Detecting Misplaced Products and Planogram Deviation, Not Just Gaps

Gap detection is the headline, not the whole job

A shelf-monitoring model's most-cited job is spotting empty facings. But planogram compliance computer vision also flags products in the wrong place — which is a different problem, and often a bigger one: "we have it, but nobody can find it because it's on the wrong shelf."

What "misplaced" means in practice

A product detected in a slot where the planogram says something else belongs, or a product from one category sitting in another's section — a customer put it down, or a restocker filled the nearest gap with whatever was on the trolley. The stock is there; the sale is lost because the shopper can't find it.

Planogram deviation

Comparing the actual shelf to the reference plan: right products, right positions, right number of facings. It's coarser than gap detection, and for a good reason — a gap is an unambiguous visual signal, but "this product is 30cm left of where the plan says" needs the model to identify the product reliably, know the reference layout, and tolerate the fact that real shelves are never pixel-perfect. So planogram checks work at "this section doesn't match the plan," not millimetre placement.

Why it's worth the extra complexity

Misplacement fixes recover sales the gap-detection view misses entirely. And planogram compliance is often a contractual obligation — suppliers pay for specific shelf positions, and proving the product is where it's meant to be has direct value. On our computer vision project misplaced-item detection is part of what the model flags alongside stockouts.

What it needs that gap detection doesn't

A machine-readable planogram per fixture, and reliable product identification — which is harder than gap detection, and where promo packaging and occlusion hurt most. Roll it out in that order: gap detection first (high signal, low complexity), then misplacement and coarse planogram checks once product ID is accurate enough on your real store conditions.

The reference-planogram problem

Most retailers keep planograms as PDFs or slide decks built by category managers. The detection system needs them as structured data — fixture, position, product, facing count. Getting there usually means a one-time digitisation project or a planogram tool that exports structured layouts. This is frequently the long pole in a planogram-compliance rollout, longer than the model work, and it's worth scoping honestly before promising a delivery date.

Keeping them current is the other half. The reference data has to update whenever merchandising changes a layout, or the deviation alerts drift into noise. Tie the planogram export to the merchandising system's release cycle so the detection always compares against the plan that's actually live on the shelf.

What a good output looks like

Not "product X is 4cm off." The useful alert is: "Bay 12, store 47 — expected 6 facings of product X, detected 3; product Y detected in product X's slot." That's actionable at the shelf without the associate needing a tape measure, and it's the level of precision the model can actually sustain on a real shelf.

Where this stops being right

• No current machine-readable planograms — if merchandising keeps them in a slide deck, that's a data project before the detection can work.
• Unreliable product identification under real conditions caps how well planogram checks work, even if gap detection is excellent.
• The millimetre-precision version some vendors imply isn't realistic — set expectations at section-level match.

FAQ

Can the model do planogram compliance, not just stockouts? Yes, at a coarse level — "this section doesn't match the plan." It needs a machine-readable planogram and reliable product ID, which is harder than gap detection.

Why is misplacement worth detecting? It recovers "we have it but it's unfindable" sales, and planogram compliance is often a paid contractual obligation to suppliers.

How precise is it? Section-level, not millimetre. Real shelves are never pixel-perfect; the model handles "doesn't match," not exact placement.

Worth doing without supplier position contracts? Still yes for the findability win — misplaced stock is lost sales regardless. But the ROI case is clearer when supplier compliance payments depend on proving products are where they're meant to be.

ISTRALLEN builds shelf monitoring that adds misplacement and coarse planogram checks once product ID is solid; see AI for Retail.