Context is such a buzzy word right now, but everyone agrees <a href="https://thedataecosystem.substack.com/p/issue-60-context-layer-ai-needs">it is the real moat for AI</a>. So imagine assembling all of that context about your customers and feeding it directly into your campaigns. That is what the Agentic CDP is after, and doing it at scale is becoming very real, very quickly.</p><h3>The Embedded Differentiator</h3><p>But context is only powerful if you have the data to create it. And this is where the embedded nature of the Agentic CDP fits in — <strong>you get the customer expertise from the marketing team integrated with the customer data of the warehouse, all in one tool.</strong></p><p>For most companies, your customer data is the most valuable data you have on hand. Loyalty behaviour, purchase patterns, lifetime value, and churn signals are the data points that decide whether you keep a customer or lose them. And while this is no surprise, that data is also among the worst quality: records go stale, identities fragment across tools and definitions don’t agree.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*8d5K0DnSpW26-3bN.png" /><figcaption>Going from middleware to embedded with agentic capabilities will make a big difference for marketing teams</figcaption></figure><p>The beauty of Databricks is that they’ve been solving this data access, governance and quality question for years (and made a pretty big name of themselves doing it), so <a href="https://www.databricks.com/blog/introducing-customerlake-agentic-cdp?scid=701Vp00001CpGGDIA3&amp;utm_medium=third-party&amp;utm_source=dylan-anderson">CustomerLake translates well into this type of architecture</a>. Either way, this is what I would want to see for Agentic CDPs to make sure marketing teams get the benefit they need:</p><ul><li><strong>Governance and PII in One Place</strong> — Customer data needs to be governed well, especially given the PII risks or GDPR rules. Every MarTech tool you copy customer data into becomes another surface to secure, audit, and honour deletion requests against. Removing copies streamlines how you govern that data. <a href="https://www.databricks.com/product/customerlake-cdp?scid=701Vp00001CpGGDIA3&amp;utm_medium=third-party&amp;utm_source=dylan-anderson">CustomerLake</a> runs this through <a href="https://www.databricks.com/product/unity-catalog">Unity Catalog</a>, which, if you already use Databricks, is likely the governance layer running the rest of your estate already</li><li><strong>Data Quality Built In</strong> — Any CDP worth its price should do identity resolution. But if schemas change in the warehouse, that doesn’t always translate across, and engineering pipelines can break. This is the data-readiness piece. An agentic system needs to combine the <a href="https://www.databricks.com/dataaisummit/session/introducing-customerlake-profile-agents">preparation of data</a> by identifying quality issues, enriching records and creating business-ready profiles with <a href="https://www.databricks.com/blog/introducing-customerlake-agentic-cdp?scid=701Vp00001CpGGDIA3&amp;utm_medium=third-party&amp;utm_source=dylan-anderson">agentic identity resolution</a>, which combines deterministic, probabilistic, agentic and human-review workflows rather than betting everything on a single matching method. Sounds like a lot but data quality is essential and agents plugged into the business data warehouse are perfectly positioned to do this with the right context</li><li><strong>Natural Language Access</strong> — The future is not SQL queries; it is natural language guiding those queries. So the tool needs features where a business user describes what they need in their own language and gets something they can act on (for Databricks and CustomerLake, this is <a href="https://www.databricks.com/product/genie/agents">their Genie tool</a>). Honestly, this would be the first thing my wife would want from her MarTech tooling</li><li><strong>Auditing and Troubleshooting</strong> — Testing and reconciliation only really works when activation tools are properly connected to the warehouse. So if a campaign misfires (the wrong segment goes out, the numbers don’t reconcile, the product isn’t linked properly), working out why is cumbersome and manual, meaning it rarely happens. An agentic CDP should query the campaign, audit what went wrong using its <a href="https://www.databricks.com/product/unity-catalog">embedded data and lineage function</a>, and set guardrails so it doesn’t happen again</li><li><strong>Connection Without Migration</strong> — Saving the best for last. Migration is a bad word in any company. Your new CDP should plug into the data platform you already own, which Databricks handles through <a href="https://docs.databricks.com/aws/en/sql/language-manual/sql-ref-federated-queries">Lakehouse Federation</a>, reading across Snowflake, BigQuery and operational databases. But it also has to plug into the MarTech and AdTech stack you already run, through Reverse ETL integrations out to your email, ad and CX platforms. Doing both a MarTech and data migration to connect a CDP is why many implementations take more time than planned. So whoever you go with, figure out the solutions architecture on both sides before signing anything</li></ul><p>When AI Agents can work directly with governed customer data in a tool the marketing team knows how to use, everything becomes easier. This is the advantage of embedded AI, and this is how you actually make Agentic AI scale to non-data and tech functions.</p><h3>Getting Value from the Agentic CDP Use Cases</h3><p>Coming back to the four areas of AI opportunity I mentioned earlier (productivity, knowledge &amp; context, new products or tools, and growth opportunities), the embedded nature of the Agentic CDP lets you start capturing value beyond productivity:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6KWctvaYM53mYbT3fvyxyQ.png" /><figcaption>The better your data is organized, the more powerful AI agents can be, especially when they are embedded in a tool like a CDP</figcaption></figure><ol><li><strong>Ask questions of governed customer data</strong> — With unified profiles, agreed definitions and governance in place, a marketer gets insight through natural language instead of waiting a week for a data team request. This is a productivity gain, letting the marketing team do existing tasks far more quickly (size audience segments, compare lifetime value across acquisition channels, draw out loyalty insights). Regular CDPs do this too, but the embedded piece delivers it faster, more reliably, and with other data to enhance insights</li><li><strong>Agent-assisted audience building and campaign design</strong> — Here we start seeing knowledge and context benefits. Using the data available, an agent can propose audiences and campaign designs based on the activation connectors and the overall strategy. Think uploading your marketing strategy into the tool and getting back an execution-ready set of audiences and campaigns built from it. That is extremely powerful, especially combined with automated campaign delivery</li><li><strong>Agent-recommended next best action</strong> — Context then starts turning into growth, with AI identifying and acting on real-time signals from sales, the market, or whatever else you’ve modelled, then measuring the result and adjusting with human guardrails in place. Catching a churn risk from a drop in usage and recommending the retention offer before the renewal date. Moving spend mid-campaign when a channel is underperforming. Or deciding not to send because that individual is tired of marketing materials. These are hard to do one customer at a time, which is exactly what Agentic AI plus a CDP enables</li><li><strong>Continuous, individualized engagement</strong> — This is what everybody is advertising, but it is still an immature area — a lot of AI engagement tools miss the mark for personalization and still write with an AI-style. That will change, and it starts with the right context, controls and loops. In CustomerLake the feature to tackle this is <a href="https://www.youtube.com/watch?v=Po1gXA67wdc">Infinity Campaigns</a>, continuous engagement loops where agents analyze real-time customer signals, decide the next-best action, act across channels, then learn and optimize around a business goal. This won’t be perfect at first (human experts need to constantly feedback to improve it), but it gives your team the right tool (not just a common AI harness) to start down this road</li></ol><p>Agentic AI is about delivering an outcome end to end, with the tool designing the process, doing the work, and overseeing the output as it goes. That said, human experts still matter enormously in guiding it. Setting the strategy and the guardrails is the human job, which then frees agents to handle the high-frequency, context-heavy decisions where they can tap into customer data. <strong>This also frees the marketing team for brand strategy, positioning and creative</strong>, where AI still does a horrible job (avoiding the sea of AI-slop sameness).</p><p><strong>These use cases are also sequential.</strong> Asking questions of governed data is common and well understood. Once teams are comfortable there, they can direct agents to build audiences and campaigns, and eventually execute against them. But the team needs to own that journey/ evolution! Each stage builds upon the next, with the Agentic CDP helping enable that progression (rather than being a silver bullet that does it automatically).</p><p>Which brings me to my caution in most articles I write about tooling: <strong>buying an Agentic CDP doesn’t make your marketing team more data-savvy and overcome a lack of modelling or agreed definitions</strong>. It gives you the tools and ability to make these things possible, but solving requires a holistic approach. It’s the same <a href="https://thedataecosystem.substack.com/p/issue-61-maturity-assessment">maturity logic</a> I apply to any data &amp; AI capability.</p><blockquote><strong>Buying the platform doesn’t move you up the maturity ladder. It just means the ladder is there to climb.</strong></blockquote><p>So if your customer data isn’t modelled anywhere, if definitions are contested across teams, or if your marketing team still doesn’t quite know how to use data to build audiences, start there instead. And there is also no need to overengineer for your own use case; if you’re running a few campaigns per year to a modest list, you probably don’t need an agentic-first tool just yet.</p><p>The other thing I would caution about is to be realistic about time. <a href="https://cdp.com/articles/cdp-implementation-guide/">CDP implementations</a> typically run three to twelve months. And while warehouse-native approaches can land in two to six weeks, that only holds if the modelling is genuinely done (e.g., governed gold tables, agreed business definitions, identity resolution rules). I highly suggest a solutions architect, internal and ideally one from the vendor, to help with implementation. When I was consulting I worked with a Databricks solutions architect on a project, which helped a ton with my client’s data migration we were going through.