Enterprise Intelligence Group

Retailers Remember Transactions. They Forget Decisions.

A retailer that can reconstruct what sold, years deep, in seconds may still be unable to reconstruct why an item was delisted, what the team believed at the last category reset, or which constraint shaped a regional assortment. That gap, the Merchant Memory Problem, matters more the moment you point AI at your business.

Why preserving commercial reasoning may be as important as preserving the data itself.

An illustrative scenario, not a client case: a new category director inherits a portfolio. Somewhere in the range is an item that, by every report she can pull, should have been delisted two resets ago: thin margin, middling velocity, duplicated by two better performers. She raises it in her first trading meeting and an experienced merchandiser winces. "We tried that in 2022. It's the item that anchors the trip for the wholesale-club shopper — take it out and the basket goes with it. We learned that the hard way."

Nothing in any system she has access to says this. The delist, the reversal, the basket analysis that justified the reversal, the argument that produced it — all of it happened, all of it cost real money to learn, and none of it was written down anywhere she could have found it. The company paid for that lesson once. It was about to pay for it again.

This is the Merchant Memory Problem: the reasoning behind a commercial decision is held only by the people who made it, not in a record the business can find. When that is the case, a successor inherits the transaction history but not the lesson, and the lesson has to be paid for again.

Very good at what. Very poor at why.

Where a retailer has invested in systems of record and retained the data, it can answer what sold, what was ranged, what price changed, what promotion ran, what margin was achieved and what inventory moved, going back years. Those systems make the what durable.

The why is another matter. Ask the same organization:

• Why was this item introduced?
• Why was that one removed?
• What assumptions were made at the time — and which of them proved wrong?
• What alternatives were considered and rejected?
• Why does one region carry a different assortment?
• Which supplier constraint or operational reality shaped the decision?
• What did the team actually learn from the last category reset?

If the reasoning was never recorded, the honest answer depends on who is still in the building.

Where it is not recorded, the reasoning behind commercial decisions sits in trading-meeting conversations, in the middle slides of archived decks, in email threads and in the memory of the people who made them. After the decision, that reasoning can fade. Once the deck is archived, it can become hard to find. When the person leaves, it can go with them.

Transaction history is not decision history

The distinction matters because the two answer different questions. Data tells the organization what happened. Commercial intelligence explains why it happened — and the why is where the judgment lives.

Where this gap exists, it can show up in ordinary operation before any technology enters the picture. The patterns below are illustrative, not client findings:

Category resets can re-litigate settled arguments. At a reset, experienced merchants may have to re-explain why the fixture is laid out the way it is, why a marginal-looking item survives and why the obvious price move was tried and failed. The organization knows these things but cannot retrieve them.

Regional differences can become folklore. For example, a group of stores may carry a different pack architecture because of a distribution constraint that was real when the decision was made. If only the resulting assortment was documented, not the constraint, nobody can say whether it still applies.

Successors can relearn by expensive mistake. A capable replacement for a long-serving buyer does not lack talent; they lack the accumulated map of which suppliers move on terms, which promotions do not pay back in a category and which rules have exceptions. Rebuilding that map through trial and error takes time, and the errors cost margin.

Post-mortems have less to work with. When a range change disappoints, the useful question is "what did we believe when we made this call, and which belief was wrong?" If the beliefs were never recorded, the review can only inspect the outcome — so the same class of mistake stays available to be made again.

Why AI raises the stakes

This has often been treated as a cost of doing business. Two things can change that.

The first is turnover and workload. Where tenure is shorter, teams are leaner or each merchant makes more decisions, there is less informal apprenticeship to carry reasoning from one generation of merchants to the next.

The second is AI. When a business points AI systems, such as search, assistants or agents, at its enterprise data, the theory is that the organization's knowledge is in there somewhere. But an AI system trained on transactions, reports and structured data can only learn what those artifacts contain: outcomes. It sees that the item was delisted and relisted. It cannot see the basket argument, the wounded supplier relationship, or the merchant's hard-won rule about club-shopper anchors — because no system ever captured them.

The risk is a familiar failure mode. A large language model connected only to enterprise documents can be fluent about what happened and unable to explain why, producing confident summaries of outcomes detached from the commercial context, constraints and exceptions that produced them. Pointed at an archived deck whose assumptions have quietly expired, it can apply a stale rule confidently and quickly.

In that situation, a better model does not supply the missing reasoning. Capturing the reasoning does.

Retail Memory

Enterprise Intelligence Group believes retailers need to treat this missing layer as an asset in its own right and manage it deliberately. We call it Retail Memory.

A useful working model:

Retail Memory = Data + Decisions + Context + Assumptions + Outcomes

Retail Memory is not another system of record, and it is not a document repository with a search box. It is a governed layer of institutional knowledge that preserves, alongside the data the organization already keeps:

• Decision history — what was decided, when, by whom, for which category, region or channel;
• Commercial reasoning — the argument that carried the decision, including the options considered and rejected;
• Assumptions — what the team believed about demand, competition, supplier behavior and operational constraints at the moment of decision;
• Context and constraints — the supplier terms, distribution realities and strategic priorities that shaped the call;
• Exceptions — where the standard rule deliberately does not apply, and why;
• Outcomes and lessons — what happened next, which assumptions held, and what the organization concluded.

Governed matters as much as captured. A memory layer nobody validates decays into a folder of opinions; the entries need owners, review, and an honest mechanism for retiring reasoning that has expired. Not everything an experienced merchant believes is still true — the point is to make beliefs inspectable, not to enshrine them.

What it makes possible

Built this way, Retail Memory can support the work retailers already do and the AI they intend to deploy.

For the organization today: category management and assortment planning that starts from the accumulated reasoning instead of re-deriving it; onboarding that hands a successor the map, not just the territory; executive decision support that can answer "have we tried this before, and what happened?" with evidence.

For AI specifically: retrieval-augmented systems become genuinely useful when what they retrieve includes decisions, assumptions and outcomes — not just documents. A category knowledge agent that can cite why the fixture is laid out the way it is, which alternatives were rejected and what the last reset taught the team is a fundamentally different tool from one that summarizes reports. And an agent proposing a range change can be required to check its proposal against recorded prior outcomes — the closest thing to institutional judgment a system can be given.

None of this replaces the merchant. The judgment stays human; that is precisely what makes it worth preserving. The objective is narrower and more urgent: to stop that judgment disappearing every time a decision is made, a presentation is archived, or an experienced person leaves the business.

The executive question

If a business can reconstruct any transaction from the last decade in seconds but cannot reconstruct the reasoning behind last year's most expensive category decision, the gap is not one that more transaction data will close. The transactions record what was done. If the reasoning was never written down, it is held by whoever was in the room at the time, and it leaves the business when they do.

An organization that closes that gap can test the benefit directly: track how long a successor takes to make an independent category decision, and how often a reset repeats a change the record shows was tried before. It also gives AI systems decisions, assumptions and outcomes to retrieve, not only documents, so its records hold not only what the business did, but why.

The data is already remembered. The question worth asking your merchandising organization this quarter is simpler and harder: where do our decisions go when the meeting ends?

Apply this to your business

EIG helps retail, CPG and distribution teams clarify business requirements, evaluate technology choices and plan implementation. The starting point is the decision you need to make—not a requirement to adopt AI.

Explore EIG's services · Discuss your project

The EIG Library  ·  Insights  ·  AI Maturity Assessment  ·  Contact