Big Box Retail & Malls
Product placement analytics is the practice of measuring what a shelf, an endcap or a floor display actually does to sales, using position data, traffic counts and transaction records instead of opinion. It replaces the planogram argument that begins with “we think” and ends with nobody checking.
Most retail spending still happens inside a building. E-commerce accounted for 17.1 percent of total US retail sales in the second quarter of 2026, seasonally adjusted, according to the US Census Bureau’s Quarterly Retail E-Commerce Sales report. The rest is decided by a person standing in front of a shelf.
This article covers what to measure, how to run a placement test that survives a skeptical category manager, which numbers are worth reporting, and the privacy rules that apply before you install a single sensor.
What is product placement analytics?
Product placement analytics is the measurement of how a product’s physical position in a store changes its sales, using three linked data sets: where shoppers go, what they do when they get there, and what they buy. Position is the variable. Sales is the outcome.
The older craft underneath it is still real. Retail product placement is the work of deciding where products appear through planning, design and negotiation, including spatial analysis of shelving and the planograms that merchandisers use to map capacity. What analytics adds is a before and after. A planogram is a hypothesis about shopper behavior, drawn to scale. Until someone counts traffic past the bay, dwell inside it, and units through the till, it stays a hypothesis.
That distinction matters because placement decisions are expensive and semi-permanent. Moving a bay costs labor, resets the category, and annoys regular shoppers who had learned where things were.
Why does shelf position move more sales than shelf space?
Because position decides who sees the product, while extra facings mostly guard against out-of-stocks. Field experiments published in the Journal of Retailing found location had a large sales effect, and facings much less once a minimum stock level was held.
That result comes from Drèze, Hoch and Purk’s 1994 study in the Journal of Retailing, which ran controlled shelf resets in real stores rather than modelling them. Two decades later, Eisend’s 2014 meta-analysis in the same journal pooled 1,268 published estimates of shelf space elasticity and reported an average of 0.17. Sales do respond to extra facings. They respond weakly, and far more slowly than the space costs.
| Placement lever | What the published evidence supports |
|---|---|
| Moving a product to a better location | Large effect on units sold |
| Adding facings above the out-of-stock floor | Small effect, weak elasticity |
| Dropping below the out-of-stock floor | Lost sales from empty shelf |
| Cross-category adjacency | Tested per store, not assumed |
The practical reading: spend the negotiation on where, not on how wide. A brand fighting for a third facing in a weak aisle is buying the wrong thing.
What data do you need to measure product placement?
Four data sets, and the measurement fails if any one is missing: spatial data on where shoppers actually walked, activity data on what they did in the store, observational data from field representatives, and sales data from the point of sale for the same hours.
The original three categories still hold. Observational data is the qualitative record your representatives compile in the field: display intact, stock present, competitor encroaching. Activity data is counts of visits and actions. Sales data is units and value over a defined period. The fourth, spatial data, is the one that turns the other three into a placement argument, because it is the only one that knows a shopper passed the display and did not stop.
| Data set | Question it answers | Typical capture |
|---|---|---|
| Spatial | Did shoppers reach this bay? | Indoor positioning, zone counts |
| Activity | How many visits and actions? | Store systems, app sessions |
| Observational | Was the display actually correct? | Field rep audits, photos |
| Sales | What sold, in which hour? | Point of sale records |
Join them on time and store, never on store alone. A weekly total hides the fact that your display wins on Saturday afternoon and does nothing on Tuesday morning, which is a staffing and restock finding as much as a placement one.
How do you run a placement test that proves a lift?
Treat it as an experiment with a control, not a rollout you measure afterwards. Pick matched store pairs, change position in one arm only, hold price and promotion constant, run across at least two full trading cycles, and measure the whole category rather than the single product.
The discipline is mostly in what you refuse to do:
- Hold everything else still. A placement test that coincides with a price change measures neither.
- Match the pairs on baseline traffic and sales, not on store size or region, and check the pre-period actually tracks.
- Measure the category, not the SKU. A product that gains 40 units while its neighbor loses 38 has moved sales, not created them.
- Keep the control arm honest. If the regional manager “helps” the test stores with extra facing work, the result is unusable.
- Write the decision rule before you look. Decide in advance what size of difference would make you reset the chain.
Cannibalization is the finding most placement reports quietly omit. It is also the one a category manager will ask about first, so bring it unprompted.
Which metrics tell you a display is working?
Four metrics, read together: capture rate, dwell, zone conversion and category transfer. Each one isolates a different failure. A display can be seen and ignored, or stopped at and not bought from, and those need opposite fixes.
| Metric | Definition | What a low value suggests |
|---|---|---|
| Capture rate | Share of store visitors entering the zone | Wrong location, poor sightline |
| Dwell | Median time spent in the zone | Nothing holding attention |
| Zone conversion | Units sold per zone visitor | Price, pack or message problem |
| Category transfer | Category units versus product units | Gain came from a neighbor |
A high capture rate with low conversion is a merchandising problem at the shelf. A low capture rate with high conversion is a placement problem, and it is the better one to have, because moving the bay is cheaper than rebuilding the offer. Report all four or the reader will pick the flattering one.
What are the limits of in-store placement data?
Real limits, and stating them is what makes the rest credible. Device-based counting measures devices, not people. Camera counting carries its own biases. And a controlled test tells you what happened in those stores, in that season, with that assortment.
Phone-based sensing undercounts in predictable ways. Shoppers who leave the phone in a pocket with Wi-Fi off, shoppers who carry no phone, and address randomization on modern handsets all cut the sample. Shared trips produce one device for two or three decision makers. None of that stops the method working, because placement analysis depends on relative change between two positions rather than on an absolute count of humans, but it does mean a reported footfall figure is an index and should be labelled as one.
