<?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 Derrick Ellis on Medium]]></title>
        <description><![CDATA[Stories by Derrick Ellis on Medium]]></description>
        <link>https://medium.com/@derrick.ellis?source=rss-59412edf57df------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*Pk81VqKhP3FeVzC9</url>
            <title>Stories by Derrick Ellis on Medium</title>
            <link>https://medium.com/@derrick.ellis?source=rss-59412edf57df------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 17:04:39 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@derrick.ellis/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[Vibe Coding: Hobbyist to Wizard to… Banned]]></title>
            <link>https://blog.stackademic.com/vibe-coding-hobbyist-to-wizard-to-banned-0d02aaa7e5d5?source=rss-59412edf57df------2</link>
            <guid isPermaLink="false">https://medium.com/p/0d02aaa7e5d5</guid>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[claude-code]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[vibe-coding]]></category>
            <dc:creator><![CDATA[Derrick Ellis]]></dc:creator>
            <pubDate>Thu, 26 Mar 2026 20:24:12 GMT</pubDate>
            <atom:updated>2026-03-27T10:56:22.368Z</atom:updated>
            <content:encoded><![CDATA[<h3>Prompt Engineering circa 2023</h3><p><strong>Three years ago</strong><em>–back in 2023–</em>I started experimenting with AI-generated content. I spun up a Wix blog, ambitiously titled “<strong><em>Prompt Prodigy</em></strong>,” as a playground to explore the potential of tools like <em>MidJourney</em> and <em>ChatGPT</em>. My question was simple: “Could I create a full multi-author blog–<em>teaching prompt writing–</em>by only writing prompts?” The answer? <strong>Almost.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1011/1*UZluBpmQK65Qp_EdhGoxPQ.png" /><figcaption>Sample article teasers from the Prompt Prodigy Blog, 4/2023 — side note, <em>humor</em> is really hard for AI still!</figcaption></figure><p>Using prompts in chat sessions and <em>Discord (MidJourney)</em>, I built a roster of ‘regular contributors,’ each with distinctive bios, writing styles, favorite topics, even MP3 profile recordings “in their own voice” (thanks to <strong>beta</strong>.elevenlabs.io). Over time, these ghostwriters co-authored two dozen blog posts, complete with vivid AI-created imagery. The articles were genuinely engaging — but achieving a consistent tone and coherent narrative across posts (with much more limited context-windows) demanded a lot of trial-and-error. At the time, I concluded: <strong>Close, but not quite there yet.</strong></p><h3>From Content Creation to Experimental Apps</h3><p>After the blog, I moved into smaller experimental features, building AI-driven tools like recommendation apps for board games, movies, and books. These apps generated suggestions based on user history and leveraged APIs (LLMs and other sites like <em>Boardgamegeek</em>) to layer generative AI functionality on top.</p><p>Over the next two years, I played around with various ideas, keeping tabs on the quickly evolving genAI landscape. I even developed tools and small applications using nothing but chat interfaces in <em>Gemini</em>, <em>Grok</em>, and <em>ChatGPT</em>. It should be no surprise, this isn’t the best approach.</p><h3>Scaling Up — Crafting Larger Projects</h3><p>This past fall, I decided it was time to tackle something more ambitious. Using <em>Cursor, ChatGPT and Grok Imagine</em>, I prototyped a website for my son’s “Driveway Basketball League.” That project led me to try <em>Replit, Grok Imagine </em>and<em> Google Colab </em>(for ML/NN training) on my next experiment: I spent a month building <strong>Board Game Dojo</strong> (<a href="https://boardgamedojo.com/">https://boardgamedojo.com</a>), a tribute to one of my favorite games — <em>Century: Golem Edition</em>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2Q84ZqxKGPKfC_dR7nHR1Q.png" /><figcaption><a href="http://boardgamedojo.com/">Boardgamedojo.com </a>— a Fan-version of Century: Golem Edition I made over the Holidays</figcaption></figure><p>The Dojo project was a really fun build. The idea was a training-ground where you could try out novel strategies, practice coming from behind, etc. I designed an AI opponent — my “<em>Sensei AI</em>” — which could make legal moves and even win. But, despite eight iterations using heuristic rules, machine learning models, neural networks, and other approaches, the AI wasn’t truly competitive. As much fun as I had crafting this fan-version of the game, I eventually hit a developmental wall and decided to move on to fresh challenges.