</p><p>So overall, this kind of tool, properly implemented, can do a lot:</p><ul><li><strong>For the data team</strong>, it provides time back without giving up control. Rather than constant analytical requests from marketing, <strong>the CDP provides the access and insight needed</strong>. Moreover, the data team can now rest easy knowing they’ve defined the trusted datasets, models, permissions and business definitions a single time, and every downstream agent and marketer inherits and uses them. This is the self-serve argument but against governed data in a tool built for the business (not a Tableau or Power BI dashboard)</li><li><strong>For the marketing team</strong>, it speeds up and evolves what they can do with data. They can readily test ideas, build segments and campaigns, and measure against them. Given ROI is so crucial for marketing right now, having a governed foundation that connects campaign decisions, activation signals, customer behaviour and business outcomes is crucial. This finally gives data and marketing teams a common measurement layer</li></ul><p>We know that Agentic AI is here to stay. I hate the buzziness of the term and how overused it is, but the benefits are starting to come through, especially in data and tech. <strong>In my view, marketing is next</strong>. I’m very excited to see how <a href="https://www.databricks.com/product/customerlake-cdp?scid=701Vp00001CpGGDIA3&amp;utm_medium=third-party&amp;utm_source=dylan-anderson">CustomerLake</a> delivers value for Databricks customers once it comes out of private preview. I genuinely think it will be a big game changer (and after consulting my wife, she does as well).</p><p>Thanks again to <a href="https://www.databricks.com/?scid=701Vp00001CpGGDIA3&amp;utm_medium=third-party&amp;utm_source=dylan-anderson">Databricks</a> for the partnership and you for reading. See you all next week!</p><p><em>A note on this partnership: I worked with the team at </em><a href="https://www.databricks.com/?scid=701Vp00001CpGGDIA3&amp;utm_medium=third-party&amp;utm_source=dylan-anderson"><em>Databricks</em></a><em> on this article. They built </em><a href="https://www.databricks.com/product/customerlake-cdp?scid=701Vp00001CpGGDIA3&amp;utm_medium=third-party&amp;utm_source=dylan-anderson"><em>CustomerLake</em></a><em>, the agentic CDP referenced throughout, and it’s worth a look if you’re thinking about how customer data and agents fit together in your stack. The overall deep dive into Agentic AI CDPs is mine and I partnered with </em><a href="https://www.databricks.com/?scid=701Vp00001CpGGDIA3&amp;utm_medium=third-party&amp;utm_source=dylan-anderson"><em>Databricks</em></a><em> because they want you to understand the category as well. The writing and opinions are mine, while Databricks helped me sanity-check the technical claims.</em></p><p><em>Originally published at </em><a href="https://thedataecosystem.substack.com/p/issue-71-the-agentic-cdp-customerlake"><em>The Data &amp; AI Ecosystem</em></a><em>, which goes out every Sunday. </em><a href="https://thedataecosystem.substack.com/"><em>Subscribe here</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3f93f55506cf" width="1" height="1" alt=""><hr><p><a href="https://medium.com/data-science-collective/the-agentic-cdp-3f93f55506cf">The Agentic CDP</a> was originally published in <a href="https://medium.com/data-science-collective">Data Science Collective</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Debt AI Creates]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/data-science-collective/the-debt-ai-creates-62c805dc890a?source=rss-e0f73d4872fc------2"><img src="https://cdn-images-1.medium.com/max/600/1*S-3gRELAQEkXuJVONvy0jQ.jpeg" width="600"></a></p><p class="medium-feed-snippet">The debt AI is slowly starting to accumulate all around us (and what to do about it)</p><p class="medium-feed-link"><a href="https://medium.com/data-science-collective/the-debt-ai-creates-62c805dc890a?source=rss-e0f73d4872fc------2">Continue reading on Data Science Collective »</a></p></div>]]></description>
            <link>https://medium.com/data-science-collective/the-debt-ai-creates-62c805dc890a?source=rss-e0f73d4872fc------2</link>
            <guid isPermaLink="false">https://medium.com/p/62c805dc890a</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[ai-debt]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[tech-debt]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Dylan Anderson]]></dc:creator>
            <pubDate>Thu, 24 Sep 2026 12:21:56 GMT</pubDate>
            <atom:updated>2026-09-24T12:21:56.268Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Getting Rid of Data Debt with AI]]></title>
            <link>https://medium.com/data-science-collective/getting-rid-of-data-debt-with-ai-1cf9aaee7555?source=rss-e0f73d4872fc------2</link>
            <guid isPermaLink="false">https://medium.com/p/1cf9aaee7555</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[data-debt]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[data-engineering]]></category>
            <dc:creator><![CDATA[Dylan Anderson]]></dc:creator>
            <pubDate>Thu, 17 Sep 2026 17:06:59 GMT</pubDate>
            <atom:updated>2026-09-17T17:06:59.147Z</atom:updated>
            <content:encoded><![CDATA[<h4>How to use modern tooling, AI, and a bit of strategic thinking to solve your organization’s data debt</h4><h4><strong>TL;DR:</strong></h4><ul><li>Data debt is the<strong> valuable information you have collected and stored, but cannot draw on</strong> (or if you can, it isn’t usable). This is a different problem from data quality; it is about <strong>accessing this hidden, valuable information</strong> and using it to draw out insights</li><li>The reason it never gets solved is that <strong>most organizations define it as a catchall</strong> which turns it into a behemoth project nobody can start. If you narrow the issue to access, it becomes more manageable</li><li>Three things have changed that make this problem solvable and very relevant: (1) The commoditization of connecting data efficiently between systems/ tools through CDC and automated backfills, (2) Tooling and AI that compensate for team gaps (especially data engineers), and (3) Effective AI genuinely requires this data to be surfaced</li><li>None of this works without <strong>figuring out the architectural design: </strong>what data matters, where it lives, what the schema should be. Then you can hand the technical work to automated tooling</li></ul><p>Data debt is pervasive: data piles up from multiple systems over long periods, gets put into a vault somewhere, and is never seen or used again.</p><p>This is the reality in many organizations. They describe an environment with <strong>a lot of data</strong> (albeit of questionable quality and not very accessible), but <strong>little to no real insight coming from it</strong>.</p><blockquote>The worst part is that <strong>most people there know this</strong>; they understand they are sitting on an enormous amount of information that could drive business value. <strong>But it still goes nowhere, because they don’t know how to use it</strong> (other than downloading a CSV and building a few Excel pivots).</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/735/1*l4mKiI1VYW7N8gH-vHM8qg.jpeg" /><figcaption>While companies know this relevant data exists, they don’t find the time to use it…</figcaption></figure><p>AI is helping make sense of that data and get more value from it, <strong>but it can’t be the answer on its own.</strong></p><blockquote>So just like technical debt (which I wrote about last week), data debt becomes a similar case study — <strong>the debt persists and accumulates because nobody has ever had the capacity, drive, or expertise to act on it.</strong></blockquote><p>And just like tech debt, I believe <strong>AI has opened up new opportunities to tackle data debt in all types of organizations.</strong> We just have to be a bit smarter about the issue, and break down the problem into modular bits, starting with the most important piece — <strong>getting access to the right data.</strong></p><h3>What Data Debt Actually Is (And What It Isn’t)</h3><p>My definition of data debt is <strong>the valuable information you have collected, paid for, stored, but cannot draw upon (or if you can, it is not usable)</strong>. Similar to tech debt, it is a debt in that the data costs money to carry it, while returning no value and accumulating/ degrading if you don’t do anything about it.</p><p>There are many dimensions to data debt.</p><p>If you look at the classic definition, the term ends up being a catch-all. Look it up, and you will see it reference everything from poor quality to inconsistent definitions to legacy systems to undocumented pipelines or unmanaged data.</p><p>For example, here are some definitions from reputable sources:</p><ul><li><strong>Google Research (Sculley et al., NeurIPS 2015)</strong> — The original serious treatment, from the paper <a href="https://papers.nips.cc/paper/5656-hidden-technical-debt-in-machine-learning-systems">Hidden Technical Debt in Machine Learning Systems</a>. Their argument focuses on the idea that data dependencies cost more than code dependencies because they’re far harder to detect. You can trace a code dependency through an import statement, whereas data dependencies are only unearthed when something upstream changes and everything downstream degrades without warning</li><li><strong>Matillion</strong> — Splits it into <a href="https://www.matillion.com/blog/the-four-main-categories-of-data-debt-explained">four categories</a> covering everything from schema design through to governance and consumption</li><li><strong>Syniti</strong> — Frames it as <a href="https://blog.syniti.com/what-is-data-debt-and-what-to-do-if-you-have-it">the accumulated cost of shortcuts and deferred decisions across the whole data lifecycle</a> (ingestion, modeling, governance, consumption)</li><li><strong>Hackolade</strong> — Breaks it into <a href="https://hackolade.com/reduce-your-data-debt.html">structural debt (bad schema design), content debt (inaccurate or duplicate records), and operational debt (undocumented pipelines and shadow systems)</a>.</li></ul><p>My biggest problem with these perspectives is that <strong>the issue becomes overbearing and impossible to tackle all at once</strong>. Most of these definitions are broad enough to contain almost any data problem you’ve ever had.</p><blockquote>I mean, tech debt is similar, but at least with technologies you can modernize and replace, one system at a time. <strong>Data is an output of many systems and these problems originate from so many different sources that it becomes extremely hard to tackle</strong>.</blockquote><blockquote>But that is only if you define it in this all-encompassing way.</blockquote><p>Unfortunately, <strong>that is how most organizations define it and try to tackle it</strong>. First, they think about data quality, whether you can trust the data or define it. Then they think of the sheer volume of data and its many sources across the organization. And soon you have a meeting room full of 10 different people describing their own individual problems, leading to a behemoth project of consolidation, data quality improvement, and analytical improvement.</p><blockquote><strong>And guess what? In the end nothing happens. Why? Because it is impossible to deal with all of this at once.</strong></blockquote><p>For example, I worked with an airline company that wanted to solve for their data debt, and to do that, they decided to build a customer data model. <strong>Their first mistake was trying to connect all 80 systems and agree on definitions across them all.