Retail Media Strategies: How to Build One That Proves It Worked

What a retail media strategy actually is

A retail media strategy is the set of decisions about where ad budget goes across a retailer's owned inventory — sponsored product listings, on-site display, off-site placements, and programmatic display, video, or audio bought through a demand-side platform — plus how you prove which of those placements actually drove a sale, rather than just claiming credit for one that would have happened anyway.

Most guides to retail media stop at definitions and format lists: what sponsored products are, what on-site display looks like, why first-party data matters in a post-cookie world. That's the easy half. The hard half is choosing the split between formats and then measuring whether the split was right. That's what this page covers.

The four decisions inside every strategy

Every retail media strategy that actually functions, regardless of retailer or budget size, is really four decisions stacked on top of each other. Skip the fourth and it doesn't matter how carefully you made the first three — you'll never know if you got them right.

A worked example: what 6.04x ROAS actually tells you

Numbers without context are marketing copy. Here's a real one, scoped honestly: across 30 advertisers managed in July 2026, the book delivered 6.04x return on ad spend, 78.4 million impressions at a $4.00 CPM, and a blended $1.42 cost-per-click. Blended cost per acquisition was $5.49 across 57,137 attributed purchases, 20.1% of them from shoppers new to the brand.

What those numbers mean together: a 6.04x ROAS on its own could mean the campaign is efficient, or it could mean it's mostly reaching people who were buying anyway and the platform is claiming the sale on the way past. The 20.1% new-to-brand figure is what makes the ROAS credible — a fifth of attributed purchases came from people who hadn't bought the brand before, which is the job display and video are supposed to do that sponsored search alone can't.

To check this on your own account: pull attributed purchases split by new-to-brand versus repeat, then run a holdout — a matched group that didn't see the DSP ads — over the same window. If the holdout converts at close to the same rate as the exposed group, the ROAS is real but the incrementality isn't. That gap is the number retail media reporting almost never shows you, because last-click attribution can't produce it. Amazon Marketing Cloud can, once DSP and sponsored ads are reconciled so they stop taking credit for each other's sales.

The mistake we've made ourselves

Campaigns often launch on a hypothesis — audio ads will lift search volume for a cookware brand, say — with a media plan attached but no measurement plan. Three weeks in, CPMs look fine, ROAS looks fine, and nobody can say whether any of it moved a number that wouldn't have moved anyway. We've shipped tests that way ourselves: a plausible idea, a clean dashboard, and no holdout group set up before launch — which means by the time someone asks whether it worked, the honest answer is that it wasn't designed to answer that.

The fix isn't more dashboards. It's deciding the measurement method — holdout, matched control, or none — before the campaign runs, not after someone asks for a readout. A strategy that skips this step isn't a weak strategy; it's not a strategy. It's a media plan with a scoreboard nobody trusts.

When the strategy isn't working — what to actually check

If ROAS drops, the instinct is to change the bid. Check three things first, in order. First: did the audience or placement mix change, or is this the same campaign just converting worse — a mix shift often gets misread as a performance problem. Second: is the attribution window or model the one you used last time you compared numbers — a lot of "it stopped working" turns out to be "we changed how we count."

Third, and the one people skip: was there ever a holdout to compare against? Without one, "it's not working" and "it never worked, we just noticed now" look identical. A brand running DSP in a considered-purchase category — the kind of space HexClad or Beardbrand sit in — will see slower, noisier signal than a low-price impulse category, and that noise gets mistaken for failure when there's no control group to separate signal from seasonality.

When a fix genuinely doesn't work, the honest move is to roll back to the last known-good state and say so — not to keep layering adjustments on top of an untested change until something looks better by coincidence.

Where Dr. DSP fits into this

Dr. DSP is Full Circle's managed version of Amazon DSP — the demand-side platform, not the delivery courier program — built around the argument on this page: last-click can't prove incrementality, holdouts and matched controls can, and every change ships with the evidence behind it, a measurement plan, and a rollback trigger already attached. Full Circle and reMKTR have managed over $500M in revenue across 100+ brands. There's no published price — a demo, the first 30 days free, and terms set on a call against real spend and scope — which is worth knowing whether you buy from us or just want to compare it against whatever your current DSP quote looks like.