</p><p>Around this time, I re-evaluated my tools. <em>Replit</em> had become my workhorse and I’d gotten comfortable with its capabilities, but its cost was getting out of hand for a hobbyist project like the Dojo. The world was starting to talk about <em>Claude Code</em>–I decided to jump in and use it for most of my development work. <em>Replit </em>is still my primary DevOps pipeline because it is really good at it (though lately I’m mostly working in my localhost).</p><h3>Building Momentum</h3><p>Fast forward through the next couple of months (January through March)–I scaled from the $20/month Claude subscription to the Max subscription–building out a bunch of interesting prototype applications, to name a few:</p><ul><li>an <strong>AI</strong> <strong>Board Game Prototyping Engine</strong> — concept brainstorming, full suite of design tools, data-driven assets, rulebook generator, video teaser generator, app stage (meant for playtesting) — all of these features have AI-native, AI-first functionality.</li><li>an <strong>AI-native QA Application</strong> — with npx registration, a clever issue-logging with screen grabs, annotations, issue queues and an ability to say “Claude go get my bugs and fix them” to really tighten the loop.</li><li>an<strong> LLM API Gateway</strong> — because AI-native application API calls can get expensive and you need to be able to track that cost AND provide fallbacks if an endpoint goes down</li><li>a <strong>Peer-to-Peer Video Chat</strong> — Google meet / Zoom style video chat with screenshare, live transcription, support up to 6 people. Also npx registered — and can even be white-labeled with custom branding</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Qd0DAPkLAvNCjRA5B0iNUg.png" /><figcaption>From scratch component design canvas for my board game concept’s Player Board w/AI-native utilities</figcaption></figure><p>Each one of these applications is really interesting —<strong> and all of them started with an, “I wonder if I could…”</strong> Do they work? Absolutely! Are they done? Not yet. Building an MVP is fast, hardening &amp; productizing is slower (that is foreshadowing). I’ve been pouring evenings, weekends, and many late nights into them, often hitting token limits in my <em>Max </em>subscription–<strong>which my fellow Vibe-Coder and frequent collaborator </strong><a href="https://www.linkedin.com/in/sgeorgievbg/"><strong>Stanislav Georgiev</strong></a><strong> jokingly calls “token jail.”</strong></p><p>As I transitioned from browser-based apps to the CLI, I tried to fully leverage my new workspace. I’ve often had five <em>Claude </em>terminal sessions open at once, each tied to a different <em>Git </em>repo and project. I began optimizing my process further, deploying multiple agents within a single terminal session to parallelize efforts whenever possible. <em>It felt like I was a wizard in a world just waking up to magic</em> — exhilarating, but exhausting. Truly Wild.</p><h3>Managing Mental Fatigue</h3><p>The constant context-switching takes its toll. The monotony of approving tools and bouncing between tabs started consuming more of my time–operational friction was sapping the fun from creating. I tried to find process optimizations–such as focusing on planning docs with each session and then triggering the build simultaneously in all sessions; this turned the monotony into a 20-min slog but at least compressed non-creative time somewhat. It still wasn’t enough.</p><p>Just as I built the QA and LLM Gateway apps to help me build better projects, I built another tool that I call the <strong>Office</strong>: a project management app with persistent sessions, clean UX, and specialized agents (‘VPs’) responsible for aspects like Security, R&amp;D, Engineering, Marketing, etc. The application quickly grew with ‘nice to haves’ like: a decidedly non-terminal chat interface, profile pics, a voice chat option (<em>WebspeechAPI</em>, <em>Whisper </em>and <em>Piper</em>) and even an <em>OpenClawfice</em>-inspired floorplan–where each of the self-registering rooms is a project running <em>Claude </em>locally on my hard drive (using <em>Phaser</em>, each with their own <em>git </em>repo!).</p><p>“Sam,” my head of the Office (a <em>Claude-Opus</em> model assigned to oversee the orchestration), coordinated tasks flowing into project rooms, each running locally on my hard drive. Using dynamic model assignment based on task complexity, I optimized token spend, reserving <em>Claude Opus</em> for complex work and <em>Sonnet &amp; Haiku</em> models for simpler tasks. I’m able to have meetings with Arctic (Security) to ask about room locks or other anomalies logged–and can ask Morgan (R&amp;D) to design and build experiments overnight. <em>But the big evolution here for me, was in leveraging different Model tiers for different types of tasks.