</strong> Unfortunately for me, that discussion (and the scale of the task) eventually killed the project. This was despite my suggestion that we start small and deal with the most important data sources, connections and definitions first.</p><blockquote>So returning to my definition of the term, it <strong>goes back to drawing upon the valuable information your company has access to, and using it to eventually draw out insights</strong>. The most important piece here in my mind? Accessing that data before it accumulates in a negative way and becomes unusable/ stale!</blockquote><h3>How Debt Accumulates &amp; Doesn’t Disappear</h3><p>Similar to tech debt, the biggest problem with data is that it accumulates in systems, and the task to unravel it and fix the sourcing of it becomes too big to do easily.</p><blockquote><strong>And this problem is brought to you by technology!</strong></blockquote><p>Every team and every person uses technology. Now each of these technologies/ tools generates data. What is more, this data is often seen as a differentiator; the executive team signs up for these tools because they generate data you can use to derive insights and improve operations.</p><p>But unfortunately, <strong>they buy the tool and forget all the steps it takes to use the data it generates.</strong></p><figure><img alt="How Data Enters Organizations" src="https://cdn-images-1.medium.com/max/1024/1*YaY7ajoWyp-LqlCSXWaHig.png" /><figcaption>The entrance of data into the organization is usually where the debt starts to accumulate</figcaption></figure><p>For most companies, they parked the data (either within the tool or on some unorganized SQL database) because storage is cheap. Say, we might need it later, and carry on with your operations because no one has time to figure out how to use that data properly. And the gap continues to widen as there is no priority to come back to that data and use it; the amount continues to grow and it becomes harder to model or wrangle it as time passes.</p><p>The consequences of this are pretty well known, so I’m not going to dive into them in too much detail:</p><ul><li><strong>Data Staleness</strong> — The data still exists but the meaning around it changes, especially when definitions or metrics shift. Using that data becomes difficult because of its staleness</li><li><strong>New Pipelines/Dashboards</strong> — Instead of going back and updating what exists, people rebuild data access. This creates both data and tech debt, duplicating costs and creating redundancy. There is a lot of this happening with AI now (which we will talk about next week)</li><li><strong>Unsustainable Data Volumes</strong> — The biggest thing holding back organizational leaders from dealing with this is the sheer scale of the problem and not knowing how to even approach it</li><li><strong>Unconnected is the Norm</strong> — Operational tools don’t get connected because it is seen as hard and difficult, so the cycle continues</li></ul><p>Fixing these things has always fallen into the realm of data engineering. Unfortunately, most of the engineering teams I see are understaffed, lack clear direction, are ticket-based, and don’t have a clear data model or strategy to work from. Therefore, the data debt problem continues to accumulate no matter how many resources you throw at it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/591/1*E6iT7C8sbv9-5cdHzKGlsg.jpeg" /><figcaption>It really is no wonder DE teams don’t have enough capability to correct data debt, look at what they have to deal with…</figcaption></figure><p>But this is stuff you know already. I<strong>f you work in a large corporate organization, you’ve seen this before</strong>, and you don’t need to be told why it happened. Instead, you need to understand what to do about it (because most leaders are afraid to start without that path), and that’s what I’ll spend most of this article on.</p><h3>How To Solve for Data Debt</h3><p>Now, I won’t contend that solving these issues is easy — especially in large organizations or if the data team isn’t well resourced. But it’s not impossible, especially with the advances in technology, AI, and connected systems.</p><p>Before, this might have been a transformation project for a large organization. And for small and medium-sized organizations, this type of task has always seemed out of reach; while they were accumulating data, they never had the team to build the capability to use it. I would often hear the phrase “We don’t have a Google-type budget to do this stuff.” In my mind, that was an excuse. And nowadays, it is a really bad excuse.</p><p>Three things have changed:</p><ol><li><strong>Connecting Systems is Being Commoditized</strong> — Connections and migrations once required bespoke engineering and pipelines. But now, <a href="https://docs.estuary.dev/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64">getting data from one place to another and from system to system can be as easy as configuring a connection</a></li><li><strong>Tooling &amp; AI Can Compensate for Team Gaps</strong> — Smaller, less specialized teams can now approach these things with the right tools, primarily because the expertise is embedded in the tools</li><li><strong>Effective AI Requires This</strong> — The raison d’être! If you want AI to live up to its expectations, you need to surface this locked-away data. You need these data systems connected so that inputs go to the right places at the right time. Essentially, alleviating data debt will be a requirement for data, analytics &amp; AI success</li></ol><p>I’m going to dig into these three components, but before I do, I need to emphasize that there is a precursor to doing this — architectural design.</p><h3>Start Solving for Data Debt with the Right Approach</h3><p>The problem with most companies and executives is that <strong>they buy a tool thinking it will be the silver bullet for all their data issues.</strong> This is even more apparent with AI (many of my conversations with COOs, and even CTOs, are me trying to convince them that they need to do more than just buy Claude).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*AhX3iKF4aY7p5YrQmzbqGg.jpeg" /><figcaption>I have this conversation more than I would like…</figcaption></figure><p>The truth is that <strong>tools are now very capable of doing the technical work</strong> (e.g., coding, monitoring, system connectivity, etc.), and <strong>companies that use them correctly are getting insane amounts of value</strong>, especially given how manual it is for some organizations to connect data/ design bespoke pipelines. For example, this article is brought to you by <a href="https://estuary.dev/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64">Estuary</a>, which I partnered with because I have seen what a difference this kind of tooling makes for clients (instead of an expensive engineering hire to build pipelines, you can connect data that had been sitting untouched for years in an <a href="https://estuary.dev/pricing/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64">incredibly cost-efficient way</a>).</p><p>But organizations need to <strong>realize they must properly plan for how these tools can deliver the promised value</strong>, especially with AI. This involves thinking about the right architecture and data model. You buy tools because they fill a gap; but to fill that gap, you need the right plan in place to make sure the tech/ tooling knows how it should work!</p><p>So before jumping to the tooling, there are four phases that can help set this up in the right way. And to be clear, running this in a looped, modular fashion is the best way to do it (otherwise you run into the immense scale problems I mentioned above with the airline data model):</p><figure><img alt="The Data Debt Planning Cycle" src="https://cdn-images-1.medium.com/max/897/1*AgaqfV8aefbPxroMHvTpPw.png" /><figcaption>The strategic process and loop behind unlocking data debt</figcaption></figure><ol><li><strong>Figure out what data actually matters</strong> — Most organizations have a lot of data sources and a lot of data debt. And while it seems logical, the worst thing you can do is focus on what data you have, because that is how the problem grows and it becomes harder to prioritize. Instead, start with the decisions you want to make within the business, identify which data will help make those decisions, and apply the 80/20 rule (80% of the solution delivered by 20% of the data). This is a mix of stakeholder management/ consultations and business analysis. It also connects directly to the use cases (which may help point you down the right path to get rid of data debt)</li><li><strong>Identify where that data lives and what it takes to reach it</strong> — Now you map the sources. Which systems hold it, who owns them, how current it needs to be, and what condition it’s actually in upon arrival. The difference now vs. five years ago is that not every source is a big engineering job. Now, a lot of sources have supported connectors, especially depending on the tools you use. Estuary, for example, connects to <a href="https://estuary.dev/integrations/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64">over 200 tools/ systems</a> (and growing), making it an easy ask to pull certain data sources. Similar to the first step, this helps you prioritize: if you have equally important data sources (for your use cases), whichever is more easily configurable should be the first choice to solve for. Here is where you also note more complex cases/ systems and potentially reference back to any tech migration efforts or timelines</li><li><strong>Design the schema and architecture</strong> — This is probably the most important step. Think about the schema for the data source as it relates to identified and potential use cases. For example, what are the entities, how do they relate, what grain do you need, what is the definition of customer or revenue, and where does the modeled data land? Ensure this is mapped out within the overall data and <a href="https://thedataecosystem.substack.com/p/issue-64-solutions-architecture">solutions architecture</a>. While AI can help you speed this up, human expertise is necessary, trust me</li><li><strong>Hand the technical work to AI and modern tooling</strong> — I’ll speak more to this in a second. But basically, once you have mapped out what you need, where it is, and what shape it should be in, the actual movement and transformation are increasingly something you configure rather than build. This is where tools like Estuary do the heavy lifting: <a href="https://estuary.dev/product/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64">capturing from the source, continuous sync, and materialization</a> wherever you’re modeling it. The point of the first three phases is that you arrive at this one knowing exactly what you’re asking the tooling to do</li></ol><p>Then you go back to phase one with the next slice or use case. If you’ve designed the <a href="https://thedataecosystem.substack.com/p/issue-14-the-forgotten-guiding-role">data model</a> well from the beginning, these passes should get faster because you are adding data sources to a proven architecture.</p><p>With that being said, some sources/ systems do create problems. But this is where the three changes I identified make it worthwhile.