Marketing

Before you cut the in-store budget, ask how the number was built

An under-measured channel and an underperforming one produce the same return on ad spend. They call for opposite decisions.

Generated by AI/Adobe Stock

October 6, 2026 by Vishak Nag Ashoka — Senior Director Of Engineering, Walmart

Summarize withChatGPTGrokPerplexityClaude

Ask a brand marketer why they have not moved more budget into in-store retail media and you get the same answer: the ROAS does not compare favorably to digital.

The industry seems to have accepted this as a verdict on in-store advertising's effectiveness. It is not. It is a verdict on its measurement, and those are different problems with opposite solutions.

The comparison is not like for like

When a marketer compares in-store ROAS to paid search ROAS, they are comparing two numbers produced by radically different levels of instrumentation.

Paid search has near-total observability. Every impression, click and conversion sits in one identity graph, timestamped and attributable within milliseconds. Fifteen years of industry investment went into making that true. Much of that number is infrastructure, not performance.

In-store has almost none of that by default. A shopper walks past a screen. Maybe they see it. Maybe they buy it forty minutes later, three days later, at a different store, or online. There is no cookie, no device ID, no login. Just a person in a physical space.

So when in-store underperforms, a significant share of that gap is not lost performance. It is unobserved performance. The conversions happened; the infrastructure to see them was not there.

That distinction decides the budget. An underperforming channel should be cut. An under-measured channel should be instrumented — and cutting it guarantees you never learn what it was worth.

Three gaps, and what each one costs you

Identity. Digital resolves a person to an ID. In-store usually cannot, and every workaround covers a partial, self-selecting slice. Loyalty matching sees loyalty program members, who shop differently and more often than everyone else. Receipt capture sees people motivated enough to submit a receipt. Mobile location sees people who opted into location sharing. None are wrong. All are narrower than the audience you reached, and the measured slice rarely represents the unmeasured one.

Delivery verification. This is the gap the industry discusses least and should discuss most. In digital, you generally know the ad rendered. In-store, most systems know the ad was scheduled. Those are not the same claim. A schedule can dispatch, a player can acknowledge it, a log can record a playout, and the screen can have been dark since before the store opened. Every impression counted on that screen then feeds an attribution model as though it happened. The error does not announce itself. It quietly lowers your measured return.

Conversion lag and cross-channel leakage. An in-store impression can produce a purchase in a different session, a different store, or a different channel entirely. Attribution windows inherited from digital assume a tighter loop than physical retail has. Set a seven-day window on a category people buy monthly and you will systematically under-credit the channel, then conclude the channel is weak.

None of these are unsolvable. All require investment that does not show up in next quarter's revenue.

The errors compound in one direction

These gaps do not simply add noise. They bias.

Unresolved shoppers drop out of the numerator. Unverified delivery inflates the denominator. Truncated attribution windows discard late conversions. Every one of those errors pushes measured ROAS down, never up. The result is a channel that reads worse than it performs, and a budget process that responds by starving it further.

That is the quiet cost of treating a measurement problem as a performance problem.

4 questions worth asking

Rather than asking a retail media network what their ROAS is, ask questions that expose how the number was produced:

1. How do you verify the ad actually played — not that it was scheduled, but that it rendered on a powered, visible screen?
2. What share of my audience can you resolve, and how do resolved shoppers differ from unresolved ones?
3. What is your attribution window, and what did you do to validate it against real purchase behavior rather than inheriting it from digital?
4. Can you show incrementality — a holdout, a matched-market test, something with a control group?

The last one matters most. Incrementality testing is the only method here that does not depend on resolving every shopper. A well-designed holdout measures the lift the channel caused, whether or not you traced each path to purchase.

Why this is a standards problem

This does not resolve through competition, because there is no common definition to compete on. Every retail media network measures slightly differently, so the numbers are not comparable to each other, let alone to digital.

That is why the standards work matters. I served on the IAB working group that produced the DOOH and In-Store Retail Media Playbook, and the most valuable part was not the document. It was watching a room of competitors discover how differently they had each been defining the same words. You cannot build a common currency until you agree what you are counting.

We are not there yet. But the last two to three years have moved faster than the previous ten.

In the meantime, the marketers who win this transition will be the ones who stop treating ROAS as a like-for-like comparison and start interrogating how the number was built. The channel is not underperforming. It is under-instrumented — and that is a far better problem to have.

About Vishak Nag Ashoka

---