</em></p><p>The Office is a great concept, but I found I was spending more of my time working to improve the Office-itself and less time exploring other interesting projects. As it bloated with features, its inefficiencies started showing: notifications overload, file contention issues, tasks spawning to handle project management-type activities instead of building–too much overhead and often with coded-but-uncommitted work. This was not working out the way I wanted, so Claude and I had a productive “heart-to-heart” about streamlining.</p><h3>“Forge <em>with Sam</em>”: Stripping Down for Efficiency</h3><p>The new version of the Office I’m building, <strong>Forge with Sam</strong>, focuses solely on streamlining workflows while minimizing distractions. You can launch any application registered with Forge with Sam from within the dashboard, you have metrics on what has changed, how many tasks for each project have been completed–how long they took and other metrics geared toward seeing at-a-glance health of the system.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*k8c7N_yugHHc33dYZGmwzQ.png" /><figcaption>Each project has a current status, and ‘easy buttons’ to initiate a queue of hardening tasks (Grok Imagine changed project names for me!).</figcaption></figure><p>Each project registered has a “release readiness” tag of Prototype &gt; Functional &gt; Deployable &gt; Production. I have an <em>easy button</em> that scans the project for missing features that are preventing it from moving towards the next step &amp; a way to create tasks for each of those work items to queue up as work. I’ve got another button for <em>hardening</em> which focuses on an <em>adaptive</em> queue of tasks needed for ‘getting the project ready for a production release.’ This audit &amp; queue work model is exactly the kind of functionality that the Office wasn’t able to support; this work of streamlining a prototype to production is intended to handle the stuff that I don’t enjoy doing but is critical to get right.</p><p><em>Claude </em>and I created a tool that had an adaptive LLM Provider and Model fallback chain–for Complex tasks I could use <em>Claude Opus</em> as my primary–if I ended up running through my tokens but still had work to do, I could dynamically switch to <em>OpenAI’s Codex</em>. I could also fall-back and up through different models–I reached my <em>Sonnet </em>weekly limit and the system fell-up to work with <em>Opus </em>even for basic tasks because that was the logic I’d implemented in my chain. The metadata I’d get back from these tools told me when tokens would be available, so instead of spamming the endpoint, I could just switch over again automatically.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8hn5F7Fs69GLCuqBeeR3bg.png" /><figcaption>My model fallback and status grid in the successor to the Office application.</figcaption></figure><p>With the Office, I saw that having an AI agent review another AI agent’s work was largely an unnecessary expense, <em>adding ceremony without real value</em>. Instead of a rigid coherence or quality gate with a code-review, I pivoted to a programmatic “<strong>Socratic Sidecar</strong>.” This mechanism interrupts my <em>TaskAI Agents</em> only when patterns from previous errors were recognized — asking simple questions like, “Did you build tests for this feature?” or “How did you solve this before?”</p><p>This model most closely matches my own role as an operational harness when I’m manually babysitting the terminal. I’m not correcting <em>Claude’s </em>code, but if I see something odd–or it seems like <em>Claude </em>is stuck in a loop–I usually interrupt. The <em>Socratic Sidecar</em> is designed to inject context and then resume the <em>TaskAI’s</em> workflow automatically, freeing me from staring at endless terminal streams.</p><h3>The Anthropic Ban</h3><p>While building and testing the <strong>Sidecar’s</strong> multi-provider interrupt-and-resume logic, my <em>Anthropic</em> account was suddenly banned. They issued a full refund for the month’s subscription and overage charges, but provided no detailed explanation.</p><p>After reviewing their Consumer Terms of Service, the ban appears to stem from routing interactions through a custom (localhost) front-end layered on the official CLI utility.</p><p>At the time, I wasn’t intentionally trying to circumvent their rules — I was simply building efficiency tools on top of the tokens I had legitimately purchased, as a natural extension of my vibe-coding experiments. I had assumed this fell within acceptable personal use.</p><p><em>Anthropic</em> terms state clearly that OAuth tokens from <em>Pro/Max</em> subscriptions are meant for direct use in their official interfaces (<em>Claude.ai</em> and <em>Claude Code</em>), not for custom wrappers or automated orchestration layers. This fits the broader industry pattern of keeping subscription usage within intended human-driven experiences while steering heavier automation toward paid API keys.