</p><h3>Why Now is the Time to Solve for Data Debt</h3><figure><img alt="Framework to solve for data debt" src="https://cdn-images-1.medium.com/max/1024/1*Rmh6vFFG5NgZcwZ_xJi3Hw.png" /><figcaption>Getting access to the right data is the first step to reducing your organization’s data debt</figcaption></figure><h4>1) Connecting Systems is Being Commoditized</h4><p>First off, <a href="https://docs.estuary.dev/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64">moving data from one place to another and from system to system can be as easy as configuring a connection</a>.</p><blockquote><strong>In the past, connections and migrations required bespoke engineering and pipelines. Now, this process can be automated.</strong></blockquote><p>Tools (like Estuary) do this through <a href="https://estuary.dev/blog/change-data-capture/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64"><strong>change data capture</strong></a><a href="https://estuary.dev/blog/change-data-capture/">, or CDC</a>, which has been around for a while but has become faster and more efficient in the past few years. Basically, CDC reads the database’s change log and only ingests/ maps the changes. Compared to the previous process — run a large extraction query against a source operational system overnight and wait for the delay to run into the morning — this is a godsend for accelerating system connections.</p><p>For data debt specifically, this goes beyond just connecting the systems. Because the data has likely sat unused for a long time, it’s important that when you stand up a connection, you (or the tool) <strong>run a </strong><a href="https://estuary.dev/blog/what-is-a-data-backfill/"><strong>backfill</strong></a> — go and get the historical data sitting in that system. This used to be a <a href="https://www.youtube.com/watch?v=ob2VqpXeUu4&amp;t=13">pain in the ass</a>, but it is much more straightforward now with modern DE tools, especially if you have the right schema set up. After that, <a href="https://docs.estuary.dev/reference/backfilling-data/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64">switch to CDC</a> to keep it up to date from then on.</p><blockquote>So basically <strong>the backfill pays down the debt you have already accumulated and the CDC allows you to stop accumulating more of it.</strong> Instead of just connecting a new pipeline, you rebuild the system in the right way, with the right context, that will keep working efficiently moving forward without constant maintenance.</blockquote><p>Speaking of efficiency, this also keeps getting faster. Backfills used to take a long, long, long time. But now tools are working off more efficient languages like Rust which takes a historical load that used to be a weekend job into something you can run anytime (Estuary recently rewrote their transaction loop and are reporting <a href="https://estuary.dev/runtime/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64">up to 8x higher throughput on a single worker</a>, sustaining around 200GB per hour). So huge volumes of historical data is <strong>no longer a data debt blocker</strong>. Worth noting that this isn’t a replacement for your data engineering function, but its capability that your data engineering team needs to be more efficient.</p><blockquote>Doing this type of backfill properly, with CDC moving only the changes, also <a href="https://estuary.dev/runtime/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64"><strong>significantly reduces the compute and processing costs</strong></a><a href="https://estuary.dev/runtime/"> of this kind of migration</a>.</blockquote><p>For example, when your database saves a set of changes that belong together — say, twenty rows across five tables — it treats them as a single unit. All of it lands, or none of it does. Most pipelines don’t preserve that. They deliver the pieces separately, at different times, so your warehouse briefly shows a version of reality that never existed in the source system: an order with no line items, a debit with no matching credit. Teams deal with this by running reconciliation jobs to confirm the numbers eventually agree and by writing “eventually consistent” into the documentation.</p><p><strong>That is data debt being created by the pipeline you built to fix data debt.</strong> By writing your captured changes to append-only logs stored in your own cloud storage, a single source transaction usually spans several of them. This means that instead of staging all these shifts in memory, your source arrives as a single transaction, no matter how many machines handle it along the way. <a href="https://estuary.dev/runtime/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64">Estuary’s CEO David Yaffe puts it as a trade-off teams have always been stuck with: speed or correctness, pick one</a>. This kind of approach (especially built in Rust) allows you to get both.</p><p>Of course I also have to mention the other part of connected systems in today’s embedded AI world — the abundance of MCPs and connectors. This allows your AI harness to read and work within systems it is already connected to. Estuary have <a href="https://docs.estuary.dev/features/mcp-integration/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64">their own MCP integration</a> for exactly this, so an agent can work against your pipelines and collections directly. While a lot of your legacy tech won’t have MCPs/ connectors, some tools that store unused data might, and that connection goes a long way toward helping you figure out how to source/ route it into your analytical data layer according to your existing data model!</p><h4>2) Tooling &amp; AI Can Compensate for Team Gaps</h4><p>Smaller, less specialized teams can now address data sourcing issues with the right tools, primarily because the expertise is embedded in them.</p><p>Think about what this all used to require:</p><ul><li>First somebody who could code.</li><li>Then for them to actually build the scripts.</li><li>Then profiling a dataset to work out what condition it’s in.</li><li>Then tracing column-level lineage so you know what breaks if you change something.</li><li>Oh and don’t forget working through what an undocumented field from 2016 means and writing the documentation to fill the gaps.</li></ul><blockquote><strong>Each one of these tasks was a ticket, and data engineers were inundated with them in a never-ending cycle. Therefore, organizations didn’t have time or money to allow engineers to reduce any data debt.</strong></blockquote><p>Now, teams have <strong>AI agents, affordable tools, and skills</strong> to do the vast majority of those things at speeds previously unheard of. That doesn’t mean you stop needing data people. But <strong>it means a much smaller team can cover ground that used to require a whole function</strong>. And extra resource can focus on these never-thought about tasks (like Data Debt).</p><p>This doesn’t stop at just connecting data — the tooling to assess, align, and fix data quality is also improving significantly, allowing teams to better maintain the pipelines they spun up. So once you solve that first barrier of getting the right data into the right place, addressing its quality issues isn’t as big a hurdle either.</p><p>This is also <strong>why the “we don’t have a Google-type budget” line is no longer defensible</strong> — all my clients right now are installing this type of thing with 0–2 data engineers on staff and running it for <a href="https://estuary.dev/pricing/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64">hundreds of dollars a month in tooling</a> (instead of thousands).</p><h4>3) Effective AI Requires This</h4><p>AI is the future and if you want to use it properly, <strong>you need to surface locked-away data.</strong></p><blockquote><strong>Essentially, alleviating data debt will be a requirement for data, analytics &amp; AI success.</strong></blockquote><p>But companies are nowhere close to that point yet. While they are hooking up AI tools and harnesses, <a href="https://www.cloudera.com/about/news-and-blogs/press-releases/2026-03-05-only-7-percent-of-enterprises-say-their-data-is-completely-ready-for-ai-according-to-new-report-from-cloudera-and-harvard-business-review-analytic-services-reveals.html">only 7% of people said their organization’s data was completely ready for AI adoption</a>. Meanwhile, Gartner predicts that <a href="https://www.gartner.com/en/newsroom/press-releases/2025-02-26-lack-of-ai-ready-data-puts-ai-projects-at-risk">through 2026 organizations will abandon 60% of AI projects that aren’t supported by AI-ready data</a>. Honestly, with the embedded AI hype starting to die down, I’m seeing that in organizations that haven’t built these foundations.</p><p>Just to be clear, this isn’t just the “connect your data or your AI initiatives will fail” argument (which honestly gets a little bit boring). <strong>Rather it is that the data your business has generated (or has stored in the back closet) is bespoke and is the secret sauce in today’s AI-driven world.</strong> Every month, models get better and cheaper, and the tooling around them gets more capable and more commoditized. Your competitor can buy the same models and the same platforms you can. <strong>But they can’t buy that data: customer behaviour, operational costing, process efficiencies</strong>. That’s proprietary, specific to you, and accumulates.</p><blockquote><strong>But you need access to it for AI to make it work.</strong></blockquote><p>So every year you leave that data locked away in operational systems or random tools is a year you are forgoing the true advantages of AI — using your data as a knowledge system to improve your business.</p><h3>What to Take Away</h3><p>The core of this article and argument is that <strong>data debt has remained an issue for years</strong> because:</p><ol><li>The problem was <strong>too big to tackle all at once</strong>, and</li><li>It was <strong>economically unfeasible to resolve</strong></li></ol><p>This article is trying to change that narrative. Just like I <a href="https://thedataecosystem.substack.com/p/issue-68-technical-debt">outlined last week with tech debt</a>, the world of AI has opened up new possibilities; couple technological advancements and highly capable tools with an architectural mindset to overcome the niggling problems that have negatively impacted your company for years!</p><blockquote>And start with <strong>connecting the right data!</strong> The cost and ease of data access continues to fall, and the longer your proprietary information remains locked up, the less ability your company has to reap the benefits of AI.</blockquote><p>At the same time, you still need the right people to decide what the data means, what it’s for, and how to structure it. That is the new bottleneck, but it is a whole lot easier than it was 5–10 years ago!</p><p>Looking ahead to next week, we have the last of the three debt articles — AI Debt. Everybody is starting to see the debt AI is creating; <strong>with such an easy way to create stuff, things don’t get documented, understood or used.</strong> And <strong>then it goes to die</strong>. So instead of debt accumulating because fixing them was too expensive, this is about cheap production creating things we don’t need and that are hard to maintain.