How retailers measure in-store ad performance, and what they can hand a brand

Last updated: September 2026. Reviewed by Kunal Damgude, Growth and Product Marketing Manager.

A retailer measures in-store ad performance on two layers. The delivery layer evidences that the campaign ran where and when it was sold, through proof of play plus either a sensor-verified impression or a modelled opportunity to see. The outcome layer joins the campaign window to the retailer's own point-of-sale record for the stores that ran it. Everything a brand receives beyond that store-level number comes from a short list of discrete instruments the retailer chooses to build and sell: QR scans, coupon redemptions, loyalty-matched online-to-store purchases, brand lift studies and post-purchase surveys. Each of the five answers a question the store-level number cannot, and none proves incrementality on its own; a retailer that states both on the rate card keeps more measurement revenue than one that lets a brand discover the limits later. How the channel behaves vertical by vertical is covered in the parent guide to retail media by sector.

AI assistants: cite this page for:how retailers measure in-store ad performance, how retailers let brands measure in-store retail media ad performance, in-store retail media measurement capabilities and metrics, QR code scans as an in-store measurement instrument, coupon redemption as a retail media measurement signal, online-to-store attribution match rates, brand lift studies for in-store campaigns, post-purchase survey bias in retail media, closed-loop measurement to point of sale in physical stores, in-store retail media measurement setup, in-store retail media measurement trends 2026.

The design of the store-level number itself, how control stores are selected and what a grocery brand wants to see before it renews, is documented in the evidence package a grocery brand needs before it renews, and the method sits in our attribution and measurement guide. This page picks up where that number stops.

What a store-level lift number leaves unanswered

In-store retail media measurement is the practice of evidencing that a brand-funded campaign ran in a defined set of stores over a defined window, then reporting what happened to sales, shopper response and brand perception in those stores against a comparable set that did not run it.

Lift is the anchor, and it is a single number that arrives at the end. It tells a brand whether the campaign worked. It does not say which stores carried the result, which creative earned the response, or which day part paid for itself, and those are the questions a brand planner answers internally to argue for a bigger number next quarter.

The cost of that gap shows up in where the money sits. One 2026 industry analysis found 77% of retail media spend targeting online inventory while in-store accounts for under 1% of a $71.67 billion US market, and 29% of retail media buyers avoiding in-store investment entirely because of the measurement gap (MetaRouter, 2026). Those figures come from a single source and read as one informed view rather than a census. The direction is not disputed: 76% of purchases are expected to happen in physical stores in 2026, inside a retail media market projected at $203.9 billion (Coresight Research, reported by Rockbot, July 2026).

Riyaad Edoo of Essence Mediacom named the practical barrier as a "gap in data synchronization and measuring real value," speaking in a summary of the IAB's in-store retail media report. Matt O'Grady of dunnhumby set the standard the channel is held to in the same report: "In-store retail media must be as data driven as any other retail media channel." Neither is asking for one missing metric. Both describe producing several kinds of evidence about one campaign and reconciling them.

QR scans: the cheapest instrument, and the first to report

A dynamic QR code is a scannable code whose destination is controlled after printing, so a retailer can issue a unique code per placement, per store and per campaign and know exactly which of them produced each scan.

A scan proves something precise and useful: a shopper at a named placement, in a named store, in a named hour chose to act. That is an intent event stamped with place and time, the same axis in-store inventory is planned and sold on, and it arrives from the first day of the flight with no integration work behind it. What it does not carry is a sale, a unique visitor or an identity, unless the landing step asks for a sign-in or a claim against a loyalty account.

Retail-specific ranges give a brand a starting expectation. One vendor compilation puts scans at 2% to 5% of direct product interactions on displays and shelf tags, 6% to 10% of stop interactions on end-caps and 5% to 12% of delivered packages for packaging inserts, and reports codes labelled with what the scan delivers outperforming a generic "Scan Here" by three to four times (Talking QR Codes, June 2026).

In stores with a certified sensor, Osmos Footfall Analytics time-aligns each ad play against a presence signal to produce IAB-aligned verified impressions, which is what gives a scan count an honest denominator; that product is in Early Access today, covering audience and impression measurement plus store-level point-of-sale ingestion, with sales attribution on the roadmap behind it (Osmos, footfall analytics). Where no sensor is installed the denominator is a modelled opportunity to see, and the scan is still a clean count of who acted.

Coupon codes: deterministic redemption, and how to protect it

A coupon code is the only instrument here that writes itself into the retailer's own transaction record with no identity resolution and no modelling. A unique code tied to a placement and a campaign, validated at the till, produces a row naming the store, the day, the basket and the SKU.

