Digital Signage Analytics: What to Report, and What to Say You Measured
Four questions a media buyer asks, and the numbers that answer them. A play count supports a delivery claim and nothing further.
An assumed scenario, constructed to show the decision. It describes no named customer and reports no measured result.
A screen network loses a retail media deal in a meeting, not in a store. The screens performed. The report said "1.2 million plays," the brand's media buyer asked what a play is worth, and nobody in the room could answer without weakening the number.
The problem is not that the network measured the wrong thing. It is that the report treats four different things as one — that the campaign ran, that the screen was working, that somebody could have seen it, and that somebody did something afterwards — and it presented all four as one figure. This guide sorts them out, and shows what a report looks like when every number says exactly what it is.
Four questions a media buyer will ask
Any brand considering your screens is trying to answer the same four questions, in roughly this order.
Did I get what I booked? This is a delivery question. It is answerable from records you already keep: what played, on which screen, at what time, and which screens were down.
How many people could have seen it? This is an exposure question, and a play count is the answer that gets given when nobody has built one. A play count measures your equipment, not your audience. If you have not built a method for estimating exposure, the honest position is to say so. Where the screen is a lobby or a hospital corridor rather than a shop, the wayfinding planning guide covers what that audience is actually there for.
Did anyone respond? A response is a defined action — a QR scan, a tagged page visit, a coupon redemption. It measures one route to a response, not all of them, and it never tells you why the person acted.
Did it change anything commercially? This is the question brands actually care about, and the one that needs the most care, because a sales increase in a store where you are also running a promotion is not evidence about the screens.
Write these four questions on one page before you design anything. Every number in your eventual report should sit under one of them, and should say which one.
Give every number a job description
The fastest way to make a report credible is to write down, for each metric, what it is and what it is not. The table below is that page. Fill it in for your own network before the next campaign. It takes about an hour, and it prevents the meeting in the opening paragraph.
| The number | What it actually describes | What it cannot tell you |
|---|---|---|
| Plays logged | Your player ran the file | Whether the screen was on, working, or seen |
| Screens online | Endpoints that reported to the CMS | Whether anyone was standing in front of them |
| Estimated audience | Whatever method you declared, at whatever coverage you stated | Attention, or a unique person, without more evidence |
| Responses | One tracked route, such as a QR scan or a tagged link | The other routes, or the reason for the action |
| Sales in the period | What the store sold | Whether the screens caused any of it |
Read the last column again. It is not a list of weaknesses. It is the reason a media buyer will believe the first column.
One rule makes the table work: when something is not measured, write "not measured." Do not substitute something adjacent. A substitute invites a comparison you cannot defend, and it usually surfaces in the second meeting rather than the first.
Delivery is where you earn the right to talk about audience
Start every report with the part you can prove, because proving it buys you patience for the harder numbers.
Reconcile the campaign against the booking, and show the whole reconciliation even where it is unflattering:
- what was scheduled, and what played;
- which screens were offline, and for how long;
- which version of the content ran, and whether it was the approved one;
- whether the promoted product was actually in stock while the campaign ran;
- anything you excluded, and why.
Networks resist publishing this list, because on the page it looks like admitting failure. Publishing it tends to read the other way. A buyer who has seen your own outage figures has a reason to trust the exposure figure that follows, because they have already watched you report against your own interest.
If you cannot measure exposure, say so and sell what you can
Exposure is the part of retail media where networks most often overreach, and the industry's own measurement bodies have been working on comparability for years. A defensible exposure claim has three parts, and all three belong in the same breath.
The method. How the estimate is produced, in one paragraph a store manager could follow.
The coverage. What share of screens and hours the method actually covers. If it covers part of the estate and you extrapolate the rest, say the coverage out loud, next to the number rather than in a footnote.
The refresh. How often it updates, and what happens when the data source stops. If a feed fails and the number stays put, the number is held, not live.
