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

What a retail media strategy actually covers

A retail media strategy is the plan for how you split budget across three layers of Amazon advertising: sponsored ads that capture people already searching, on-site and off-site display that build consideration, and Amazon DSP — the Demand-Side Platform, Amazon's programmatic buying tool, not the Delivery Service Partner courier franchise — that reaches shoppers off Amazon using purchase data. Get the split wrong and you either starve demand capture or waste money building demand on top of demand that already existed.

Most brands build the sponsored ads layer first because it's easiest to measure: someone searches, clicks, buys, and the attribution chain is direct. The mistake is stopping there. Sponsored ads can't reach someone who hasn't started searching yet. Display and DSP exist to create that demand before the search happens — and they need a different way of proving it worked, because there's no click to point to.

A worked example, using real numbers

Here's what the funnel looks like when it's running. Across 30 of those advertisers 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.

Break that down and the strategy becomes visible. A $4.00 CPM buying 78.4 million impressions is roughly $313,600 in display and video spend doing awareness and consideration work. The $1.42 CPC sits on top of that, doing the conversion work sponsored ads and DSP retargeting are built for. The 20.1% new-to-brand figure is the number that tells you whether the top of the funnel is pulling in people the brand didn't already have — if that number is low, the display spend is expensive retargeting dressed up as prospecting.

The ratio between what's spent building awareness and what's spent converting it is the real strategic lever. Total budget matters less than that split.

Why last-click can't tell you if it's working

Last-click attribution gives every conversion to whichever ad ran last — usually a sponsored ad retargeting someone who was going to buy anyway. That's fine for judging sponsored ads on their own. It cannot tell you whether the display and DSP spend above it caused any of those sales, because last-click was never built to test what would have happened without the ad. It just records what happened after.

The only way to answer that is a holdout: hold back a matched group of shoppers from seeing the campaign, run the rest normally, and compare outcomes. The gap between the two groups is incremental lift — sales the campaign actually caused. Reconciling that inside Amazon Marketing Cloud matters too, because DSP and sponsored ads log against the same shopper. Without dedup, one purchase gets counted twice and display looks like it did more than it did.

Common mistakes, including ones we've made

• Chasing ROAS instead of incrementality. A campaign can post a strong ROAS number built almost entirely on people who would have bought anyway. We've shipped campaigns that looked good on ROAS and, once tested against a holdout, added close to nothing. The fix isn't killing display — it's testing before scaling it.
• Turning on DSP before sponsored ads are converting well. DSP finds new shoppers; if the listing they land on isn't ready, they bounce and you've paid to lose them once.
• Skipping AMC dedup. Without it, DSP and sponsored ads both claim the same purchase, so the combined reported return is inflated against what actually happened.
• Sizing off-site budget by what's left over rather than by the addressable audience beyond what search already captures. Off-site spend should be sized to opportunity, not to leftover quarter budget.

What to do when the number is bad news

Sometimes the holdout comes back small, or the new-to-brand rate is low. That's a real result, not a failed test — it tells you something specific needs to change: creative, audience, placement, or the split between on-site and off-site. The wrong response is to keep the spend running because the sponsored ads layer is still hitting its own target. DSP needs its own bar, checked separately.

Every change should carry three things before it goes live: the evidence behind it, a measurement plan for how you'll know if it worked, and a rollback trigger — the number that, if hit, reverts the change automatically rather than waiting for a monthly review. That discipline matters more than any single tactic, because a retail media strategy that can't be tested is just a budget with a story attached.

Where a managed DSP practice fits into this

If you're running all of this in-house and it's working, keep doing it — none of the above requires a vendor. Where a managed practice tends to help is the measurement layer: running holdouts on a regular cadence, reconciling in AMC every cycle, and having guardrails set before a change ships rather than after it underperforms.

That's the piece Dr. DSP is built around: Amazon DSP run as a managed product by Fable 5, part of Full Circle, which has managed more than $500M in revenue across 100+ brands. You set the autonomy level — full approval, supervised, or autonomous inside guardrails you define — and the reporting suite, Orbit, is included. There's no published price. It's a demo, the first 30 days free, and a quote set against your actual spend and scope on the call.