Beacons are sometimes the right answer. For one aisle, one promotional display, or a single store where the question is narrow and the budget is small, Bluetooth Low Energy hardware at the bay is simpler than instrumenting the building. The trade-off arrives later, in battery replacement and in the fact that a beacon knows proximity rather than a path. Ultra-wideband sits at the other end, with tighter positioning and higher hardware cost per square meter. Choose by the question, not by the specification sheet.
What privacy rules apply to in-store product placement analytics?
Two separate obligations: whether your sensing produces personal data at all, and what you may then do with it. Both have been enforced, and the enforcement has been about the promises retailers made rather than the technology they installed.
Under the EU’s General Data Protection Regulation, personal data in Article 4(1) covers information relating to an identified or identifiable person, and Recital 26 puts genuinely anonymous data outside the regulation. The European Data Protection Board’s Guidelines 3/2019 on processing of personal data through video devices, adopted in 2020, draw the same line for cameras, separating simple counting algorithms from biometric techniques, and noting that the regulation does not apply where a device collects no information relating to a natural person.
In California, “precise geolocation” is treated as sensitive personal information, and the statute defines it as data used to locate a consumer inside a circle with a radius of 1,850 feet, per the California Consumer Privacy Act text published by the California Privacy Protection Agency. Any in-store position estimate is well inside that radius.
| Regime | What it turns on | Practical effect in store |
|---|---|---|
| GDPR Article 4(1), Recital 26 | Whether a person is identifiable | Aggregate at source, document it |
| EDPB Guidelines 3/2019 | Counting versus biometric processing | Keep counting algorithms non-identifying |
| CCPA sensitive data rules | Location inside 1,850 feet | In-store position qualifies |
The two US enforcement actions are worth knowing by name. In April 2015 the Federal Trade Commission settled with a retail tracking firm over misleading claims about in-store opt-out; the complaint alleged the company collected information on about nine million mobile devices in the first nine months of 2013, while the in-store opt-out it had promised did not exist. In January 2024 the FTC ordered a location data broker to stop selling sensitive location data and to delete what it had collected.
Neither case was about measuring display performance. Both were about onward use. Keep placement analytics aggregated, keep it inside the business that collected it, and the exposure stays small.
How do you present placement data to a retailer to win shelf space?
Lead with the retailer’s category, not your brand. Show the gap in their current layout, the test that measured it, the category-level effect, and the restock implication. A pitch that promises your product will sell more is ignored. A pitch that shows where the category loses units gets a meeting.
Four things carry weight in that conversation:
- A matched-pair test with a named control arm and a stated run length
- Category transfer numbers, offered before anyone asks for them
- Out-of-stock evidence from field audits, with dates
- A specific position request, with the sightline reasoning for it
The older tactics from this article still apply underneath the data. Co-brand the display where it helps the category. Identify the exact prime real estate you need rather than asking for more width everywhere. Fight for eye level in the bays where capture rate shows traffic actually arrives. Consider a broker when you have no field team. Keep the display tidy and cohesive, because a crowded fixture suppresses dwell and the measurement will show it.
Where should you start?
Start with one category in a handful of stores, not a chain-wide program. Four weeks is enough to produce a defensible number, and a small first test buys the credibility to run a larger one.
| Week | Action |
|---|---|
| 1 | Pick category, match store pairs, pull baseline |
| 2 | Instrument zones, verify counts against manual audit |
| 3 | Move position in test arm only, freeze promotions |
| 4 | Read all four metrics, check category transfer |
The manual audit in week two is the step people skip. Stand in the aisle for an hour, count by hand, and compare with the sensor output. Mapsted works on indoor positioning, indoor mapping and location analytics, and the same rule holds for any such system: a zone count nobody has ever validated against a human with a clicker is a number, not a measurement.
Frequently Asked Questions
What is the difference between product placement analytics and footfall counting?
Footfall counting tells you how many people entered the store or passed a door line. Product placement analytics ties position to outcome, so it needs zone-level spatial data joined to point-of-sale records for the same hours. A door counter cannot tell you whether a shopper walked past your endcap and ignored it, which is the finding that changes a layout.
Do I need shopper consent to measure display performance?
It depends on whether the processing produces personal data. Under GDPR Article 4(1) and Recital 26, genuinely anonymous aggregate counts fall outside the regulation, while anything that identifies a person does not. California treats location within an 1,850-foot radius as sensitive personal information. Get a data protection assessment before installation rather than after, and never promise an opt-out you have not built.
How long should a product placement test run?
Long enough to cover at least two complete trading cycles, so weekday and weekend patterns both appear in both arms. Shorter tests pick up weather, paydays and local events instead of placement. If a stakeholder wants a result in a week, run the week as a dry run of the instrumentation and keep the decision for the full period.
Why is wayfinding relevant to product placement?
Wayfinding determines which parts of the store get traffic at all, and placement analytics can only optimize within the paths shoppers actually take. If signage funnels everyone down two aisles, the bays on the other side have a capture-rate ceiling no merchandising will lift. Fixing the route is often cheaper than fighting for a better position inside a dead zone.
Can analytics replace planograms?
No. A planogram is still the instruction a merchandiser executes, and spatial analysis of shelving and capacity still produces it. Analytics changes how that diagram is chosen, and how it is checked once the shelf is built. The diagram stays. The guesswork behind it goes.
If you want to know which displays in your own store are being walked past, the first step is a conversation about your floor plan and your zones. Book a scoping call and bring your current planogram.