A network without a defensible exposure method still has something to sell: verified delivery. Some brands buy reach. Many buy proof that a campaign ran where it was supposed to run. Selling delivery honestly is a smaller business than selling an audience number you cannot support, and it is a business that survives its first audit.
Design the test before you run it
Most screen networks run a pilot first and work out what it measured afterwards. That order produces a report nobody can act on.
Four decisions belong before the campaign starts.
Choose the primary response now. Pick the one measure that will decide the outcome, and write it down. If you choose whichever number improved after the fact, you have not run a test; you have collected anecdotes.
Choose the comparison deliberately. Two similar stores, or the same store alternating weeks. Each removes a different kind of noise, and neither removes the fact that stores differ in traffic, staffing and stock. Write down the differences you could not control.
Hold everything else still. Placement, offer, price and availability should stay the same across the comparison if the question is about the message.
Decide in advance what a small sample will mean. One screen in one store for two weeks produces a small number, and a small number is still useful if you describe it as one.
Two calculations, and the words that go with them
Delivery completion is qualifying plays divided by booked plays, with the campaign, the screens and the period named every time you use it.
Response per play is qualifying responses divided by logged plays. Call it response per play. If you want to call it conversion, you need an audience denominator measured separately from the player. Otherwise both halves of the fraction come out of the same box, and the rate does not mean what the name says.
For anything you compare over time, show the underlying counts alongside the percentage. A percentage on a small base moves a long way on very little, and a buyer who spots that will stop reading the rest of your report.
When you only have a little data, say that clearly
Small pilots are normal in this business, and they are not a failure. What fails is describing a small pilot in the language of a large one. A summary that survives review sounds like this. Every figure in it is invented for illustration:
Example only. These numbers describe no real campaign, and no measurement method behind them is published. Across four screens in two stores over three weeks, the campaign played as scheduled in 97% of booked slots, with one endpoint offline for two days. We measured 214 responses against that delivery.
Two things about that paragraph matter more than the figures. The first is that it names what the numbers are and what they are not. The second is that it claims no direction: with no control group and no comparison message, a higher response count is a description of what happened, not evidence that the message caused it. A summary that says "the direction favors our message" has stepped past what an uncontrolled sample can support, and it is the sentence most likely to be challenged in the meeting.
That paragraph will not win a budget by itself. It gets you into the meeting where the budget is decided, which is more than a percentage with no denominator will do.
Six things to check before you send the report
- Endpoints that did not report. Were they excluded, or quietly counted as delivered?
- Duplicate events. A player that retries after a network drop can log the same play twice unless you define what counts as one play.
- Time zones. A network across several zones will mis-state dayparts unless every event carries a stated zone.
- Content changes mid-campaign. An edit halfway through changes what the exposure figure describes.
- Stock. A play for a product that was out of stock is not the same as a play for one on the shelf.
- The response route you counted. A QR code measures the people who scanned it, not the people who walked in later.
One more note, on in-store sensing. If your exposure method uses cameras or other sensing, do not treat a camera-derived attribute as a verified identity, and do not assume a method is acceptable because the equipment can do it. Notice requirements, permissions and data-handling rules are a decision for the project owner and the responsible specialists, and they belong before the pilot rather than after it.
What this guide does not cover
It publishes no benchmark response rate, uplift figure or comparative performance claim, and it covers no advertising or privacy law. Whether your players and CMS can log what a report needs is a hardware and software question, and worth settling on paper before it is promised to anyone.
Send us the reporting requirement
Tell us which player and CMS you run, what your report needs to record, and whether you need to track content versions and endpoint uptime. We will tell you what the display and player configuration can and cannot log, against the documentation for that configuration, before you commit to a campaign schedule.
Tell us what the report has to answer →
Sources
- IAB: A Framework for Maturing In-Store Media Measurement
- IAB and IAB Europe: In-Store Retail Media Standards
Compiled September 29, 2026. The method is the author's; the external sources are listed above, and the opening scenario is constructed rather than reported.
