Before You Buy the Sales Copilot, Audit What It Will Read

MSC is rebuilding its sales model around AI, and the trade press is cheering the org chart and the task lists. Nobody is asking the more useful question: what is the model actually reading when a rep asks it for help? A sales copilot is a retrieval system laid on top of your product data, and it can only be as good as the catalog underneath it. For most of the industry, that catalog is not good enough yet.

What MSC actually said, versus what got covered

Read Distribution Strategy Group's writeup of MSC's reset closely and the AI claims are thinner than the headline suggests. MSC consolidated overlapping sales coverage (accounts that had "2, 3, 4 or even 5" reps calling on them, down to one) and pointed to AI adoption in planning, procurement, and distribution-center operations. No specific rep-facing copilot is named. The growth numbers cited, a supplier forum generating roughly $500 million in near- and long-term opportunity and a 10-basis-point margin gain, are attributed to sales process discipline, not to a model.

That distinction matters, because the surrounding coverage fills in the gap with enthusiasm. Distribution Strategy Group's 2024 piece on inside-sales "superheroes" describes AI generating daily task lists — cross-sell prompts, churn flags, reorder reminders — that turn order-takers into "trusted advisors." It's a clean pitch. It's also entirely about the interface. Every one of those six functions is a query against product and account data that has to already exist in usable form. The article never asks whether it does.

That's the pattern across most of this coverage, ours included until we started measuring it: intense focus on what the rep sees, no interest in what the system is querying to produce it.

A copilot is a retrieval system, not a brain

Strip the marketing layer off any sales AI tool and what's underneath is retrieval-augmented generation: a model that looks up structured facts and composes an answer around them. Ask it "what's a comparable bearing if this SKU is backordered," and it isn't reasoning from first principles — it's cross-referencing attributes like bore diameter, load rating, and seal type across your catalog. If those attributes aren't captured as structured fields, there is nothing to cross-reference. The model either hallucinates a plausible-sounding substitute or, worse, states something false with total confidence.

This isn't a hypothetical failure mode. One industry analysis put the global cost of AI hallucinations at roughly $67 billion in 2024, headed toward $112 billion in 2025, and pointed to a case where hallucinated product specifications drove a 25% spike in returns for an electronics brand. Gartner has been blunter about the root cause: it expects 60% of AI projects to be abandoned through 2026 for lack of AI-ready data. The same gap shows up wherever AI meets a messy system of record. McKinsey's 2025 State of AI survey found that 88% of organizations now use AI in at least one business function, but only 6% report meaningful bottom-line impact, and CRM vendors trace a real chunk of that gap to incomplete records with nothing for the model to work from. A product catalog is a harder retrieval surface than a CRM pipeline. It has more fields, more variance across supplier feeds, and a buyer on the other end who notices immediately if the "equivalent" part doesn't fit.

What we measured, and why it matters here

This is exactly the gap our Digital Readiness Index was built to catch. We score four pillars and fourteen signals off each distributor's own live site — not self-reported claims, not a survey. One of those pillars is structured spec depth: how much of the catalog carries the kind of machine-readable attribute data a retrieval layer needs to actually answer a question, versus a PDF spec sheet or a product name and a price.

Across the Top Distributors 2026 field — 200-plus distributors, six operating archetypes, from national full-liners to regional specialists — the majority score short on that pillar. Not because these companies are careless. Because structured attribute data is expensive to build and has historically had a thin business case: a human buyer squints at a spec sheet and figures it out, a rep calls the vendor when they're unsure. AI removes that slack. It doesn't squint, and it doesn't call the vendor — it answers from whatever's in the field, or it doesn't answer at all.

The audit, before the contract

So the sequencing question an operator should be asking isn't "which copilot." It's: can I pull ten SKUs at random from my top-selling category and get complete, structured specs — not marketing copy, not a PDF — for every field a rep would need to answer "what's the alternative to this"? Do my supplier feeds populate attributes consistently, or does completeness depend on which vendor sent the data last? If the honest answer is no, a copilot built on top of that catalog is a very confident intern with nothing to read. It will sound right up until the moment a customer catches the substitution that doesn't actually fit — and by then the trust cost lands on the rep who repeated it, not on the software vendor who sold it.

I'm not arguing against AI in sales. I'm arguing about sequencing. The catalog is the retrieval layer whether or not anyone thought of it that way going in, and it's worth auditing before you buy the interface that sits on top of it. That's the layer we work in. Anglera sits on top of whatever PIM a distributor already runs and does the enrichment work to get catalog data to the structured depth an AI layer actually needs, without asking anyone to rip out what they have.