The Resolution Layer · Case Study in BriefEKOM
A national industrial-electrical distributor.
Two real data problems, two different right answers — and a catalog that became the doorway to something bigger. One of a set of anonymized EKOM analyses.
The discipline · right tool per problem
Two problems. Two different right answers. And the honesty to prove which is which.
EKOM ran two of this distributor's real data problems. On one, an AI read was the right tool. On the other, a plain deterministic method beat AI outright — we tested whether a language model could do the hard slice better, it scored worse, so we didn't use it and didn't bill for it. In both, the lever on quality was connecting their own authoritative source data, not piling on inference. That's what a team that told us their product data was "suspect" actually wants: judgment, not hype.
What that looked like
AI where it reasons
Faceted fuse attributes normalized and reconciled against the distributor's own association and manufacturer feeds — class values unified, value/unit split, titles rebuilt. ~96% resolved from data already in their ecosystem.
Deterministic where codes decide
Tens of thousands of un-segmented products classified into the reporting taxonomy straight from their own ERP codes — no AI. Held-out accuracy: Segment ~94%, Group ~86%, most rows shipping automatically.
The restraint, proven
We measured whether a language model could do the hard slice better. On the fine-grained call it couldn't — so we didn't run it. Right tool per problem, and honest about which is which.
The lever is source data
Where a source held the answer, we nailed it; where nothing did, we said so rather than guessing. The unlock isn't more inference — it's connecting the authoritative feeds and adding the fields that are missing.
Where it goes next · beyond product data
The catalog is the entry point, not the ceiling.
After the first pass, this distributor asked EKOM to take the same discipline past product data entirely — linking it to customer records and sales history to surface which accounts are buying below their potential, and generating those opportunities in the reps' ERP. Product intelligence becomes commercial intelligence, held to the same evidence.