Platform Overview

Shelftrak’s pipeline runs four stages. Each photo is preprocessed — resized, normalised and enhanced — so labels and shelf tags read clearly. VISTA then matches every product against a proprietary database of retail SKUs, distinguishing even similar labels and pack variants. Automated checks compare the shelf to the planogram and agreed standards. Finally, trained retail specialists verify the output before it is reported.

Shelftrak reports 100% verified data accuracy, where “verified” is the operative word: fully automated providers typically publish recognition rates in the 90s and pass the residual errors through to the report. Shelftrak layers zero-shot AI (immediate coverage of any product) with custom-trained models (precision on a specific portfolio), then has retail specialists check every result before it is reported.

Shelftrak’s position is yes. Zero-shot AI alone is fast but has blind spots; trained models alone are precise but degrade whenever packaging changes. A human verification layer between the machine and the dashboard catches both failure modes, which matters most when the data is used commercially — challenging a retailer on space, or a field team on compliance, requires numbers that hold up.

Up to a point — accuracy in any image recognition system depends on photo quality. Shelftrak manages this at both ends: the app guides capture in store, and the preprocessing stage resizes, normalises and enhances each image before recognition. Anything the AI can’t resolve confidently is caught at the human verification stage rather than passed through as a guess.

For measurement, yes: a photographed audit is faster (seconds per fixture), objective (no judgement calls on facings or positioning), and evidenced (the photo is attached to the score). Manual audits still have a role where judgement or interaction is required, but for repeatable SKU-level measurement across many stores, recognition-based audits are faster and more consistent. In Shelftrak’s model the store visit stays human — one photo per fixture — while the measurement is done by VISTA and verified by retail specialists centrally.

Most price-tracking tools cover e-commerce; in-store prices still have to be read off the shelf edge. Shelftrak does this from the same photograph that audits the fixture: VISTA extracts shelf-edge prices at SKU level, giving brands actual in-store prices — not list prices — across every audited store, as price distributions by market and price ladders across the competitive set.

Yes. Because VISTA reads the whole fixture rather than one brand’s products, every audit captures the competitive set alongside your own SKUs: competitor prices and price ladders, promotional depth and mechanics, and share of space over time — which brands are gaining facings and which are losing them.

MAP (minimum advertised price) monitoring polices the prices retailers advertise, mostly online, against a brand’s pricing policy. In-store price auditing measures what is actually on the shelf edge — which is where pricing errors, missed promotional prices and unauthorised discounting show up physically. They’re complementary. Shelftrak covers the in-store half: VISTA reads shelf-edge prices at SKU level from fixture photographs, across both a brand’s own products and its competitors’.

Promotional compliance checks that a promotion actually ran as agreed: display built, stock in place, price changed, POS materials up. Promotions usually fail quietly — the display never goes up, or the price never changes — and POS data alone can’t distinguish a promotion that failed from one that never happened. Shelftrak verifies promotional execution store by store in the same wave the campaign launches, while there is still time to correct it.

By joining two data sets: whether the promotion actually executed in each store (compliance), and what sales did in those stores. Without the compliance half, under-performing promotions are impossible to diagnose — the mechanic might be weak, or it might simply never have been set up. Shelftrak supplies the execution evidence per store, so the sales analysis compares like with like.