</p><p>I’ve seen similar reports from other heavy users. My appeal is still pending amid long support queues. <strong>If the ban is lifted, I’ll stick strictly to their official tools going forward and avoid any custom front-ends on the consumer subscription.</strong></p><p>Losing access to my primary coding partner mid-project has been a real setback after months of intense evenings and weekends. It’s a reminder that subscription “unlimited” access has real guardrails — especially once you start adding your own optimizations. After three months of heavy effort, it feels strange to suddenly lose my main workhorse tool.</p><p>These projects — spread across the last three years — have deeply sharpened how I think about building with AI. It’s no longer just about clever prompting; it’s about designing resilient, cost-aware, human-in-the-loop systems and understanding the real work required to take them from prototype to production.</p><p>Still, the iterative process of building, failing, adjusting, and learning has made the whole journey incredibly meaningful. <strong>We’re living in WILD times.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0d02aaa7e5d5" width="1" height="1" alt=""><hr><p><a href="https://blog.stackademic.com/vibe-coding-hobbyist-to-wizard-to-banned-0d02aaa7e5d5">Vibe Coding: Hobbyist to Wizard to… Banned</a> was originally published in <a href="https://blog.stackademic.com/">Stackademic</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Looking Beyond the Golden Record: Unified Profiles in Salesforce Data Cloud]]></title>
            <link>https://medium.com/@derrick.ellis/looking-beyond-the-golden-record-unified-profiles-in-salesforce-data-cloud-ec23bf17bfb5?source=rss-59412edf57df------2</link>
            <guid isPermaLink="false">https://medium.com/p/ec23bf17bfb5</guid>
            <category><![CDATA[salesforce]]></category>
            <category><![CDATA[golden-record]]></category>
            <category><![CDATA[salesforce-data-cloud]]></category>
            <category><![CDATA[salesforce-architect]]></category>
            <category><![CDATA[master-data-management]]></category>
            <dc:creator><![CDATA[Derrick Ellis]]></dc:creator>
            <pubDate>Tue, 03 Dec 2024 15:50:30 GMT</pubDate>
            <atom:updated>2024-12-03T16:20:25.285Z</atom:updated>
            <content:encoded><![CDATA[<p><em>This post is a collaboration between </em><a href="https://medium.com/u/59412edf57df"><em>Derrick Ellis</em></a>, <em>Director Success Architect — Data Cloud Customer Success</em> &amp; <a href="https://medium.com/u/c2d2d333212e"><em>Stanislav Georgiev</em></a><em>, Director Product Management — Data Cloud Identity @ Salesforce.</em></p><p><strong>It’s time to talk about Data Cloud’s unified profile. </strong>We know what you’ve been thinking — this is your new customer master, your shortcut to that perfect <em>‘golden record.</em>’ If you’re nodding along with this statement, it’s time to reset some of your expectations. We’re going to take some time to help you really understand the unified profile, enabling you to leverage it effectively in your enterprise architect designs.</p><p>In discussions with customers, architects, and partners on these topics, we’ve found that many don’t understand the difference between a golden record and a unified profile. Sometimes this discovery makes them worry that they won’t be able to use unified profiles how they intended. Don’t worry, Data Cloud’s unified profile can still help you meet your business needs — you just have to use it differently than you might be thinking. You’ll be OK and you might just get excited by all of the potential! Ready? Let’s do it!</p><p>Let’s start with some basic concepts to build a shared foundation for our discussion. By the end, you’ll see unique ways the Unified Profile can benefit your business. We’ll also hint at future articles exploring Data Cloud’s relationship to your master data management (MDM) strategy. There’s a lot to cover, but we have to start somewhere! In this blog post, we’ll focus on<strong> </strong>four things.</p><p><strong>(1) Data Cloud as a system of reference</strong><br><strong>(2) The mutability of the unified profile</strong><br><strong>(3) The key ring model versus the golden record approach </strong><br><strong>(4) Additional advantages of this model</strong></p><p><strong><em>Spoiler Alert</em></strong><em>: Data Cloud’s Unified Profile is NOT a golden record and that’s a good thing!