Your Retiring Counter Pro's Brain Is a Catalog Problem, Not a Chatbot Problem

Every distributor with a graying counter staff is running the same experiment right now: interview the veteran, feed the transcript to an AI vendor, ship a chatbot. That's the wrong container for what's actually being lost. Most of what a 30-year counter pro knows isn't a story worth transcribing. It's product judgment, and product judgment belongs in the catalog, not in a conversation.

The knowledge crisis isn't a training problem

Distribution Strategy Group has argued that the industry faces a genuine "knowledge crisis" as technical staff retire, and that generic chatbots trained on the open internet are the wrong answer because they hallucinate on technical specifics. That part is right, and it's not hypothetical in distribution. A wrong torque spec or a wrong voltage rating, delivered confidently by an assistant, isn't just an annoying wrong answer. It's a returned order or a warranty claim.

But DSG's fix is still a better AI layer: a proprietary, "governed" chatbot anchored to internal documents instead of the open web. A separate DSG piece goes further, arguing that "most reps don't know the products they're selling" because the old apprenticeship model can't keep pace with SKU count, and that AI should compress hours of research into seconds so junior reps can operate at veteran competency.

Both pieces treat this as a delivery problem: get the right knowledge to the right person, faster. Neither asks the more basic question. Where does that knowledge live once the veteran is gone, in a form nobody has to re-earn from scratch every time a vendor swaps models?

What's actually in the veteran's head

Sit with a counter pro for a day and almost none of what makes them valuable is a story. It's judgment calls, made instantly, that never got written down anywhere. This gasket replaces that obsolete OEM part number, because they've fielded the return calls that prove it. This actuator bolts to that panel, but only with the older mounting bracket, a fact that lives in nobody's spec sheet. Two bearings rated the same on paper aren't interchangeable in this application, for reasons the manufacturer's PDF doesn't spell out. A duty-cycle rating reads one way in the datasheet and means something narrower out on the shop floor.

None of that is a conversation topic. It's a set of facts about specific SKUs, which is exactly the kind of thing that belongs as a structured attribute on a product record: cross_ref, fits_with, substitute_for, field_duty_note. A chatbot can retrieve those facts once they exist. It cannot originate the judgment behind them, and at most distributors today, nothing captures that judgment at all. APQC has found that only 8% of organizations consistently capture knowledge from departing employees, and 16% don't even try. That gap tracks with a separate Gartner finding that 47% of digital workers struggle to find the information they need to do their jobs. The knowledge was never structured anywhere a system, or a new hire, could find it.

The demographic clock makes this urgent rather than theoretical. The U.S. Census Bureau reports that the share of wholesale trade employment at firms where at least a quarter of the workforce is over 55 rose from 14% in 2000 to more than 40% by 2022, one of the sharpest aging trends of any sector tracked. There isn't a decade to figure this out gradually.

A chatbot is a rendering, not a record

The distinction matters because of what each format survives.

A chatbot answers the question a customer happens to type. A completed attribute answers every future question that touches that field: the ecommerce filter, the AI Overview that cites your spec page, the agent a customer's procurement system runs next year to auto-source a replacement part. Write the fact once, in a structured field, and it works everywhere a catalog gets read, including by software that doesn't exist yet.

That's also where the "governed chatbot" framing runs into a shelf-life problem DSG doesn't really address. The knowledge base under a proprietary assistant is still just documents with a retrieval layer bolted on top. Swap the AI vendor or change the embedding model and you're re-indexing, except the underlying documents were never verified in the first place, only retrieved. An attribute doesn't have that fragility. cross_ref: OEM-4471 to SKU-88291 is either true or false, checkable against a source, and portable to any downstream system you plug in later, chatbot included. Done right, the chatbot is a rendering layer sitting on top of that attribute. It was never a substitute for writing it.

Interview the veteran, but ask catalog questions

None of this is an argument against talking to the retiring expert. It's an argument about what you ask them. Not "walk me through a typical week," but "which SKUs get returned because customers picked the wrong one, and why." Not a transcript destined for a fine-tuning job, but a punch list of missing cross-references, fit notes, and substitution rules, typed into the product record field by field and checked against a spec sheet or a return log. Not left sitting as an unstructured paragraph that a model will later have to guess at.

We built Anglera around this bet: a PIM should hold the structured judgment, and enrichment work is fundamentally attribute completion, not chat transcript curation. Our Top Distributors 2026 index measures exactly this gap, how much verified, structured product data 200-plus distributors actually expose versus how much still sits, uncaptured, in someone's head. Your PIM stores the data. Anglera does the work of getting the veteran's judgment into it before the retirement party.