Golden Records Aren't a Data Project. They're a by-product
In the last piece we made a claim almost in passing that deserves its own article: Which product substitutes for which, which supplier actually ships what, from where, at what price. The relationships that matter most in procurement, are exactly the data your master-data systems only partially hold. Attribute-based systems are good at describing what a product is. They are poor at capturing how products, suppliers, and prices actually behave.
Almost every large distributor and manufacturer has the same answer to this problem, and it's the wrong one: a master-data project.
The project that never ends
You know the shape of it. There's a budget, a steering committee, a consulting partner, and a golden-record engine somewhere in the middle. There are cleansing rules, deduplication passes, a data-governance charter, and a roadmap measured in years. And after all of it, the data is still wrong, not because the team was incompetent, but because of a flaw baked into the premise.
The premise is that perfect master data is a state you can reach: cleanse everything once, declare victory, maintain from there. But master data isn't a state. It's a flow. The moment your golden records are clean, reality moves on. A supplier renames a SKU, a pack size changes, a new price list lands, a product is discontinued and a substitute appears. Static records begin decaying the day they're finished. So the project is never really done; it just runs until the budget does, and then the decay quietly resumes.
The deeper problem is that these projects try to reconstruct, in a warehouse, a truth that is already flowing past them somewhere else.
The operational stream is the master-data stream
Here's the reframe. The ground truth about your products, suppliers, and categories isn't hiding in a data lake waiting to be cleansed. It's moving through your procurement operations every single day.
Every order confirmation tells you something true: this supplier ships this SKU, at this price, in this pack size, from this origin, with this lead time. Every price list update tells you what's current and what's changed. Every supplier email carries relationships, this item substitutes for that one, this line is discontinued, these two are usually bought together. The relational truth that no attribute system holds is precisely the truth that flows through the operational layer.
An agent like Lisa AI already sits in that stream. She reads confirmations, reconciles them against the purchase order, resolves the deviations., That's her operational job. But notice what that job is, structurally: a continuous act of comparing what the record claimed against what reality confirmed. Every one of those comparisons is a free master-data check. That's exactly what a master-data project sets out to do in batch, after the fact, on stale extracts. Doing it that way is slow and expensive; doing it in the operational flow, the way Lisa does, is neither.
Reconciliation is validation
This is the mechanism, and it's worth being concrete about it.
When a confirmed price disagrees with the price in your system, the operational agent treats it as an exception to resolve. But it is also a signal: one of these two numbers is stale, and now you know which record to question. When a supplier ships a substitute, that's not only a fulfillment event, it's a product-relationship fact that just wrote itself. When a confirmation comes back in a different pack size than the PO assumed, that's a unit-of-measure correction landing in your lap for free.
None of these required a data project. They fell out of the work. The agent that resolves exceptions is, as a side effect, continuously observing, validating, and correcting the very records a golden-record initiative is chartered to produce. The clean product, supplier, and category records don't have to be manufactured in a separate stream. They are a byproduct of an operational agent doing its job well, and because they're generated by live operations, they never go stale. Every document that flows through re-validates them. It's a golden record that heals itself.
The honest boundary
This is the point where a less careful pitch would overclaim, so let's be precise about what the byproduct is and isn't.
Operational flow gives you behavioral truth: how products, suppliers, and prices actually act, continuously reconciled against reality. It is extraordinarily good at that, and it's the layer legacy MDM is worst at. What it does not hand you for free is canonical identity, the plumbing question of deciding that "Widget-A" from one supplier and "Widget A / rev 2" from another are, in fact, the same thing. That still needs deliberate governance. Nor does the stream see what never flows through it: the products you don't currently buy, the suppliers you don't currently use. Its coverage is bounded by your activity.
So the honest version is this: the operational agent collapses most of the master-data problem, the relational, behavioral, decay-prone part, which is the expensive part, into a byproduct of operations. It does not abolish the need for canonical identity resolution and governance. It makes that remaining job small, focused, and worth doing, instead of a multi-year excavation that's obsolete on delivery.
From operational buyer to category strategy
Now the part that changes who you're talking to.
Once you have clean, live, continuously-validated product, supplier, and category records, a substrate rather than a snapshot, a whole tier of work becomes possible that was previously manual, slow, or simply not done. Assortment rationalization. Private-label substitution analysis. Supplier consolidation. Price-corridor analysis across the category. Spend visibility that's actually current. These aren't operational-buyer tasks. They're the daily concerns of the category manager and the CPO.
That's the strategic lift. An agent that enters the organization to handle order confirmations, an operational, cost-center conversation, produces, as its byproduct, the exact substrate that category and product management have been trying to build for years. We call that substrate Product Intelligence: the living layer of relational truth about products, suppliers, and categories that accumulates through operations and then feeds strategic decisions. It sits on top of the same intelligence layer, Recall Intelligence, that runs the operational work.
The commercial consequence is a natural expansion path. You land operationally, where the pain is obvious and the ROI is immediate. You expand strategically, because the operational work has quietly produced an asset the CPO's office cannot buy off the shelf and cannot easily build, a category data asset that is current by construction.
The reframe
Stop funding golden records as a destination. They are not a project you charter, staff, and finish; the ones built that way rot before the ink dries. Let clean master data fall out of an operational agent that's already standing in the stream where the truth flows, reconciling records against reality every time a document moves.
Do that, and two things happen at once. The master-data problem stops being a line item and starts being a byproduct. And the procurement conversation stops being about clicks saved and starts being about category strategy, because the same agent that resolved this morning's exceptions has been building, quietly, the data asset your whole commercial organization runs on.
At Recall Space, Lisa handles the operational stream, confirmations, deviations, supplier communication and the clean product, supplier, and category records emerge from that work as Product Intelligence. If your master-data effort has quietly become a permanent line item, that's a good first conversation to have.