</em></p><h3>Data Cloud is a System of Reference</h3><p>By now you’ve heard that Data Cloud brings all your customer data together to give you a <em>single view of your customer</em>. Data Cloud should be ingesting data from every system of record you have, bringing all those master data fragments together to find relationships between them and surface that crucial single view. Important to note, <strong>Data Cloud isn’t becoming your new system of record</strong>; it’s using data from your existing systems <em>to find connections </em>between those disparate data sets.</p><p>With such varied customer data, how does Data Cloud find relationships? First, you map the different datasets to a common set of objects (data model objects, aka DMOs) and fields, harmonizing<em> </em>all data models into a single shared model within Data Cloud. Next, you explore data across systems, using the rules you set in identity resolution to link records, thus allowing you to uncover new insights, patterns, and trends.</p><figure><img alt="Data Cloud is a System of Reference that harmonizes all System of Record data into a common set of objects and fields" src="https://cdn-images-1.medium.com/max/733/1*EETMVq3n3GtypFln9dB6zA.png" /><figcaption><strong>Figure 1: </strong>Data Cloud’s Unified Profile does not master data, it <em>links</em> existing data together based on the relationships found through identity resolution.</figcaption></figure><p>The unified profile isn’t taking over your data mastery; it’s helping you discover the full view of your customer across all of the records you already have! Additionally, because each record’s data lineage is preserved in the Unified Profile, you can trace every bit of information back to the original system of record. That’s why <strong>Data Cloud is a system of reference</strong>.</p><h3>Data Cloud’s Unified Profile is Mutable</h3><p>Data Cloud’s unified profile is designed to evolve and improve automatically as your customer data evolves and improves. Your data isn’t perfect (don’t worry, we won’t tell anyone!). Some of your systems of record need quality spring cleaning and data stewardship. Can you still bring flawed or incomplete data into Data Cloud? Absolutely! How else can we help improve things and provide unique insights? Most systems have messy data — but the data you bring doesn’t have to be of the highest quality data to help your business.</p><p>Your customer data will change over time, becoming more complete, accurate, and robust through customer actions or structured internal data quality efforts. As identity resolution incorporates the latest data from various systems of record and the relationships between fragments of master data change and improve, your unified profiles also evolve. This mutability ensures you always have the latest and best unified profile that can be built from your customer records. For example, two records that were once related to different unified profiles might now be linked through a single unified profile after the latest identity resolution run.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xYQu2cLh6uv6ZmZtVBozGw.png" /><figcaption><strong>Figure 2: </strong>The relationship between source records and the Unified Profile can change over time.</figcaption></figure><p>What does this mean for the UUID (the Unified Profile’s ID)? It means you shouldn’t use it as a permanent ID in your external applications — <strong>it’s not your new Subscriber Key or Customer ID!</strong> Using an old UUID means missing out when your profiles are updated with the latest updates to your data, leading to an outdated single view of the customer.</p><p>Unified profiles prioritize accuracy first and stability second. Will the makeup of each unified profile’s set of linked records change constantly? No. But some relationships will change with each run. To take advantage of improvements in your unified profiles, your model needs to support this reality. Don’t worry, these changes to unified profiles won’t be random, and they’re always based on the rules you give to identity resolution. Expect shifts based on factors like data quality changes in source records, new data source, new records, and changes to your identity resolution match and reconciliation rules.</p><p>We know what you’re thinking, “If the unified profile can change for any given source record, how can we use the profile?” It’s time to talk about the key ring.</p><h3>Unified Profile as Key Ring, not a Golden Record</h3><p>When people think about distributed customer data fragments, they often conclude that they must have a single golden record for each customer to handle different email addresses, phone numbers, or missing information like birthdates or middle names. This can be one way to approach your customer information challenge, and many enterprise MDM solutions are designed to help with this.