</p><p><em>A note on this partnership: I worked with the team at </em><a href="https://estuary.dev/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64"><em>Estuary</em></a><em> on this piece. They do real-time data movement — </em><a href="https://estuary.dev/blog/change-data-capture/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64"><em>change data capture</em></a><em>, streaming and batch across </em><a href="https://estuary.dev/integrations/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64"><em>200+ connectors</em></a><em> — which is exactly the “getting your data somewhere you can use it” problem this article is about. The bit that made them the right partner for this particular piece is that a connection </em><a href="https://docs.estuary.dev/reference/backfilling-data/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64"><em>backfills your history and then keeps it current</em></a><em>, so you’re paying down the data debt you have and not building more of it. They’ve also just shipped a </em><a href="https://estuary.dev/runtime/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64"><em>ground-up rewrite of their runtime</em></a><em> in Rust, and their whitepaper is worth checking out. And there is a </em><a href="https://estuary.dev/pricing/?utm_source=substack&amp;utm_medium=referral&amp;utm_campaign=tde64"><em>free tier</em></a><em> if you just want to try it out. Thanks to them for supporting the newsletter!</em></p><p><em>Thanks for the read! Comment below and share the newsletter if you think it’s relevant! Feel free to also follow me on </em><a href="https://thedataecosystem.substack.com/"><em>Substack</em></a><em> and </em><a href="https://www.linkedin.com/in/dylansjanderson/"><em>LinkedIn</em></a><em>, or reach out if you are looking for some </em><a href="/service/mailto:dylan@daecosystems.com"><em>top-notch freelance consulting input</em></a><em>! See you amazing folks next week!</em></p><p><em>Originally published at </em><a href="https://thedataecosystem.substack.com/p/issue-68-data-debt"><em>The Data &amp; AI Ecosystem</em></a><em>. It goes out every Sunday. </em><a href="https://thedataecosystem.substack.com/"><em>Subscribe here</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1cf9aaee7555" width="1" height="1" alt=""><hr><p><a href="https://medium.com/data-science-collective/getting-rid-of-data-debt-with-ai-1cf9aaee7555">Getting Rid of Data Debt with AI</a> was originally published in <a href="https://medium.com/data-science-collective">Data Science Collective</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Understanding Technical Debt (and how to approach retiring it)]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/data-science-collective/understanding-technical-debt-and-how-to-approach-retiring-it-de61f7ba29fc?source=rss-e0f73d4872fc------2"><img src="https://cdn-images-1.medium.com/max/1518/1*Xd8wph_eWJVX5ycrJR6l3g.png" width="1518"></a></p><p class="medium-feed-snippet">The dreaded combination of legacy technology with no appetite to solve it; explaining it and offering a solution</p><p class="medium-feed-link"><a href="https://medium.com/data-science-collective/understanding-technical-debt-and-how-to-approach-retiring-it-de61f7ba29fc?source=rss-e0f73d4872fc------2">Continue reading on Data Science Collective »</a></p></div>]]></description>
            <link>https://medium.com/data-science-collective/understanding-technical-debt-and-how-to-approach-retiring-it-de61f7ba29fc?source=rss-e0f73d4872fc------2</link>
            <guid isPermaLink="false">https://medium.com/p/de61f7ba29fc</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[tech-debt]]></category>
            <category><![CDATA[technology-strategy]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Dylan Anderson]]></dc:creator>
            <pubDate>Thu, 10 Sep 2026 12:26:51 GMT</pubDate>
            <atom:updated>2026-09-10T12:26:51.816Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Data & AI Ecosystem Map]]></title>
            <link>https://medium.com/data-science-collective/the-data-ai-ecosystem-map-3fb895957d48?source=rss-e0f73d4872fc------2</link>
            <guid isPermaLink="false">https://medium.com/p/3fb895957d48</guid>
            <category><![CDATA[ai-strategy]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[ecosystem]]></category>
            <dc:creator><![CDATA[Dylan Anderson]]></dc:creator>
            <pubDate>Tue, 25 Aug 2026 14:52:53 GMT</pubDate>
            <atom:updated>2026-08-25T14:52:53.189Z</atom:updated>
            <content:encoded><![CDATA[<h4>A 12-minute overview of how the Data &amp; AI industry has shifted in the last 3 years and what that means</h4><h3>TL;DR</h3><ul><li>The Data Ecosystem map is a single-page view of everything that must work together for an organization to derive value from its data: business drivers on the left, the data lifecycle in the middle, consumption and decision-making on the right, management underneath, and external influences around the edges</li><li>AI runs through every domain, which is why this newsletter is now called The Data &amp; AI Ecosystem</li><li>The two new zones on the map are Data &amp; AI modeling and the Context Layer, and Embedded AI</li><li>The previous data domains still exist, but each one has changed in how it’s done, who does it (not always humans now), and what it produces</li></ul><p>Before I started my newsletter (<a href="https://thedataecosystem.substack.com/">The Data &amp; AI Ecosystem</a>), a simple Google search of the data ecosystem would often return a technical view of a company’s tooling landscape.</p><p>My argument was that the ecosystem is so much more than that; it encompasses business strategy, organizational design, stakeholder management, relationships, data flows, strategic considerations, new technologies, etc.</p><p>But no one was explaining it like that, and all the material out there was specific to each domain, without doing the work to connect the dots between them. So I did my best to do that, and created this newsletter. This included my view of the <a href="https://thedataecosystem.substack.com/p/issue-3-pulling-back-the-data-ecosystem">entire Data Ecosystem on one page</a> (which has been redrawn many times). Business drivers on the left. The data lifecycle running through the middle. Consumption and decisioning on the right. Underpinning management along the bottom, holding all of it up, and the other influences/ considerations around the edges.</p><figure><img alt="The Data Ecosystem pre-AI (circa 2024)" src="https://cdn-images-1.medium.com/max/1024/1*6GvbwfdxIWgKIHru8Ce8tg.png" /><figcaption>The Data Ecosystem pre-AI (circa 2024)</figcaption></figure><p>This infographic was great for explaining the complexity of data, especially to consulting clients. It drove home the point that your data problems aren’t contained to one domain but need to be viewed holistically. And while it is complex, data largely flows through in a process-like way, from strategy on the left to the data lifecycle hosted on a platform, with analytics turning it into an insight/decision at the end.</p><p>But in the past year, AI has gotten so prominent that it changed the map; in fact, it changed the ecosystem!</p><p>Instead of being a domain or a piece of the process, it runs through the entire ecosystem. If data is the sun that powers the ecosystem, AI is the air that each domain interacts with to function.</p><p>This shift has <a href="https://thedataecosystem.substack.com/p/the-data-ai-ecosystem">led me to rename my newsletter The Data and AI Ecosystem</a>. It has also led me to write a lot more about AI, as you’ve seen in the past 10 or so articles.</p><blockquote><strong><em>The Data &amp; AI Ecosystem is a single-page view of everything that must work together for an organization to derive value from its data and AI</em></strong><em>: business drivers, the data lifecycle, consumption and decision-making, the management that underpins them, and the external influences at the edges.</em></blockquote><p>But before I go too far down more topics, domains, and articles on AI, I want to holistically evaluate how the data ecosystem has evolved to incorporate the new power of AI.</p><h3>Why the Data Ecosystem became the Data &amp; AI Ecosystem</h3><p>I will always be the first to say that <a href="https://thedataecosystem.substack.com/p/issue-11-dispelling-the-ai-hype-train">AI is not magic</a> and that there are many gaps in the technology. I’ve created countless memes making fun of it and how companies are <a href="https://thedataecosystem.substack.com/p/capitalist-vibes-of-ai">over-hyping it in their sales and marketing channel</a>s to drive sales/ revenue.</p><p>But, like all technology, AI has evolved and advanced to a point where it is legitimately changing the industry at a pace no one has ever seen before.</p><p>And I’m still skeptical of AI, but <strong>I’m skeptical of bolt-on AI</strong>, where somebody expects it to overcompensate for foundations that don’t exist. Without reliable definitions to work from, agreed/ governed metrics, data lineage, and any other additional context, the model won’t give you what you want, especially if you are prompting it with one vague sentence (like we so often do). And <a href="https://www.cloudera.com/about/news-and-blogs/press-releases/2026-04-14-nearly-80-percent-of-enterprises-say-ai-is-held-back-by-data-access-challenges-cloudera-report-finds.html">most data isn’t ready for AI (only 7% of orgs according to HBR)</a>, so bolt-on models really don’t work!</p><p>Now when we move from bolt-on AI to embedded AI, that is where we are starting to see stuff that looks like magic. This is not a surprise to anybody who has read this newsletter before, but the companies and individuals who started with data and <a href="https://thedataecosystem.substack.com/p/issue-60-context-layer-ai-needs">context at the foundational level</a> and embedded AI from there are not only getting exceptional value from AI but <strong>are changing how we work and how the economy functions.</strong></p><blockquote><strong><em>For reference, bolt-on AI expects the model to compensate for foundations that don’t exist. Embedded AI starts from data and context and builds outward.</em></strong></blockquote><p>So instead of AI being a domain tacked on to the data science/consumption part of the process, it flows through from start to finish. Every domain therefore needs to evolve, because the domains now work differently, both in how they are done, who does them (not always humans anymore), and what they output.</p><p>And so, the Data Ecosystem map has evolved to the Data &amp; AI Ecosystem map!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*n1BvOrJ-kGSd4xxxoaLx0w.png" /><figcaption>The Data &amp; AI Ecosystem map</figcaption></figure><p>The rest of the article digs into this (well as much as I can in a twelve-minute article). We are going to walk through the updated map and how the different domains within each section have evolved with AI.</p><h3>The Eight Zones of the Data &amp; AI Ecosystem</h3><h4>1) Business Drivers Still Come First</h4><p>Data never starts with Python, SQL or any technical code; it starts with the business.</p><p>That hasn’t changed with AI. But what we have seen is that the business is now obsessed with AI, and thereby inadvertently obsessed with data (even if they don’t know it).</p><p>The <a href="https://thedataecosystem.substack.com/p/issue-52-explaining-business-models"><strong>business model</strong></a> is the clearest example, which <a href="https://thedataecosystem.substack.com/p/issue-53-business-models-and-data">I’ve written about extensively</a>. Companies have <a href="https://thedataecosystem.substack.com/p/issue-54-refactoring-business-model">changed how they operate because of AI</a> (e.g., how they staff/ resource, generate revenue, go-to-market, etc.). This flows into the <a href="https://thedataecosystem.substack.com/p/issue-7-where-it-all-begins-the-business"><strong>business strategy</strong></a><strong>,</strong> which now has to incorporate AI into the planning cycle as a core facet rather than the “nice-to-do” enabler that data was.