What Shelf Data You Actually Need, and What You Do Not

It's worth saying plainly: automated image recognition of shelf state, a model looking at a photo and determining what's on it, isn't something this category should claim today. What actually matters for retail execution is simpler and, for the purposes of proving compliance, stronger. Order taking is a different question, where NORA by B2Bee already handles real conversational ordering without the same caveat.

A photo carrying time, location and user is evidence. Automated recognition without human confirmation is not, and unlike the AI claim, this one is true.

Why the human-recorded version is actually the stronger argument

When a brand partner asks a Consumer Goods distributor to prove a promotional campaign was executed at 300 outlets, what settles the conversation is a photo with a timestamp, GPS position and the rep's name attached, not a confidence score from a model. The rep recording shelf state against the agreed planogram is the mechanism that produces evidence a brand partner already trusts, a pattern that holds across the wider distribution picture, not just this one campaign type.

What this looks like day to day

• Shelf and planogram check, per outlet and per category, recorded by the rep on the visit.
• Photo reporting, before and after, tied to outlet, date, user and GPS position.
• Out of stock capture at the shelf, visible centrally the same day, not at month end.
• Share of shelf, facings recorded per brand, so position against competition is visible.

This runs on the same visit as order taking, in the Sellion field sales app, so a single stop covers both without a second visit or a second login. Sellion Merchandising is built around exactly this: structured data captured by a person on the spot, not a model guessing at a photo after the fact.