</p><p>A golden record is the result of mastering the “best” single value (or values) for all customer attributes across the enterprise. Activating that golden record often means propagating the winning values to original source systems, overwriting losing values that originated in those systems, or it means setting up a central service to query the golden record for the best values. Designing and implementing solutions for this model takes significant time and compromise to align all business and IT stakeholders. Consensus is hard-won and involves decisions around business logic for survivorship, merges, splits, and the use of the golden record.</p><p>Data Cloud’s unified profile isn’t attempting to determine the single best values for customer attributes or to propagating “winning” values. Instead, it focuses on discovering relationships across your fragmented customer data and linking those records together. <strong>The goal is to build a holistic view of your customer, not to merge the data into a single “best” record.</strong></p><p><strong>Enter the Key Ring</strong><br>Think of the unified profile as a like you think of the ring of keys in your pocket. Suppose your key ring holds a car key, a house key, and a USB key. These keys are linked together because they belong to the same person. Each key is unique and interacts with a specific external system to perform a specific function.</p><p>Let’s say you have a truck and an SUV — each with its own key on your key ring. Would it make sense to pick the best vehicle key and use it for both? Can the truck key unlock or start the SUV? Can the USB key unlock your house’s front door? Of course not. Each key has a specific purpose, value, and set of information associated with it. Like your key ring, the unified profile holds multiple keys, each with its own purpose.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EIgFCFC25n_aNmirTBRgsA.png" /><figcaption><strong>Figure 3: </strong>Keys on a Key Ring have specific value, purpose and set of information associated with them.</figcaption></figure><p>Extending this metaphor to a key’s related information — how important is it to understand the context and lineage of that data? Both of your vehicles have a key fob for remote starting. The truck’s fob shouldn’t remote start the SUV — that’s specific key-related information. But what if we aggregate related data at the key ring level? Can we activate based on that?</p><p>Consider related data for each vehicle’s remote start capabilities, such as fuel level, door lock status, and proximity to the operator. You need to be close to the vehicle for remote start to work. Does your proximity to the truck matter if you want to start the SUV? No. Does the SUV’s fuel level matter if you’re trying to start the truck? Of course not. This remains true even if both vehicles are the same make, model, and year.</p><p>In this example, it’s clearly incorrect to pick the best vehicle, best fuel level, latest door lock status, or average proximity without considering the proper context — specifically, which vehicle the data is related to. Ideally, we would keep all the data’s context (it’s lineage) intact and maintain the mastered information and relationships associated with each key. This is what Data Cloud’s unified profile does: it retains the full mastered dataset and the relationships for each key linked by the key ring.</p><p>It can be useful to know if you have “2 vehicles” or if you tend to fully fuel your vehicles or wait until they’re almost empty. These are potential Data Cloud Calculated Insights, aggregate insights built from data related all the keys on your profile. We could segment unified profiles associated with more than one vehicle, but when activating or surfacing data to a customer, we drill down to the key-specific level. These unified insights drive significant business value, deeper customer understanding, advanced support interactions, and even loyalty and marketing strategies. However, the primary means of engaging with the customer remains the relationship of the actual data to its specific context. <strong>We want you focus on context-specific activation.</strong></p><p>Looking at Data Cloud’s unified profile, the key ring doesn’t contain 2 vehicle keys, a house key, and a USB key. Suppose a particular unified profile holds various records (with IDs) from Marketing, Service, Sales, Commerce, POS, Order Management, and other systems of record across your enterprise. Each key will have a full set of related data specific to that key ingested into Data Cloud along with it. Sometimes the data contains a single key from each system, and sometimes it may contain several keys from a single system.