Pilots Don't Fail at the Demo — They Fail at SKU 5,001

Your pilot isn't lying to you. It's just describing a distributor that doesn't exist — one where every SKU has a spec sheet, an image, and three years of clean transaction history. Scale-up doesn't fail because change management collapsed or the vendor oversold. It fails because SKU 5,001 has none of that, and nobody budgeted time to fix it.

The demo is real. The catalog isn't.

Walk into almost any pilot review and the data on screen is genuinely good — full attributes, current pricing, clean images, a sales history that goes back years. That's not a coincidence and it's not dishonesty. It's what "pick a representative sample" turns into in practice: whoever scopes the pilot reaches for the SKUs that are easiest to pull clean data on, because a pilot that fails on data quality reads as a failed pilot, and nobody wants to own that. So the A-items get the demo. The other 80-95% of the catalog — the long-tail SKUs with a part number, a vague description, and nothing else — waits for "phase two."

Phase two is where the trouble starts, and it isn't a training problem. Distribution Strategy Group has written about the "pilot paradox" — the gap between a successful small-scale test and a stalled enterprise rollout — and points to controlled pilot environments, legacy-system integration surprises, and under-invested change management as the drivers. Those forces are real. But in every enrichment rollout we've watched across distributor archetypes — branch-dense regionals, catalog-native specialists, scale aggregators — the actual failure event looks the same: someone loads category 40 of 60 and finds 4,000 SKUs with three populated fields and a scanned PDF nobody can find. The rollout doesn't stall because users resisted a new interface. It stalls because there's no data behind the interface to work with, and manually chasing it down at 30-45 minutes a SKU turns a two-week phase into a fourteen-month one.

The scope of the problem is not small. One mid-market electrical distributor found during an ERP data-quality audit that 22% of its item master hadn't seen a transaction in three years, and 18% of the catalog had no product category assigned at all — the kind of long tail that never shows up in a curated pilot set because nobody would choose it on purpose. Gartner's oft-cited estimate puts the ongoing cost of poor data quality at $12.9 million a year for the average organization — and distributors, sitting on more SKUs per revenue dollar than almost any other sector, absorb more than their share of it.

This isn't unique to distribution, either. MIT's widely cited 2025 study on enterprise AI found that 95% of generative-AI pilots delivered no measurable P&L return, and traced the failure to integration and workflow gaps rather than model quality — a structural finding, not an execution-team indictment. Distributors running enrichment, search, or AI-catalog pilots are living a specific, sharper version of the same pattern: the demo works because the data behind it was cherry-picked, and cherry-picked data is not a rollout plan.

Why "more change management" doesn't fix this

The Distribution Strategy Group prescription — invest in change management dollar-for-dollar with the technology spend, keep executive sponsorship past go-live — is sound advice for a different failure mode. It assumes the tool works and the organization needs to catch up to it. That's true for CRM adoption, where DSG has separately made the case that leadership has to stay engaged well past launch or usage decays. It's a people problem there because the data — customer records, contact history — is mostly usable from day one.

Product data pilots have a different shape. The tool doesn't need users to adopt it more enthusiastically; it needs input it was never given. No amount of training closes a gap where the attribute simply doesn't exist yet. You can run the best change-management program in the sector and still watch the rollout stall at SKU 5,001, because the problem isn't behavioral, it's structural: the pilot was scoped by convenience, and convenience excludes exactly the SKUs that make up most of the catalog.

Run the pilot on the tail, not around it

The fix is not a cleaner pilot. It's a differently scoped one. Pick one full category — not a sample, the whole thing, tail included — and run it end to end. If a category has 6,000 SKUs and 4,800 of them are long-tail, the pilot touches all 4,800. That surfaces the real data gap in week two, while there's still budget and attention to do something about it, instead of month fourteen, when it looks like the project failed.

The second half of the fix is budgeting for what week two turns up. Data remediation — sourcing spec sheets, standardizing attributes, filling the gaps a supplier feed never had — has to be a pilot-phase line item, not a rollout-phase surprise. Treat it as part of the pilot's cost of goods, not overhead discovered after the purchase order is signed.

At Anglera, this is close to the whole thesis behind our Digital Readiness Index: we measure the long tail because that's where digital readiness is actually decided, not the top 5% every distributor already has in order. Your PIM stores whatever data you feed it; it can't invent the spec sheet for SKU 5,001. That's the work — full-category remediation, tail included, priced in from day one — and it's the difference between a pilot that predicts your rollout and one that just flatters your cleanest data.