Retail media and closed-loop measurement, explained

Can you prove an ad sold a unit? Retail media is advertising sold by a retailer or a retail network on its own properties, and closed-loop measurement is the attempt to answer that question by connecting ad exposure and purchase inside one data environment. The purchase half of the loop is point-of-sale (POS) scan data.

What is retail media, exactly?

It's the retailer acting as a publisher. The inventory can be a website's search results, an app's placements, a screen at the register, or signage in the store, and the pitch to advertisers is proximity: the message lands minutes or seconds from the purchase decision rather than days. For a CPG (consumer packaged goods) team, that proximity is the appeal and the trap at once, because closeness to the sale is precisely what makes causation hard to untangle from coincidence.

The independent channel enters this conversation later than the big chains did, for a structural reason: one corner store is not a media network. Thousands of them on a shared POS platform are a different matter. A networked channel, like the independent stores running the NRS point-of-sale (nrsplus.com), can be addressed in aggregate, and its registers already record the outcome side of the loop. That structural fact is what moves this discussion from theory to something a media plan can actually use.

What makes measurement closed-loop?

Open-loop measurement lives in two systems: the ad ran over here, sales are reported over there, and an analyst argues the connection. Closed-loop measurement keeps both ends in one environment, so exposure and transaction can be compared directly. In an in-store version, a campaign runs in a defined set of stores for a defined period, and scan data reads those stores against comparable stores that didn't run it.

The honest core of any closed-loop read is the comparison group. Sales rising during a campaign proves little by itself; seasonality, weather, or a distributor push could have done it. Sales rising in campaign stores while matched control stores stay flat is a real argument. The control set is the measurement; everything else is bookkeeping.

Designs vary, and the store-split is only the cleanest of them. Pre-and-post reads in the same stores are common because they're easy, and they're also the most fragile, since anything else that changed in those weeks rides along in the result. When a split isn't possible, insist at minimum on a long pre-period baseline and a read that extends past the campaign's end.

What should you ask of any closed-loop claim?

Start with incrementality versus attribution, because vendors blur them. Attribution asks which sales can be associated with exposure; incrementality asks which sales would not have happened without it. Attribution flatters campaigns by harvesting purchases that were coming anyway. Incrementality, measured against controls, is the number a budget decision deserves.

Then ask about the mechanics. How were control stores chosen, and would you call them comparable if you saw the list? What's the time window, and does it include the weeks after the campaign, when borrowed sales sometimes get paid back? Is the claimed effect in units or dollars? Units resist the flattery that price moves lend to dollar figures. A measurement partner with good answers will welcome the questions, and evasive answers are themselves data.

None of this requires taking anyone's word for a number, which is the point. The channel's baseline behavior is observable in the monthly same-store reporting NRS Insights publishes, and the report archive shows that baseline month after month, which is the backdrop any campaign effect has to stand out against.