</p><p><strong>A Word about Duplicate Data</strong><br>We know what you’re thinking — “several keys from a single system on the same unified profile? Bad data! We need Data Cloud to dedupe.” Okay, take a breath. First, understand that Data Cloud isn’t designed to dedupe your data — remember, we aren’t picking winners; we’re linking. Second, while some systems might have bad or unintentional duplicate customer records, it’s also possible that some systems intentionally create multiple records that appear to be the same person. This doesn’t mean you did a bad job; it might be the right choice based on that system’s data model or workflow rules.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZSD_qJadiaQAzwZJFz-y9w.png" /><figcaption><strong>Figure 4: </strong>The Unified Profile’s Key Ring will often have many keys; sometimes multiple keys from each system</figcaption></figure><p>For example, Marketing Engagement excels at tracking subscriber key-level (contact point) details, but its data model doesn’t support more than one email address per contact. The same goes for active mobile numbers in SMS campaigns. If a customer has multiple contact points, you’ll end up with multiple subscriber keys. These aren’t duplicates; they’re the best way to track that data in Marketing Engagement.</p><p>Each subscriber key has specific context — like business unit, personally identifying information (PII), subscription status, and engagement history, including knowledge of current journeys for that subscriber key. These details are specific to the mastered subscriber key-based data and don’t generalize across multiple subscriber keys. This is how Marketing Engagement can have multiple subscriber keys linked into a single unified profile, with each referencing a specific contact point. In practice, this might result in different Subscriber Keys for your work and home email addresses — each has its own distinct related PII and subscription preferences that have been explicitly set by the customer.</p><p>B2C Commerce often lets customers check out as guests, meaning they don’t log in or connect a profile to the order. This requires mapping the Order object in B2C Commerce to both the Sales Order and Individual DMOs in Data Cloud, so the guest can be registered as an Individual on the unified profile. Linking these individual records enables valuable insights, like lifetime value inclusive of both logged-in and guest checkouts, driving strategy across commerce, loyalty, and marketing. Arbitrarily deduping these mapped Individuals would be counterproductive, as we might need to send a back-in-stock notification, order status update, recall notice, or other important communications. These actions should rely on the mastered data lineage rather than some mashed-together record from a bunch of separate systems. The shipping address from a guest order should be used in the context of that specific order, rather than picking a “winner” address from all addresses the customer has used.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6ncgfad_NLsiDmJuf_8qnA.png" /><figcaption><strong>Figure 5: </strong>Unified Profile insights can come from an aggregate view, but through the Key Ring, you can trace back to the specific Order record to find the right shipping information, the product purchased, the transaction date, and other details.</figcaption></figure><p><strong>Key Ring’s Magic: Unified Link DMOs</strong><br>The unified profile’s key ring (found in the Unified Link DMOs) offers the best of both worlds: it maps relationships between source records to retain the full data lineage of those records and their related data. You can use the unified profile to drive aggregate signals, understanding, views, and segmentation while carefully activating along system of record-defined sets of mastered data. <strong>One of Data Cloud’s biggest strengths is its ability to engage directly with operational and transactional systems using keys those systems understand</strong>, driving business workflows and engagement. Can you use these views to improve your data stewardship efforts? Absolutely! Take a look at the AppExchange for Data Cloud-native stewardship applications. To be clear, Data Cloud can’t magically fix everything — each system of record has different business requirements and stakeholders with specific logic to incorporate into the process. One size definitely doesn’t fit all!