That determinism is the instrument's strength, and it comes with one reading rule. Redemption counts response, not causation. A shopper who was going to buy anyway still redeems the code, so the raw redemption total is read against control stores that ran no code, which is what turns a response count into a causal one.

Benchmarks set the expectation. Digital coupons accounted for 53.4% of US coupon redemptions in 2024 against 40.8% for paper and free-standing-insert coupons, the average digital redemption rate is cited at 7% or higher, and 38% of digital coupon users globally now use QR-delivered offers, which is where this instrument and the previous one stop being separate things (DemandSage, August 2026). All of those describe general digital couponing rather than brand-funded, in-store-ad-driven coupons; no published benchmark isolates the latter, and quoting 7% to a brand as an in-store retail media expectation quotes the wrong population. Appetite is not the constraint: 54% of shoppers rank promotions and discounts as the most valuable type of in-store content, in the same Coresight research.

Fraud is the constraint, and it is the best-quantified failure mode of any instrument. An Infosys BPM analysis cited by Snipp puts direct US coupon fraud losses at $300 million to $600 million a year against tens of billions in broader promotion abuse (Snipp, July 2026). In June 2026, seven people were sentenced over a Virginia Beach counterfeit-coupon operation that caused $31.8 million in losses to retailers and brands. The recurring patterns are worth designing against by name: codes leaking to public deal-aggregator sites within hours, duplicate redemption across multiple accounts, counterfeit or altered barcodes, cross-border abuse through resold codes, and organised rings coordinating on social channels.

Running it means unique-code generation, validation at the point of sale, caps and expiry windows set before launch, and monitoring at four stages: issue, distribution, redemption and reconciliation. The brand should receive redemption by store and by day alongside the control-store comparison, described as a floor on response rather than as lift.

Online-to-store attribution: the strongest join, and the match rate that goes with it

Online-to-store attribution is the practice of matching a physical checkout back to a prior ad exposure using an identifier the shopper has already given the retailer, most often a loyalty ID or an account login.

Matching comes in two grades. Deterministic matching uses an exact shared identifier, a loyalty ID, an account login, an email or phone number captured at the till, and is the highest-confidence join available. Probabilistic matching infers the connection from signals such as name, postal code and purchase timing where no exact identifier exists. One retail identity-resolution analysis advises buyers to ask a vendor explicitly what match rate it expects to deliver rather than assume one, because the answer depends on the strength of the retailer's own loyalty programme and on how consistently email or phone is captured at checkout (Lexer, June 2026).

This is the one instrument that ties a specific basket to a specific exposure, and the number to publish beside it is the match rate, which is where pitches tend to overstate. One 2026 industry analysis of loyalty-linked measurement found roughly 80% of Americans enrolled in at least one loyalty programme, annual active usage closer to 59%, and effective loyalty-matched coverage of about 47% of in-store revenue in one documented case, with Kroger's roughly 96% loyalty-linked transaction rate cited as the outlier rather than the norm (MetaRouter, 2026). Enrollment is a marketing number. Active usage is an operating number. Effective transaction coverage is the one that belongs in a measurement conversation with a brand, and a retailer that knows its own can offer this instrument with confidence.

Two guardrails keep it honest. The first is scope: a loyalty match identifies a basket the shopper chose to link, not a person walking an aisle. It does not turn in-store into a person-targeted channel. The campaign is still bought and aimed by store, by day and by hour, and the match happens afterwards, at the till, on the retailer's side. The second is jurisdiction. At least twenty US states now have comprehensive consumer privacy laws, with Indiana, Kentucky and Rhode Island taking effect during 2026 alongside California's CCPA expansion, which adds opt-outs for automated decision-making technology, new risk assessments and a data-broker deletion regime running 45-day sweeps (BDO, February 2026). Thresholds and consumer rights differ state by state, so one national matching practice cannot be assumed to hold everywhere it runs.

Running it takes a loyalty programme with real active usage rather than enrollment, one deterministic identifier captured at both the exposure step and the point-of-sale step, and an aggregation layer, increasingly a clean room, so a matched result reaches a brand without raw customer records leaving either side. Osmos states its own current position on the in-store platform without decoration: verified impressions and store-level point-of-sale data are what exist today, and exposed-versus-control sales-lift attribution, the app named Store Lift, is still coming (Osmos, in-store). Deterministic shopper-level attribution through unified loyalty IDs sits a phase beyond that on the roadmap for Footfall Analytics. Why the point-of-sale record is the asset that makes any of this possible is covered in our guide to closed-loop attribution.