</p><p>AI also changes the operational components of this section. The <a href="https://thedataecosystem.substack.com/p/issue-8-deliver-on-the-data-needs"><strong>business needs and requirements</strong>, <strong>stakeholder alignment</strong></a>, and the <a href="https://thedataecosystem.substack.com/p/issue-13-defining-the-data-operating"><strong>operating model</strong></a> all have to take into account that the business is already using AI tools, and that getting them to work well in the organization starts from the beginning. Organizations now expect AI Agents to do a lot of the work, which impacts how we need to scope out new products/ workflows, enable existing employees, and even <a href="https://thedataecosystem.substack.com/p/issue-18-organisational-structures">rethink the <strong>org structure</strong></a>. The key thing here is to be realistic with expectations and learn to crawl, then walk, and then run before you expect that rocketship AI growth that executives are all talking about.</p><p>Then you have the <a href="https://thedataecosystem.substack.com/p/issue-57-data-strategy-direction"><strong>Data </strong></a><strong>&amp; </strong><a href="https://thedataecosystem.substack.com/p/issue-59-ai-strategy"><strong>AI Strategy</strong></a> which should encompass all these changes and help organizations build on the right foundations. This will help show the path to success and sustainable AI usage, but only if it considers the underlying data components that enable AI. The strategy will also inform any <strong>Data &amp; AI investment decisions</strong>, the overall <strong>approach/ philosophy</strong> and the <a href="https://thedataecosystem.substack.com/p/issue-65-data-and-ai-use-cases-spec"><strong>use cases</strong></a> that both business and data teams will execute against.</p><h4>2) Data &amp; AI Modeling and the Context Layer</h4><p>Data and AI modeling and the context layer is the first of two new zones on the map. You used to scope the use cases and then dive right into sourcing and ingesting data to build solutions. That process isn’t as straightforward anymore.</p><p>Instead, people are spinning up POCs like it’s nothing. To do it correctly, however, you need to start with the <strong>AI engineering</strong> and the <strong>context graphing</strong>. First, think about the <a href="https://thedataecosystem.substack.com/p/issue-25-role-of-data-archtitecture"><strong>Data &amp; AI architecture</strong></a> in how the <a href="https://thedataecosystem.substack.com/p/issue-64-solutions-architecture">use cases/ solutions should be structured (e.g., is it a deterministic script or completely AI-enabled)</a>, and <a href="https://thedataecosystem.substack.com/p/issue-14-the-forgotten-guiding-role">how the data model fits into that</a>. Second — and building on the data model — how is<strong> </strong>all the data and business knowledge represented in the organization. This is the <a href="https://thedataecosystem.substack.com/p/issue-60-context-layer-ai-needs">context that everybody is talking about right now</a> (e.g.,<strong> ontologies and business glossaries</strong>, the <strong>semantic and context layer</strong>, the <strong>knowledge graph)</strong>, which is poorly done at most organizations because they haven’t put the time and effort into this increasingly important domain. Now, <a href="https://futurumgroup.com/press-release/enterprise-data-analytics-survey-finds-59-investing-in-semantic-layers-as-critical-ai-infrastructure/">Futurum estimates 59% of enterprise decision-makers</a> are directing new budget toward semantic layers, so this capability is coming.</p><p>Oh, I also added in agentic retrieval, because more often than not, humans are handing over information retrieval to AI agents, where the system decides for itself what it needs to fetch rather than having a fixed query. Without a well-modeled data foundation or context, this can go quite poorly.</p><p>On my old map, knowledge graphs and the semantic layer were listed under surrounding influences, filed as new technology worth keeping an eye on. Well, embedded AI has gone and made them really relevant, really fast, and honestly, I view these as one of the biggest enablers of your whole Data &amp; AI Ecosystem if done right.</p><h4>3) The Data Lifecycle Looks the Same But Works Completely Differently</h4><p>Despite the promises of AI, <a href="https://thedataecosystem.substack.com/p/issue-10-the-data-lifecycle">companies still need to source, ingest, process and serve data</a>.</p><p>How they go about it, however, has changed with AI. Data sourcing is the mildest shift, as the types of data haven’t changed, but the ease of pulling unstructured and operational content into your processes has made this so much easier.</p><p>For <strong>integration and storage</strong>, the <a href="https://thedataecosystem.substack.com/p/issue-36-real-life-advice-on-building">classic ETL approach is no longer the only way</a> — and in many cases, it isn’t the right way. <a href="https://thedataecosystem.substack.com/p/issue-64-solutions-architecture">Your architecture now has to be designed for agents as consumers.</a> Meanwhile, <a href="https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report">AI has sped up engineering to a ridiculous degree</a> (but with more data and builds, it usually leads to more engineering requests). The <a href="https://thedataecosystem.substack.com/p/issue-21-demystifying-the-buzzy-data">platform infrastructure also looks differen</a>t, especially as tools themselves embed AI agents, skills, or features that make jobs easier.</p><p>Processing and serving got more complicated. You have to know what you want from your processing and serving layer because you could have deterministic jobs, LLM-driven outputs, or completely self-serve. The solutions architecture and orchestration within that will continue to evolve at pace, and is definitely something to keep an eye on. This bleeds into the Data Analytics &amp; Consumption segment, which used to be the fourth box within the Data Lifecycle Process, but it now belongs with consumption given all these changes with AI.</p><p>Overall, the lifecycle process looks similar, but AI has significantly changed how practitioners work inside it. If you aren’t using (or at least thinking about) AI to do your job in these domains, you are falling behind.</p><h4>4) Embedded AI: agents, evals, guardrails and MCP</h4><p>Embedded AI is the second new zone: <a href="https://thedataecosystem.substack.com/p/issue-63-the-emergence-of-agenticops"><strong>agentic workflows</strong></a>, the <strong>AI harnesses and models</strong> underneath them (think Claude or ChatGPT), <strong>AI literacy and policies</strong>, <strong>skills and prompting</strong>, <strong>evals and guardrails</strong>, and <strong>MCPs</strong> connecting all of it to the systems. When I was skeptical of AI, it was as a bolt-on technology. But when it is properly embedded, it is a huge differentiator.</p><p>Done well, these things touch everything else. It should read from the context layer, act on/ with the data lifecycle, and produce outputs that are consumed by both business and data stakeholders. <a href="https://thedataecosystem.substack.com/p/issue-51-ai-governance-considerations">It also needs guardrails, governance, and literacy for people to make the most of it.</a></p><p>This box will probably change in the next year. Most of these components weren’t common terms six months ago, and there is no established best practice for any of it. Nonetheless, this is quickly becoming one of the most important areas of the Data &amp; AI Ecosystem, not least because it enables the true democratization of data &amp; AI across the organization.</p><h4>5) Analytics and Consumption Is No Longer the Last Mile</h4><p>AI’s biggest shift for most companies is the fact that analytics and consumption is no longer dependent on the data team, it can be prompted through an AI harness.</p><p>The question of whether <a href="https://thedataecosystem.substack.com/p/issue-32-effective-dashboard-design"><strong>BI</strong>, <strong>ML &amp; analytics</strong> tools</a>, and <strong>data/ AI product builds</strong> should be one-off prompted through an AI harness is a completely different story. But, you know what, we are here and it is happening so we may as well not ignore that truth…</p><p>So Data &amp; AI democratization is here. After spending a decade doing Tableau and SQL trainings, the trick ended up being natural language conversational interfaces that can connect right into the data platform. For some, this is a win, but at the same time, data teams need to make sure the previous steps (data lifecycle, data &amp; AI modeling, context layer) are done well so that some COO isn’t pulling a report that is completely wrong and off-base. There also has to be strong data &amp; AI management to underpin that (but more on that later).</p><h4>6) Business Decisioning &amp; AI Operationalization</h4><p>Everything comes back to this. Making decisions from your data or AI tools is why everything else is done.</p><p><strong>Data literacy</strong> has become even more essential; with people now using AI daily, they need to understand the role of data in those interactions. This ranges from data privacy, regulations, <a href="https://thedataecosystem.substack.com/p/issue-15-the-data-quality-conundrum">garbage-in-garbage-out</a>, etc. Data literacy isn’t just BI or reading visualizations anymore.</p><p>On the <a href="https://thedataecosystem.substack.com/p/issue-66-value-driven-prioritisation"><strong>business case</strong> and <strong>value creation</strong> sides, AI has falsely made it seem that everything is cheap to build. In reality, there needs to be more scrutiny on how this process takes place</a>. To consider the short-term gains and the long-term impact in managing these tools that make decisions for the business.</p><p>The new addition here is AI operationalization, which is all about how AI runs in an organization, makes decisions, and takes action without doing something incorrectly.</p><p>Oh, I’ve also included a feedback loop. Given the importance of AI to the business model and strategy, there <a href="https://thedataecosystem.substack.com/p/issue-54-refactoring-business-model">needs to be better feedback mechanisms to ensure the organization knows what it can and should do with AI and how it’s performing</a>.</p><h4>7) Underpinning Management Just Gets More Important</h4><p>The names of most of these domains have not changed. But what does change is that management is no longer just done by and for humans; nope, we need to consider how AI and agents play into it all.</p><p><a href="https://thedataecosystem.substack.com/p/issue-47-role-of-data-governance"><strong>Data &amp; AI governance</strong></a> has the hardest job. It’s a domain that <a href="https://www.dataversity.net/white-papers/dataversity-trends-in-data-management-2025-survey-results/">never got enough investment</a>, and now it has to help govern AI usage within the systems that manage data. <a href="https://thedataecosystem.substack.com/p/issue-50-ai-governance-definition">When I wrote about AI Governance</a>, the domain was still early, so I led on governing from a value-led perspective. But now, this has extended into guardrails, boundaries and policies for embedded and agentic AI usage.