Planogram Compliance Robots

Compliance reporting has a credibility problem that has nothing to do with robots. Merchandising leads are asked to defend a percentage to a supplier, a category manager, or an auditor, and the percentage is routinely produced by a sampling process with an undocumented definition. When the number is challenged, the programme loses, regardless of whether the shelf is right.

A scanning robot does not automatically fix this. It replaces an inconsistent sample with a consistent census, which is a large improvement, but it also makes the definition problem visible: the report will faithfully apply whatever scoring rule it was configured with, at scale, in writing. Getting that rule right before the first pass matters more than the hardware.

Why Compliance Numbers Disagree Between Teams

Four definitional choices account for almost all of the variance between two compliance reports of the same shelf.

• Whether the denominator is facings or positions. A four-facings-wide block that is fully stocked but shifted one position left scores 100% on a facing-count basis and 0% on a position basis. Both are defensible. Reporting one while implying the other is not.
• Whether gaps count as non-compliance. An empty facing where the planogram calls for the product is out of stock, not a planogram error, but a naive matcher counts it as a mismatch. Every out-of-stock event then depresses the compliance score, and the merchandising team learns to distrust the number precisely when the store is trading badly.
• Whether the tolerance band is symmetric. Most planogram software accepts a one-position offset as compliant in one direction and not the other, because the planogram is anchored to a shelf end. If the robot's rule and the auditor's rule use different anchors, results diverge systematically at every aisle end.
• Whether the basis is a sample or a census. A 10% sample of facings gives a compliance estimate with a confidence interval; it does not give a store score. Presenting a sample estimate as a census figure is the single most common source of overstated precision in retail compliance reporting.

The fix is not subtle and does not require new technology: publish the definition, in writing, alongside every number. A compliance score without a stated definition, tolerance, and basis is an opinion.

The Three Metrics That Matter

Three metrics, reported together, cover the decisions the business actually makes. Reporting a fourth or fifth usually adds noise rather than control.

Metric	Definition	What it drives	Healthy band
Facing compliance	Count of correctly stocked facings divided by count of planned facings, excluding facings flagged out of stock	Merchandising execution quality; supplier compliance claims	92% to 97% at steady state
Position accuracy	Proportion of product blocks whose horizontal start falls within ±1 facing of the planogram anchor	Navigation and shopability; whether customers can find what they expect	88% to 95%
Share-of-shelf variance	Absolute difference between planned and actual share of linear shelf space, per brand block, in centimetres	Supplier disputes; category management credibility	Median under 8 cm per block

The exclusion in the facing compliance definition is the important one. Separating out-of-stock from merchandising error is what makes the metric actionable: an out-of-stock routes to replenishment, a facing error routes to the merchandising team, and a position error routes to whoever executed the last reset. A single blended number routes to nobody.

What a Robot Measures That a Human Auditor Cannot

It is worth being precise about the advantage, because it is not "better vision". A trained auditor with a planogram in hand reads a shelf better than any current production vision model. The robot's advantage is procedural, and it has three parts.

Repeatability. The same unit, applying the same rule, produces the same score on the same shelf state. Two auditors produce two scores, and the difference between them is invisible in the report. Where compliance is a contractual or supplier-facing number, repeatability is the whole product.

Cadence. A robot can run a route three to five times per week at zero marginal labour cost. A manual audit programme typically runs monthly or quarterly because auditor hours are expensive. Cadence changes what the metric is: a monthly audit measures whether the shelf was right on audit day, and everyone involved knows which day that is.

Census. A full pass covers every facing on the route, not a sample, which means the score is a population figure rather than an estimate, and the report can name the specific 37 facings that failed instead of estimating that there are probably 37.

The honest counterweight is that a human auditor catches things a camera does not: a product in the right facing with the wrong variant, a damaged pack, a shelf-edge ticket describing the wrong size. The best programmes run the robot weekly and a short human verification pass monthly, using the robot report to target where the human looks.

Scoring Arithmetic With a Worked Store

Work the numbers for a 24,000-facing store across 18 aisles, with a two-unit fleet running five passes per week.

Measure	Value	Derivation
Total planned facings	24,000	From planogram set
Facings flagged out of stock	1,180 (4.9%)	Gap detection, excluded from compliance denominator
Compliance denominator	22,820	24,000 − 1,180
Facings failing the match rule	1,461	Wrong product, wrong count, or unreadable association
Facing compliance	93.6%	(22,820 − 1,461) / 22,820
Blocks outside ±1 facing	612 of 4,100 blocks	Position rule
Position accuracy	85.1%	(4,100 − 612) / 4,100
Median share-of-shelf variance	11.4 cm	Per brand block, absolute

Three numbers, three different owners, three different remedies. The facing number at 93.6% is inside the healthy band and indicates a working process with normal drift. The position number at 85.1% is below band and points at reset execution quality, not at daily facing discipline. The share-of-shelf variance at 11.4 cm is above the median target and is the number a supplier will use in a negotiation, so it needs to be accurate and it needs a tolerance band agreed in advance.

Note what the out-of-stock exclusion does to the headline. Without it, the store reports 87.5% compliance and the merchandising team is blamed for a replenishment failure. That is a 6-point swing caused entirely by a definition, which is exactly why the definition has to be written down before the first report is circulated.

Manual Audit Cost Versus Fleet Cost

The comparison that decides most programmes is not accuracy, it is the fully loaded cost of the audit it replaces. Manual compliance audits are usually bought two ways: internal auditor hours, or an agency visit with a per-store fee.