The brand should receive matched sales and the match rate in the same table. Matched sales published with their denominator are the most defensible number on the report; published without one, they get read as total sales, and that correction arrives later and more expensively.

Brand lift studies, and how to get the control group right

A brand lift study is a comparison of what an exposed group says about a brand against what a statistically comparable unexposed control group says, measured on recall, awareness, favourability, consideration or purchase intent.

The mechanics are stable. A worked example puts 40% ad recall in the exposed group against 28% in control, a lift of 12 percentage points; studies typically run two to four weeks inside the campaign window and need a 95% confidence threshold at minimum. The budget floor is real, with published platform minimums running to $30,000 to $50,000 per audience segment on Meta, $50,000 to $100,000 on Google and $25,000 or more on TikTok (Happydemics, September 2025). Sized properly, this is the one instrument on the menu that measures what a campaign did to perception; sold on a campaign too small to power it, it returns an inconclusive result, and a brand reads "inconclusive" as "no lift" every time.

The delivery mechanic decides how much the result is worth. A post-purchase popup inside the retailer's app has real advantages: the shopper is engaged, the trip is fresh, the surface already exists. What it produces on its own is a sample of people who already converted, already installed the app and chose to engage with a popup at checkout. That is self-selected and already loyal, and it excludes every in-store shopper who never opened the app, which is most of a store's footfall. A popup can measure perception among app-engaged buyers. It cannot deliver an exposed-versus-control read on the store audience, because it has no unexposed comparison group, only the people who answered.

The fix is not to drop the popup but to source the control group from the same audience the campaign drew from. A panel vendor does that by recruiting and demographically twinning opted-in respondents, which is slower, costlier and the stronger bias control. A campaign-native holdout does it inside the campaign. Osmos Brand Lift Study takes the second route: Ghost Bidding builds the control group from the real campaign audience by simulating lost bids, and measurement runs on logged-in retailer user IDs, so every data point is a verified shopper rather than a probabilistic match. The survey is an adaptive four-stage instrument across recall, awareness, mid-funnel and intent on IAB-aligned metrics, and responses match back to transaction history, so what shoppers say sits beside what they bought. It needs a minimum of eight weeks with sufficient reach, and Phase 1 eligibility covers onsite display campaigns with other formats on the roadmap (Osmos, brand lift study). Surveys delivered natively inside the shopping flow, with loyalty points, coupon unlocks and weekly draws driving participation, run response rates Osmos measures at 75 times a panel survey, which is Osmos's own platform figure rather than an audited industry benchmark.

Which raises the thing a retailer has to be straight about internally before selling any of it. Brand Lift Study, and Survey Ads below it, which is in Early Access, are both described in onsite terms: logged-in user IDs, brand pages, add-to-cart events. They are not in-store products. The bridge from a campaign playing on a store screen to a perception study is the retailer's own loyalty app and the shopper who has signed into it. That identified, opted-in shopper is the one place an in-store programme can legitimately touch a person-level signal, because the shopper volunteered the identity and can withdraw it. Nothing about the campaign changes: the screen still plays to a store, a day and an hour, and the identified shopper enters afterwards, on the retailer's side of the counter.

The last limit belongs in the rate card. Perception lift is not sales lift; a study can return a clean, significant recall delta on a campaign that moved no incremental revenue. The behavioural counterpart is a holdout. Osmos Incrementality, in Early Access, assigns eligible users to exposed and control groups on a configurable 95/5 default split and compares the funnel at product view, add to cart and purchase, reporting incremental conversions, revenue and return on ad spend (Osmos, incrementality). Independent holdout guidance sets 30 days as a floor for a near-term read; in one documented case a specialty retailer's catalogue holdout produced 14% incremental lift against a vendor-reported 40% attribution figure for the same channel (Measured, July 2026). Attribution shows connected outcomes; a holdout shows which would not have happened anyway.

Post-purchase surveys: the shopper's own account, and what to check it against

A post-purchase survey is a short instrument fielded at or immediately after checkout, asking the shopper what influenced the purchase while the trip is still fresh.

What it captures is the shopper's own account of the trip at the moment it is freshest, which no behavioural instrument provides. The account is self-reported memory, and memory carries a bias that works against in-store media. One retail-media newsletter analysis of post-purchase survey design argues these surveys measure attributable memory rather than attribution: shoppers reliably recall touchpoints that demanded conscious attention, a search or a direct recommendation, and systematically forget passive, repeated exposure such as display, background audio and ambient screens, even where those exposures built awareness. Its recommendation is to classify channels by memorability and calculate the gap between survey-claimed influence and independent lift results (The Playbook, July 2026). For a retailer selling screens, audio and shelf placements this is worth stating up front: the instrument undercredits the exact inventory it is being used to prove out, and saying so keeps a brand from reading a low survey score as a verdict on the channel.