</p><p>This flexibility and value potential is made possible by the unified profile’s key ring model. As discussed, a golden record’s survivorship decisions involve complex business logic to pick the best values across systems and stakeholders. Data Cloud’s identity resolution includes a process called “Reconciliation” — which <em>sounds </em>like survivorship and even generates a single Unified Individual (or Unified Account) DMO with values picked from the set of keys on the unified profile. <strong>This resulting DMO can be useful as metadata to describe the key ring</strong>. For example, this set of keys belongs to Derrick and has these attributes. Are those attributes ‘the best’? No. They are simply a heuristic, an at-a-glance way to describe that specific unified profile. They aren’t meant for mastering a new golden record — and they won’t always be the “best” values to activate, because “best” always depends on the context in which a value is needed.</p><p>When are these reconciled views useful? Architecturally, you can access these values via the Profile API with very low latency. They are valuable for use in a Lightning Web Component header block (an at-a-glance summary view of basic characteristics) for the unified profile. They can also label or describe some of the aggregate Calculated Insights you’ve generated at the unified level. However,<strong> the real value lies in the Unified Link DMOs</strong>, where you access the underlying key ring. You may surface those reconciled attributes to quickly describe the unified profile, but should follow up by loading specific keys to provide a more nuanced, lineage-based view. Traversing the key ring is also important when you are forwarding actionable data to your internal users and when you’re exposing PII directly to your customers.</p><h3>Other Advantages of Data Cloud’s Key Ring Model</h3><p>Using both the aggregate insights of the Unified Individual and the full lineage and context of your data makes your control in activation incredibly valuable. Beyond honoring the context and mastery of your existing systems of record, there are further advantages:</p><ul><li><strong>Alignment with decentralized data reality</strong>: Data Cloud’s key ring design aligns with your existing enterprise landscape, promoting insights without disrupting data ownership structures.</li><li><strong>Smoother organizational adoption</strong>: This model reduces friction and resistance within your company by linking identifiers rather than overhauling data structures, streamlining cross-functional buy-in.</li><li><strong>Scalability and adaptability</strong>: The key ring model simplifies the addition of new systems or updates to existing ones, enhancing scalability and adaptability in a dynamic enterprise environment.</li><li><strong>Faster integrations</strong>: This approach speeds up integrations by connecting to any enterprise data source without requiring cleansing or pre-processing, removing common blockers.</li><li><strong>No survivorship conflicts</strong>: The key ring model avoids debates and conflicts over survivorship and reconciliation logic, yielding faster consensus and smoother implementations.</li><li><strong>Reduced life-cycle complexity</strong>: By focusing on identifier mapping rather than data attribute consolidation, maintaining the key ring (unified profile) is simpler and less complex over time.</li></ul><h3>In Conclusion</h3><p>Phew! That was a lot to cover, but it was necessary. Let’s revisit where this journey started: <em>“We know what you’ve been thinking — this is your new customer master, your shortcut to that perfect ‘golden record.’”</em> By now, you should be shouting, “<em>No! It’s a key ring!</em>” <strong>The advantage of this model is that it never overwrites your mastered data but links those fragments or records together, providing the best of both worlds</strong> — preserving the full, intact data lineage for your customer data and creating critical insights from across the unified profile. Use those unified profile-based calculated insights to drive deeper understanding and power internal dashboards, business intelligence, segmentation, or outreach strategies based on a single view of your customers. When you activate or engage a customer directly, use the right key and that key’s related data in the right context (or in the right system).</p><p>This combination of the unified profile and easily access to system-specific related customer data gives you tremendous flexibility and removes the implementation barriers that typically accompany survivorship or mastery-oriented solutions.</p><p>With your new understanding of this powerful system and the value it provides, you’re ready to have better follow-up conversations within your company and design quality solutions that take advantage of this value proposition to address your business’ needs. Data Cloud’s relationship to your master data management strategy is another topic worth exploring in-depth, but we’ll tackle that in another post.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ec23bf17bfb5" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>