</p><p><a href="https://thedataecosystem.substack.com/p/issue-44-upstream-data-observability"><strong>Observability</strong> </a>now goes beyond determining when pipelines run and has extended to understanding how the AI works and routes through the data systems, especially when something doesn’t go right. <a href="https://thedataecosystem.substack.com/p/issue-45-data-contracts-testing"><strong>Data contracts</strong></a> has evolved to not only have humans as customers, but also agentic systems. The agentic evolution makes them that much more important too, because if you want your AI to work properly, then contracts are a good solution. And of course <a href="https://thedataecosystem.substack.com/p/issue-43-data-quality-today"><strong>data quality</strong> has become even more important</a> as there is no data team between the query and the output, so ensuring reliability, accuracy, and consistency helps ensure outputs can be trusted. <strong>Master data management</strong> follows the same path, focused on repurposing the highest value data assets in the organization for direct AI usage.</p><p><strong>Data privacy and security</strong> concerns continue to grow as the efficacy of AI models have increased, especially when it comes to hacking/ cybersecurity. Hence the new domain of <strong>AI security</strong>, which concerns how to protect your own systems from AI and ensure your AI works properly within your own boundaries. <strong>Data catalogues/ lineage</strong> becomes more important to inform/ audit the context layer. Finally, <a href="https://thedataecosystem.substack.com/p/issue-62-explaining-devops-vs-dataops"><strong>DevOps/DataOps</strong> principles are essential</a> and need to be considered when building new AI products; hence <a href="https://thedataecosystem.substack.com/p/issue-63-the-emergence-of-agenticops">the birth of <strong>AgenticOps</strong></a>.</p><h4>8) The Other Edge Considerations</h4><p>Finally, the last bit is the outer edges of the map.</p><p>All of these have evolved due to AI. Within the <strong>strategic considerations</strong>, obviously the <a href="https://thedataecosystem.substack.com/p/issue-19-developing-an-overarching"><strong>technology strategy</strong> needs to now consider AI tooling</a> and how that embeds in the platform. A holistic view of this is essential, so <strong>enterprise architecture</strong> becomes even more important, planning your technology estate for the fast-paced world of AI. The other big one is <strong>change management</strong>. Every organization out there is having a very tough time with AI enablement, and proper change management principles/ delivery is going to be essential to driving proper adoption (rather than AI usage without the right training/ oversight).</p><p>As some <strong>surrounding influences</strong> have graduated into the other realms, the rest have just intensified with AI. The <strong>market</strong> has completely shifted with new AI tools and companies. <strong>Ethics</strong> now includes the question of how do we work with AI responsibly, which creates another level on top of the data layer. <a href="https://thedataecosystem.substack.com/p/leaders-outsource-to-ai"><strong>Training and development</strong> becomes even more important</a> as the whole organization is working with AI and they need the skills to do so properly. I’ve also added <strong>AI advancements</strong> as an influence in its own right, because the capability will continue to evolve and companies need to pay attention to how that changes/ shifts.</p><figure><img alt="The Data &amp; AI Ecosystem Map with associated context" src="https://cdn-images-1.medium.com/max/1024/1*_v9PSfKA4mOk23vXQepsBA.png" /><figcaption>The whole map, with the new additions and large-scale changes marked.</figcaption></figure><h3>Why An Ecosystem View Matters</h3><p>Nobody knows what AI will look like in a year or two. I don’t.</p><p>But I do know it will play a part in how we work, especially for those in the data industry.</p><p>And I also know that <strong>a </strong><a href="https://thedataecosystem.substack.com/p/issue-1-we-need-to-rethink-data"><strong>holistic, ecosystem perspective</strong></a><strong> (rather than a siloed domain view) will help you understand and enable AI in your organization/ life</strong>. <a href="https://thedataecosystem.substack.com/p/issue-55-the-ai-sociotechnical-system">AI is about systems thinking</a>, and taking a holistic approach is one of the best ways to do that.</p><p>I mean, don’t trust me. As you’ve probably heard 1,000x, AI is only ever as good as the data underneath it. And that data is produced, managed and manipulated by an entire ecosystem (including humans, tools, and AI).</p><p>And this is why I’m still excited about writing this newsletter; some people get bored of content creation after a year or two, but for me, this is constant learning and teaching on a subject that has so much room to evolve.</p><p>So thank you for sticking with me all this time. Or if you’re new to the Data &amp; AI ecosystem, thank you for joining in.</p><p>There is a hell of a lot to dig into over the next while, and I hope you find this newsletter helpful for everything you’re doing in the data and AI space.</p><blockquote>Next up, I’m digging into a long-running blocker in this industry — debt. First we start with <strong>tech debt, then data debt, and finally a newly emerging form of debt, AI debt</strong>. Honestly, this topic has been sitting in the corner of my map as for three years, and it is one of the biggest blockers to progress, so it’s nice to finally dig into it, especially with the capabilities of AI at our disposal!</blockquote><p>If this was useful, please send this article to one or two people who’d get something out of it. Otherwise, have a great week!</p><p><em>Thanks for the read! Comment below and share the newsletter if you think it’s relevant! Feel free to also follow me on </em><a href="https://thedataecosystem.substack.com/"><em>Substack</em></a><em> and </em><a href="https://www.linkedin.com/in/dylansjanderson/"><em>LinkedIn</em></a><em>, or reach out if you are looking for some </em><a href="/service/mailto:dylan@daecosystems.com"><em>top-notch freelance consulting input</em></a><em>! See you amazing folks next week!</em></p><p><em>Originally published at </em><a href="https://thedataecosystem.substack.com/p/issue-67-the-data-and-ai-ecosystem-map"><em>thedataecosystem.substack.com</em></a><em>. The Data &amp; AI Ecosystem goes out every Sunday, with a shorter mid-week piece that doesn’t run anywhere else.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3fb895957d48" width="1" height="1" alt=""><hr><p><a href="https://medium.com/data-science-collective/the-data-ai-ecosystem-map-3fb895957d48">The Data &amp; AI Ecosystem Map</a> was originally published in <a href="https://medium.com/data-science-collective">Data Science Collective</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Data Ecosystem: Where do you even start?]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://dylansjanderson.medium.com/the-data-ecosystem-where-do-you-even-start-6b5768ea6ab0?source=rss-e0f73d4872fc------2"><img src="https://cdn-images-1.medium.com/max/600/0*vzn9xzb9w4dEuVMx.gif" width="600"></a></p><p class="medium-feed-snippet">How to take that next step in your data knowledge</p><p class="medium-feed-link"><a href="https://dylansjanderson.medium.com/the-data-ecosystem-where-do-you-even-start-6b5768ea6ab0?source=rss-e0f73d4872fc------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://dylansjanderson.medium.com/the-data-ecosystem-where-do-you-even-start-6b5768ea6ab0?source=rss-e0f73d4872fc------2</link>
            <guid isPermaLink="false">https://medium.com/p/6b5768ea6ab0</guid>
            <category><![CDATA[data]]></category>
            <category><![CDATA[data-strategy]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[data-analytics]]></category>
            <category><![CDATA[data-ecosystems]]></category>
            <dc:creator><![CDATA[Dylan Anderson]]></dc:creator>
            <pubDate>Tue, 10 Sep 2024 16:53:59 GMT</pubDate>
            <atom:updated>2024-09-10T16:53:59.727Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Forgotten Guiding Role of Data Modelling]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/data-science/the-forgotten-guiding-role-of-data-modelling-a76f69364284?source=rss-e0f73d4872fc------2"><img src="https://cdn-images-1.medium.com/max/1000/0*dpe5a8qO72HOheB4.png" width="1000"></a></p><p class="medium-feed-snippet">Getting to the bottom of what structuring your data responsibly really means</p><p class="medium-feed-link"><a href="https://medium.com/data-science/the-forgotten-guiding-role-of-data-modelling-a76f69364284?source=rss-e0f73d4872fc------2">Continue reading on TDS Archive »</a></p></div>]]></description>
            <link>https://medium.com/data-science/the-forgotten-guiding-role-of-data-modelling-a76f69364284?source=rss-e0f73d4872fc------2</link>
            <guid isPermaLink="false">https://medium.com/p/a76f69364284</guid>
            <category><![CDATA[data-modelling]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[data-structures]]></category>
            <category><![CDATA[data-architecture]]></category>
            <category><![CDATA[data-engineering]]></category>
            <dc:creator><![CDATA[Dylan Anderson]]></dc:creator>
            <pubDate>Wed, 21 Aug 2024 20:11:06 GMT</pubDate>
            <atom:updated>2024-08-21T20:11:06.668Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[A No-BS Story of Why You Need A Data Strategy]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.datadriveninvestor.com/a-no-bs-story-of-why-you-need-a-data-strategy-30a8503b01ab?source=rss-e0f73d4872fc------2"><img src="https://cdn-images-1.medium.com/max/2600/0*jLuZetyngMPsCtj9" width="6240"></a></p><p class="medium-feed-snippet">The reasons you need to do data strategically, in terminology your grandparents would understand</p><p class="medium-feed-link"><a href="https://medium.datadriveninvestor.com/a-no-bs-story-of-why-you-need-a-data-strategy-30a8503b01ab?source=rss-e0f73d4872fc------2">Continue reading on DataDrivenInvestor »</a></p></div>]]></description>
            <link>https://medium.datadriveninvestor.com/a-no-bs-story-of-why-you-need-a-data-strategy-30a8503b01ab?source=rss-e0f73d4872fc------2</link>
            <guid isPermaLink="false">https://medium.com/p/30a8503b01ab</guid>
            <category><![CDATA[data]]></category>
            <category><![CDATA[business]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[strategy]]></category>
            <category><![CDATA[analytics]]></category>
            <dc:creator><![CDATA[Dylan Anderson]]></dc:creator>
            <pubDate>Sun, 30 Apr 2023 05:31:50 GMT</pubDate>
            <atom:updated>2023-05-01T01:00:26.507Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Consulting Could Be The Start That Your Career Needs]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/illumination/consulting-could-be-the-start-that-your-career-needs-2569a8c717d9?source=rss-e0f73d4872fc------2"><img src="https://cdn-images-1.medium.com/max/2600/0*vz9fLQzNQwW6gWvT" width="5402"></a></p><p class="medium-feed-snippet">A career that affords flexibility and personal growth, why not check it out?</p><p class="medium-feed-link"><a href="https://medium.com/illumination/consulting-could-be-the-start-that-your-career-needs-2569a8c717d9?source=rss-e0f73d4872fc------2">Continue reading on ILLUMINATION »</a></p></div>]]></description>