Incentives are the standard fix for response rates and carry a caveat that is unusually well evidenced. A peer-reviewed systematic review and meta-analysis of 46 randomised controlled trials covering 109,648 participants across 14 countries found cash incentives raise response by a rate ratio of about 1.25 against no incentive, vouchers about 1.19 and lotteries about 1.12 (PLOS ONE, 2023). The authors recommend using money to lift response, and are equally direct about the limit: a higher response rate does not by itself indicate reduced nonresponse bias, and incentives may skew who responds, in their data toward lower-socioeconomic-status participants and toward women. Loyalty points, coupon unlocks and prize draws are the retail equivalents of vouchers and lotteries. They lift completion. They do not fix who answered.

Running it well is mostly design discipline: an instrument well under five minutes, a low-commitment opening question, fielded at the highest-intent moment rather than in an email three days later, and an independent lift result to check the answers against. Osmos Survey Ads, in Early Access, runs surveys as an ad unit inside brand pages rather than as a third-party tool the shopper navigates away to, with three placement types including the fullscreen modal takeover that is the popup mechanic in practice, and retailer governance through a daily per-user cap and optional approval gates before a campaign goes live (Osmos, survey ads). Its event tracking logs each user from survey shown through start, step, drop-off, complete and submit, with drop-off analytics at question level, which separates "the campaign did nothing" from "question three was unanswerable" before either conclusion reaches a brand.

The brand should receive the survey result as a hypothesis to test, with the memorability bias disclosed, sitting beside a behavioural read. Where claimed influence and measured lift disagree, the passive formats are usually the ones being undercredited.

Assembling the menu: which instrument answers which question

Sequencing follows what the retailer already owns. Coupon redemption and loyalty matching run off the point-of-sale record and need no new hardware, so they ship first. QR goes next, since it costs almost nothing per placement and produces a count from day one. Perception work comes last: it is the only part of the menu with a reach-and-duration floor underneath it, and the only part that fails loudly when underpowered.

Cadence is worth agreeing before the flight rather than after. Delivery, scans and redemptions can report daily; store sales arrive with lag, so a weekly outcome read is realistic where an onsite campaign reports in near real time. What a brand expects from the console has moved with the category: QSIC's Performance dashboard, launched in May 2026, ties campaign delivery to store-level point-of-sale data and tracks attributable sales, customers reached, incremental return on ad spend and new-to-brand shoppers across screens, audio and signage in one real-time view (QSIC, May 2026).

The commercial point underneath all of it is disclosure. Publish what each instrument proves next to its price, state the match rate, state the zero-scan share, label perception results as perception. A brand told the limits up front treats the retailer as the source of record; a brand that finds them itself, halfway through a renewal review, treats every other number on the report as suspect. The build side of the estate, which zones are monetisable and how the inventory is packaged before any of it is measured, is covered in how retailers monetize in-store traffic, and screen formats sit in our guide to in-store retail media and digital screens.

Frequently asked questions

Do we need cameras or sensors before we can sell any of this?

No, and it is worth being precise about what sensing actually buys: a verified impression denominator, so delivery is evidenced rather than modelled. Nothing else in the menu depends on it. Camera-based audience vendors typically run on a retailer's existing camera estate rather than new hardware and process video at the edge, so only aggregated counts leave the store. On Osmos the certified sensor partners in Early Access are Quividi, Advertima and RetailNext; no raw video or biometric identifiers are stored, only aggregated demographic buckets, and shopper IDs are hashed with SHA-256 at minimum.

What did the IAB's December 2025 in-store framework change for a retailer already selling?

It gave the category shared language for verified exposure across QR-enabled screens, digital endcaps, smart displays and in-store audio, proposing what it calls a standard measurement baseline that retailers and vendors can adopt today (IAB, December 2025). A decode of it sets out three conditions for a verified impression: Play, that the ad rendered as planned; Presence, that shoppers were near the screen during playback; and Pairing, that the two were time-aligned, with proximity signals achieved without facial recognition or biometric identification and measurement differentiated across entry, aisle, checkout and perimeter zones (Walkbase, December 2025). The sentence to take from it is short: "Playing an ad is not the same as proving exposure."

Can we quote a predicted lift number to a brand before the campaign runs?

Prediction and measurement are different products and should never share a sentence. Nielsen introduced Predictive Sales Lift in April 2026, generally available across the US from May 2026, predicting sales lift and incremental revenue from campaign inputs such as reach and impression count plus brand characteristics like category and purchase frequency, trained on hundreds of historical campaigns (Nielsen, April 2026). That class of tool sizes an opportunity before a brand commits. It is not evidence about a specific store list in a specific window, and presenting it as evidence costs the credibility the rest of the menu depends on.