• Internal audit. A trained merchandising auditor covers roughly 6,000 to 9,000 facings in an hour at a defensible accuracy, so a 24,000-facing store is 3 to 4 hours of pure audit work. Add travel, report writing, and management review, and a realistic weekly programme costs 6 to 8 hours per store at a $28 to $40 loaded rate. That is $9,000 to $16,600 per store per year, and it buys a monthly cadence at best.
• Agency audit. Per-visit fees in the $400 to $900 range are common, covering a defined scope and a report. At monthly cadence that is $4,800 to $10,800 per store per year, with the same single-visit limitation and less control over methodology.
• Robot fleet. A two-unit store carries $30,000 to $90,000 capex and $9,600 to $19,200 per year in service, software, and energy, delivering five passes per week rather than one per month.

At those numbers the robot is not cheaper than a monthly agency audit on a pure cost basis; it is roughly comparable on annual spend and delivers twenty times the cadence. The business case has to come from what the extra cadence is worth, not from labour substitution alone. If the answer is "nothing", the programme should not proceed, and an operator who cannot articulate a weekly use for the report will not act on a weekly report either.

From Compliance Score to Sales

Every proposal claims a sales lift from compliance. Very few programmes can prove one, and the reason is almost always the study design rather than the effect.

The credible design is a stepped-wedge rollout across matched stores: identify 20 to 40 comparable stores, split them into two matched groups on baseline sales, traffic, and format, then switch the programme on for group A while group B continues, and switch group B four to six weeks later. Compare within-store change against the matched control, not against the same store's previous period, because the previous period contains seasonality, promotional calendars, and price changes that will confound the result.

Two design details decide whether the answer is usable. First, hold the promotional calendar frozen for the comparison window, or the effect you measure will be the promotion. Second, measure the intermediate variable as well as sales: if compliance rose 8 points and sales did not move, the compliance definition is probably measuring something the customer does not notice, which is useful to know and rarely reported.

The honest version of the sales argument is that compliance protects revenue rather than creating it. A well-executed planogram makes the shelf navigable and keeps promotional stock where the promotion says it is. The value shows up as the absence of an incident, which is why it is hard to measure and why operators who have run without it for a quarter tend to fund it afterwards.

Reset Day: The One Day Compliance Actually Matters

Most stores run a reset cycle, and the 72 hours after a reset are when the planogram changes and the shelf either follows or does not. Everything else is drift. A compliance programme that does not specifically cover reset verification is measuring the wrong period.

The rule worth adopting: a reset is not complete until a full scan has been run within 72 hours of the reset finishing, with a signed exception list. The 72-hour window is a compromise. Too soon and the merchandising team is still working, so the report is noise; too late and the drift from the new planogram has already blended with pre-existing error, and the team cannot tell which failures were theirs.

This single rule changes the reporting calendar. Instead of a uniform weekly pass, the route becomes a baseline pass, a reset-triggered pass, and a cadence pass in between. Operators who adopt it usually find that the reset-triggered pass accounts for most of the compliance improvement in the first two quarters, because it converts an invisible execution failure into a specific, same-week correction.

The Reporting Contract

A weekly report that survives a challenge contains, at minimum, the following. If a system cannot produce these fields, the report will be argued with rather than acted on.

• The scoring definition, tolerance, and basis (census or sample), stated on the report itself rather than in a methodology appendix.
• Out-of-stock facings reported separately and excluded from the compliance denominator.
• Per-aisle breakdown, because a store score hides the fact that one aisle is carrying the whole failure.
• A named exception list by facing, with the planogram position, so the correction is a task rather than an investigation.
• The pass timestamp and the unit serial, so the report is attributable.
• A trend line of at least 12 passes, because a single number cannot distinguish a stable process from a recovering one.

Rolling It Out Without a Merchandising Mutiny

Compliance programmes fail politically far more often than technically. The report lands as an accusation, the store manager defends, and by month three the report is being generated and ignored.

Three practices prevent that. Give the store manager ownership of the score rather than visibility of it: the region reports the score, the store owns the exception list and closes it. Never publish a store league table in the first quarter, because ranking before the process is stable teaches stores to game the measurement rather than fix the shelf. And make the first two months explicitly a baseline period, with no targets, so the number is understood as a measurement rather than a verdict.

The exception workflow matters as much as the score. A weekly list of 1,461 failing facings is not actionable; a list sorted by aisle with the 40 highest-value corrections at the top is. Where the platform supports it, routing the exception directly to the person who owns that aisle, with a photo, converts a report into a work queue, and a work queue is the only form in which a compliance score changes a shelf.

If you want the compliance score specified against your own planogram set before committing to a fleet, send us an aisle sample and your current reporting format and we will return a scoring specification and route plan. The out-of-stock detection workflow covers the gap-detection half of the same pass, and the scanning hardware comparison covers sensors and mounts. If you are still deciding whether the audit layer is worth funding at all, the ROI framework sets out how to value a measurement capability, and the ESL comparison covers the display-layer alternative.

The Number Is the Product

Every vendor in this category sells hardware and software, and every operator buys a number. The hardware determines whether the number is available; the definition determines whether it means anything; the workflow determines whether it changes anything.

An operator who writes the definition down, separates out-of-stock from execution error, verifies resets within 72 hours, and routes exceptions to named owners will get more from a modest single-unit fleet than a larger operator who buys more units and never agrees what compliant means. The robot is a census instrument, and a census only has value when someone has agreed what is being counted.