Home › Measurement

Measuring in-store retail media without fooling yourself

The fastest way to lose credibility in in-store retail media is to over-claim measurement. The honest position is that in-store is harder to measure than on-site, that the good methods are old and unglamorous, and that the flashy metrics are usually the least trustworthy. Saying so up front is what makes the numbers you do report believable.

What you can actually measure

• Sales lift against a control group. Run a campaign in some stores and not in comparable ones, then compare. This is the workhorse of in-store measurement — boring, robust, and hard to argue with when the store sets are matched well.
• Proof of play. Whether a given spot actually ran, where and when. Not an outcome, but the precondition for trusting every outcome number on top of it.
• Category and basket effects. Whether the promoted item moved, and whether it pulled or cannibalised neighbouring products.

What you can't (yet) measure honestly

• Individual exposure to individual purchase. In a physical store the screen and the register are usually not linked at the person level, and pretending otherwise invents precision that isn't there.
• "Impressions" as a hard number. Footfall past a screen is an estimate, not a log line. It is useful for planning and dangerous when sold as fact.

The metrics worth arguing about

Attention measurement — cameras or sensors estimating whether people looked — is the most contested area. It can be genuinely useful for optimising placement and creative. It becomes a problem when it is dressed up as a conversion metric or when it quietly crosses into identifying individuals. The defensible use is aggregate and anonymous: did this placement earn more glances than that one, not who looked.

A simple standard

Report one outcome metric you would stake your reputation on (usually control-group sales lift), support it with proof of play, and label everything else — footfall, attention, dwell — clearly as an estimate. A program measured this way grows on trust. One that leads with attention dashboards and incrementality claims it can't defend tends to get audited, and then unwound.

Next: a first pilot that's small enough to measure honestly →