Measuring ROI on an AR campaign
This is the question that decides whether a second AR campaign gets funded, and it is usually answered badly, with a scan count and an uplift statistic nobody can source. Here is what to instrument instead, and what the numbers can and cannot honestly tell you.
Why is AR ROI so often reported badly?
Because the easy number is the useless one. Total scans is trivial to capture and rises with footfall, media spend and stand position, none of which describe the experience. A campaign can report a large scan figure while almost nobody completed the experience, and the report will still read as a success.
The second reason is that AR usually sits mid-funnel, between awareness and purchase, where attribution is genuinely hard. A visitor tries a sofa in their living room on Tuesday and buys it in-store on Saturday. Nothing connects those two events unless you decided in advance that they should be connected.
So the honest approach is to stop trying to produce a single ROI figure and instead measure the funnel well enough to know which step to fix. That is less satisfying in a deck and far more useful in the next planning round.
What should you instrument?
Six events, in order, so you can see where people leave.
| Event | What it tells you |
|---|---|
| Scan or link open | Whether the invitation was visible and appealing. A low figure is a placement or creative problem, not an AR problem. |
| Camera permission granted | How much trust the entry point carried. Denials cluster where the prompt arrives without explanation. |
| Tracking lock achieved | The most diagnostic metric available, and the one almost nobody captures. A poor rate is environmental and fixable. |
| Time in experience | Whether the thing was interesting once it worked. Very short sessions after a successful lock point at the content, not the build. |
| Completion or key interaction | Whether people reached the moment the campaign was built for, such as configuring a product or capturing an image. |
| Onward action | Share, save, add to basket, click to product, book an appointment. The only step that connects to commercial value. |
Which single metric matters most?
Tracking-lock rate, and it is barely ever reported. It is the proportion of sessions that reached a stable tracked state, meaning the experience actually started rather than merely loaded.
It matters because it separates problems you fix with engineering from problems you fix with creative. If lock rate is low, the environment or the build is defeating people, and no amount of better content will help. If lock rate is high and completion is low, the build is fine and the experience is not interesting enough.
Without that split, teams iterate on the wrong half. A campaign gets a new creative treatment when the real issue was that half the audience never got past a dark corner of an exhibition hall.
How do you connect AR to revenue?
Decide the connection before you build, because it has to be designed in. The usual mechanisms are a session-scoped identifier carried through to your existing analytics, a voucher or code issued at the end of the experience, an appointment booking, or a basket link that carries the configuration the customer built.
For considered purchases with long cycles, direct attribution is usually not worth chasing. A cleaner approach is a holdout comparison: run the activation in some locations or with some segments and not others, and compare. It is less precise than it sounds and still more honest than attributing every subsequent sale to a scan.
For online purchases, the configuration handoff is the strongest link available. If a customer builds a specification and that specification arrives in a basket or with your sales team, the connection is causal rather than inferred.
Why won't you quote industry uplift statistics?
Because we could not verify them. The AR engagement figures that circulate in decks, the multiples on dwell time and conversion, trace back to vendor marketing rather than to a published methodology, sample or control group. They are repeated until they acquire the texture of fact.
They are also structurally optimistic. The campaigns that get written up are the ones that worked, run by brands with the budget to do them well, measured by the people who sold them. Nothing in that pipeline produces a number you can plan against.
The useful benchmark is your own first campaign. Instrument it properly, accept that the first set of numbers is a baseline rather than a verdict, and compare the second against the first. Any agency quoting you a specific uplift figure before understanding your products, audience and channel is selling, not forecasting.
What does a realistic first-campaign result look like?
A funnel with an obvious weakest step, which you then fix. That is genuinely the goal. First campaigns are for learning where your audience drops out, and the value of the exercise is that the second campaign is designed with that knowledge.
Expect the drop-off to be somewhere unglamorous. Codes printed too small to scan at the distance people stand. A permission prompt appearing before any explanation of why the camera is needed. Assets too heavy for the connection in the room. None of those are creative failures, and all of them are cheap to fix once measured.
Judge an agency on whether they instrumented the campaign well enough for you to know which one it was.
Common questions
- What is the most important AR campaign metric?
- Tracking-lock rate, the proportion of sessions that reached a stable tracked state so the experience actually started. It is the metric that separates engineering problems from content problems: a low rate means the environment or build is defeating people, and better creative will not help.
- Why are total scans a poor measure of AR success?
- Because scans rise with footfall, media spend and placement, none of which describe the experience. A campaign can report a large scan count while almost nobody completed the experience. Scans measure the invitation, not what happened after it was accepted.
- Can you attribute sales to an AR campaign?
- Partially, and only if it is designed in advance. The reliable mechanisms are a session identifier carried into your existing analytics, a code issued at the end of the experience, or a configuration handed off into a basket or to your sales team. For long purchase cycles, a holdout comparison between locations or segments is more honest than attributing later sales to a scan.
- What AR engagement uplift should we expect?
- We will not quote a figure, because the ones in circulation trace to vendor marketing rather than to any published methodology or control group. Instrument your first campaign properly and treat it as your baseline. Any agency quoting a specific uplift before understanding your products, audience and channel is selling rather than forecasting.