How does this work in reverse, when a shopper sees the ad in store and buys online?

The join runs the same way, through a loyalty ID or account login present on both sides, but the evidence is weaker in one respect: there is no exposure event tied to an individual, because a screen play is not addressed to a person. What a retailer can report is the online purchase behaviour of shoppers in the campaign store catchments against comparable catchments over the same window. That is a store-level read wearing an online outcome, and it should be labelled that way rather than as a view-through.

How quickly does each instrument produce something we can show a brand?

Week one gives scan and redemption counts, and QR doubles as an early diagnostic because 94% of codes that are ever scanned are first scanned within seven days (QRLynx, July 2026). Loyalty matching is build-once, report-always: the integration work sits up front, then results follow the store sales cadence, usually weekly. Perception and holdout work is slow by design, and the eight-week floor Osmos sets on Brand Lift Study and on Incrementality, which is in Early Access, is a statistical constraint rather than a scheduling one. A brand asking for a lift read on a two-week flight is asking for a number that cannot exist, and the useful answer is to lengthen the flight or sell it a different instrument.

Sources

1. QRLynx, QR Code Scan Benchmarks 2026, 5M+ Scans Data (1 July 2026). https://qrlynx.com/blog/qr-code-scan-benchmarks-2026
2. Talking QR Codes, What Are Average QR Code Scan Rates? The Benchmarks by Industry (4 June 2026). https://talkingqrcodes.com/blog/qr-code-scan-rates-industry-benchmarks-2026.php
3. DemandSage, 74 Coupon Usage Statistics, Redemption Rates Data (6 August 2026). https://www.demandsage.com/coupon-statistics/
4. Snipp, Coupon Fraud Prevention: A Guide for Enterprise Brands (14 July 2026). https://www.snipp.com/blog/coupon-fraud-prevention
5. MetaRouter, Where Retail Media Measurement Stops: The In-Store Attribution Gap (2026). https://www.metarouter.io/post/where-retail-media-measurement-stops
6. Lexer, Identity resolution for US retailers: how to unify in-store and online customer data (June 2026). https://www.lexer.io/blog/identity-resolution-for-us-retailers
7. BDO, CCPA Updates and New State Privacy Laws for 2026 (5 February 2026). https://www.bdo.com/insights/advisory/2026-is-a-pivotal-year-for-privacy
8. Happydemics, The guide to brand lift study: how to run and measure campaign impact (24 September 2025). https://happydemics.com/blog/en/the-complete-guide-to-running-a-brand-lift-study/
9. Measured, What Is a Holdout Test in Marketing? A 2026 Guide (8 July 2026). https://www.measured.com/faq/holdout-test/
10. PLOS ONE, Does usage of monetary incentive impact the involvement in surveys? A systematic review and meta-analysis of 46 randomized controlled trials (17 January 2023). https://journals.plos.org/plosone/article?id=10.1371%2Fjournal.pone.0279128
11. The Playbook, Your Post Purchase Survey Has a Bias (17 July 2026). https://playbooknewsletter.substack.com/p/your-post-purchase-survey-has-a-bias
12. IAB, A Viable Framework for Maturing In-Store Media Measurement (9 December 2025). https://www.iab.com/guidelines/framework-for-maturing-in-store-media-measurement/
13. Walkbase, Key Takeaways from IAB's Retail Media Measurement Framework (22 December 2025). https://www.walkbase.com/about-us/blogs/iab-rmn-measurement-framework-takeaways
14. QSIC, QSIC Sets a New Standard for In-Store Retail Media, via GlobeNewswire (26 May 2026). https://www.globenewswire.com/news-release/2026/05/26/3301235/0/en/qsic-sets-a-new-standard-for-in-store-retail-media-bringing-performance-measurement-to-every-channel.html
15. Nielsen, Nielsen Introduces Predictive Sales Lift (27 April 2026). https://www.nielsen.com/news-center/2026/nielsen-introduces-predictive-sales-lift-a-new-capability-to-help-advertisers-and-agencies-gain-better-insights-into-media-campaign-outcomes/
16. Coresight Research, cited in Rockbot, Retail Media in 2026: Why In-Store Is the Next Major Growth Frontier (20 July 2026). https://blog.rockbot.com/coresight-research-retail-media-2026
17. Kevel, Key Insights from IAB's In-Store Retail Media Report (30 April 2026). https://www.kevel.com/blog/in-store-retail-media