            <link>https://medium.com/illumination/consulting-could-be-the-start-that-your-career-needs-2569a8c717d9?source=rss-e0f73d4872fc------2</link>
            <guid isPermaLink="false">https://medium.com/p/2569a8c717d9</guid>
            <category><![CDATA[consulting]]></category>
            <category><![CDATA[career-advice]]></category>
            <category><![CDATA[career-paths]]></category>
            <category><![CDATA[personal-growth]]></category>
            <category><![CDATA[business]]></category>
            <dc:creator><![CDATA[Dylan Anderson]]></dc:creator>
            <pubDate>Thu, 13 Apr 2023 12:27:06 GMT</pubDate>
            <atom:updated>2023-04-13T12:27:06.995Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The 5 Soft Skills Data Professionals Need & How to Build Them]]></title>
            <link>https://medium.com/analytics-vidhya/the-5-soft-skills-data-professionals-need-how-to-build-them-a1df614b4336?source=rss-e0f73d4872fc------2</link>
            <guid isPermaLink="false">https://medium.com/p/a1df614b4336</guid>
            <category><![CDATA[skills]]></category>
            <category><![CDATA[personal-development]]></category>
            <category><![CDATA[communication]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[job-search]]></category>
            <dc:creator><![CDATA[Dylan Anderson]]></dc:creator>
            <pubDate>Fri, 10 Mar 2023 14:44:10 GMT</pubDate>
            <atom:updated>2023-03-10T14:44:10.412Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A roadmap to soft skill success!</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*XLZXb0rVEgLUJxiG" /><figcaption>Photo by <a href="https://unsplash.com/@krakenimages?utm_source=medium&amp;utm_medium=referral">krakenimages</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>The newest trend data influencers are focusing on is talking about how you need to know soft skills to get the job.</p><p>Every other post on LinkedIn is <strong>“Soft skills &gt; technical skills” </strong>or “<strong>Learn how to communicate</strong>”.</p><p>So yes, you do need to improve your soft skills as a data person.</p><p>But what are those soft skills? How do you develop them? What value do you need to showcase?</p><p>As somebody who works with, hires and manages data engineers, analysts, scientists, and architects, here is my list of five.</p><h4>1. 𝐂𝐨𝐦𝐦𝐮𝐧𝐢𝐜𝐚𝐭𝐢𝐨𝐧 📡</h4><p><strong><em>Why it is important</em></strong> — Data people suck at communication. FULL STOP.</p><p>Therefore, if you can do this well, you will excel in your career.</p><p>It matters because few people in the business understand what data people do without your ability to <strong>translate it into the language of the business</strong>.</p><p>And the reason everybody mentions it is, <strong>if you fail at communicating, all your data work will be useless</strong> and just sit in an unused file folder on your computer.</p><p><strong><em>How to develop it</em></strong> — Three things:</p><blockquote>1) Pay attention to other leaders and how they communicate so you can speak their language</blockquote><blockquote>2) Edit your work and attempt to make your work more concise</blockquote><blockquote>3) Draw out the relevant ‘so what’ insights, results, and implications that people care about, not the high-level data that doesn’t mean anything without further context.</blockquote><p><strong><em>What to showcase</em></strong> — Talk about the process of <strong>how you communicated data outputs</strong> <strong>and</strong> <strong>insights to other stakeholders in the business</strong>, and what the <strong>result of that ended up being</strong>. Showing how you accomplished this demonstrates the ability to translate between the data and the business world, <strong>a huge differentiator on any data team</strong>.</p><h4>2. 𝐏𝐫𝐨𝐛𝐥𝐞𝐦 𝐒𝐨𝐥𝐯𝐢𝐧𝐠/ 𝐀𝐧𝐚𝐥𝐲𝐭𝐢𝐜𝐚𝐥 𝐓𝐡𝐢𝐧𝐤𝐢𝐧𝐠 💡</h4><p><strong><em>Why it is important</em></strong> — Companies don’t pay top dollar for data resources that can’t give the business a competitive edge.</p><p>Low code tools, commercial ML algorithms and countless data products can do a large swath of the data responsibilities, but linking those activities to deliver better business results still needs to be done by a smart employee. To do this, data people <strong>need to know how to properly approach and solve organisational problems in a way that is efficient, effective and inclusive of all stakeholders</strong>.</p><p><strong><em>How to develop it</em></strong> — The key to build this skill is to <strong>think about what matters to your audience</strong>.</p><p>Then practice creating frameworks and processes for how to get to the answer and how you find what you need to include in that process. The key is to <strong>ensure the logic flows from beginning to end</strong>, <strong>it is simple to follow</strong> and any output you create <strong>has evidence to back it up.</strong></p><p><strong><em>What to showcase</em></strong> — Demonstrate your ability to <strong>solve difficult problems in a creative way that is different </strong>from the usual ‘I approached a problem and solved it’ response.</p><p>And make sure you <strong>highlight the value you created</strong> (either monetary or resource-wise). For example, tell me when you cut your dev time in half by identifying how to unlock the business value at the beginning of the process rather than at the end.</p><h4>3. 𝐒𝐭𝐚𝐤𝐞𝐡𝐨𝐥𝐝𝐞𝐫 𝐌𝐚𝐧𝐚𝐠𝐞𝐦𝐞𝐧𝐭 👋</h4><p><strong><em>Why it is important</em></strong> — A project done entirely within the data team is a project that <strong>never makes it out of the data team</strong>.</p><p>Learning to engage with stakeholders outside of your immediate team helps ensure that there is business logic to complete the project, that everybody knows what is going on and that the end result is as effective as possible. This skill is especially relevant for data managers and leaders who are doing this on a daily basis.</p><p><strong><em>How to develop it</em></strong> — Build yourself a <strong>stakeholder management list of to dos</strong>. This might include identifying what stakeholders may need to be involved, asking what kind of contact they want, and scheduling meetings to keep individuals involved in projects.</p><p>In addition to this practice writing <strong>concise emails</strong>, <strong>keeping meetings organised</strong> with an agenda and figuring out how to best <strong>build trust with key stakeholders</strong>.</p><p><strong><em>What to showcase</em></strong> — Talk about how you <strong>brought together different departments/ individuals</strong> across the organisation <strong>to produce something special</strong>.</p><p>What was your process for doing this and how would you replicate it? Finally, stress how aware you are of the importance of bringing in stakeholders from across the organisation, I have never seen a leader not be impressed by that kind of cross-functional awareness.</p><h4>4. 𝐀𝐭𝐭𝐞𝐧𝐭𝐢𝐨𝐧 𝐓𝐨 𝐃𝐞𝐭𝐚𝐢𝐥 🔍</h4><p><strong><em>Why it is important</em></strong> — Finding somebody who has strong attention to detail skills is basically about <strong>finding somebody you trust to execute</strong>. And <strong>this is so difficult to find</strong>!</p><p>Why? Because people often neglect this skill.</p><p>In the data world — where the simplest mistake can ruin a multi-million dollar project — attention to detail can often be the <strong>subtle difference between two strong candidates</strong>.</p><p><strong><em>How to develop it</em></strong> — After you complete something, <strong>look over it 3–4x more</strong>. Identify where you are most likely to make errors in code or communications <strong>so you know where to look moving forward</strong>. Most of all, <strong>get people to review your work</strong> and tell you where they made changes. Do all of this until you can spot the smallest of errors in your work.</p><p><strong><em>What to showcase</em></strong> — The key is to <strong>demonstrate your ability to find needles in a haystack</strong>, especially ones from other team members or ones that were missed by managers.</p><p>An example could be when your debugging skills or ability to find spelling errors in code saved the team a week in dev time or fixed a problem that would have crashed the app in production. <strong>Prove yourself as a last line of defence against errors</strong>, and you will be handsomely rewarded!</p><h4>5. 𝐈𝐧𝐢𝐭𝐢𝐚𝐭𝐢𝐯𝐞 📈</h4><p><strong><em>Why it is important</em></strong> — Data is not an industry that stays still.</p><p>No, it is <strong>constantly evolving and advancing</strong>.</p><p>I expect top performers to do the same, otherwise you will just be seen as a cog in the machine, not an engine that drives the team forward.</p><p><strong><em>How to develop it</em></strong> — Figure out <strong>what matters to your team outside the normal job description</strong> and couple it with what you enjoy doing.</p><p>Then take 1 hour a week to create great outputs that <strong>get to your team’s goal in this extra-curricular activity or just demonstrate your skills in the area</strong>.</p><p>By sharing your wins and learnings, you prove that you can go above and beyond in areas that matter to your team and your own expertise.</p><p><strong><em>What to showcase</em></strong> — Talk about how you <strong>go beyond the role description at your past jobs</strong> and are able to accomplish impressive feats that provide value to your company.</p><p>This can include job-related activities or can be bits related to leadership, management and culture-building! Showcasing personal examples of initiative (blogs, portfolio, website, etc.) is also a great talking point.</p><h4>Final Thoughts</h4><p>Soft skills are essential. Period.</p><p>And from my years in data, these are five of the most crucial. So write them down, follow the steps to gain them and take action to improve them. Just remember:</p><ul><li>Communication — Carve out the ‘So What’ for your audience</li><li>Problem Solving/ Analytical Thinking — Be logical with your deliverables</li><li>Stakeholder Management — Build trust with key people</li><li>Attention to Detail — Become the last line of defence</li><li>Initiative — Be known for going above and beyond</li></ul><p>Take these away and kill it at your data job, and hopefully this article is way more informative than the 1,000th “Soft Skills Matter” post on LinkedIn with no context or roadmap.</p><p>Thanks for reading and please let me know if you find this useful!</p><p><em>Want to keep getting the best career value for your social media dollar? Follow my </em><a href="https://dylansjanderson.medium.com/"><em>Medium</em></a><em> and </em><a href="https://www.linkedin.com/in/dylansjanderson/"><em>LinkedIn</em></a><em> advice about excelling in your data, consulting and corporate career! No fluff, no bs, some humour, all value (otherwise I will give you your $0 back)!</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a1df614b4336" width="1" height="1" alt=""><hr><p><a href="https://medium.com/analytics-vidhya/the-5-soft-skills-data-professionals-need-how-to-build-them-a1df614b4336">The 5 Soft Skills Data Professionals Need &amp; How to Build Them</a> was originally published in <a href="https://medium.com/analytics-vidhya">Analytics Vidhya</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>