DSP retargeting, incremental lift vs reported ROAS across 80 audits

6.2x reported ROAS. 1.4x incremental ROAS. That’s the average gap we found across 80 DSP retargeting audits run between Q2 2024 and Q1 2026.

The 4.8x delta is not measurement noise. It is purchases Amazon would have closed without the ad, attributed to the ad anyway, then billed to your CPM line.

If you run DSP retargeting and your reporting deck shows 5x+ ROAS without an incrementality test next to it, you are funding a number that is mostly fiction. Below is what the audits showed, how the math actually works, and what to cut on Monday.

What 80 accounts told us about reported vs incremental ROAS

The audit set: 80 mid-market Amazon brands, $5M-$120M in annual GMV, all running DSP retargeting against ASIN viewers and cart abandoners. Median monthly DSP spend: $42,000.

We isolated the retargeting line items. We compared Amazon’s reported ROAS against an incremental ROAS calculated via geo holdouts, audience holdouts, or, when neither was available, a regression on weekly spend volatility against organic baseline.

The headline: 71 of 80 accounts had a reported-to-incremental ratio above 3:1. Twelve had a ratio above 6:1. Three accounts had negative incrementality. They were paying Amazon to retarget customers who would have purchased faster without the ad nudging them into a comparison-shopping window.

The accounts with the worst gaps shared three traits. Heavy reliance on viewed-ASIN audiences, no frequency cap below 6 impressions per 14 days, and a creative library refreshed less than once per quarter.

Why retargeting ROAS reports lie by default

Amazon’s DSP attribution model is last-touch within a 14-day window. If a shopper viewed your detail page on Monday, got served a DSP retargeting impression on Wednesday, and purchased on Friday, that purchase is credited to DSP. Even if they were going to buy on Friday anyway. Even if the DSP impression was a 1-pixel below-the-fold display unit they never consciously saw.

This is not a bug. It is how every walled-garden ad platform reports. The bug is that agencies and in-house teams take the report at face value and let the CFO believe a 6x ROAS is real cash on cash.

Incrementality testing tells you what would have happened without the ad. The math is simple. Run a holdout. Hold 20% of your retargetable audience out for four weeks. Measure organic conversion in the holdout vs the exposed group. The delta, divided by spend, is your real ROAS. Not the inflated last-touch number.

Of the 80 accounts we audited, 7 had ever run a holdout. Seven. The other 73 were optimizing toward a number that did not measure what they thought it measured.

The four buckets of inflated ROAS we kept finding

Across the audit set, the inflation grouped into four patterns. Each maps to a specific tactical mistake.

Bucket 1, Cart abandoner audiences with sub-1-day lookback. A shopper who added to cart 6 hours ago is mostly going to convert. A DSP impression in that window steals credit from organic. Reported ROAS 8-15x. Incremental ROAS 0.6-1.8x. We saw this in 34 of 80 accounts.

Bucket 2, Viewed-ASIN audiences with no exclusions for recent purchasers. If you do not exclude buyers from the last 60 days, you retarget loyal customers who would have repurchased anyway. Reported ROAS 5-9x. Incremental ROAS 1.1-2.3x. We saw this in 41 of 80 accounts.

Bucket 3, Frequency-uncapped retargeting hitting the same shopper 12+ times. Diminishing returns kick in hard after impression 4. Past 8, you are paying to remind buyers who already decided. Reported ROAS holds because the last impression always gets credit. Incremental drops to near-zero. We saw this in 28 of 80 accounts.

Bucket 4, DSP credit for sales that Sponsored Products closed. When DSP and SP both touch a shopper and SP gets the click, Amazon’s view-through window often gives DSP credit anyway. Reported DSP ROAS gets a free ride on SP intent. We saw this in 19 of 80 accounts where the DSP and SP teams reported separately and never reconciled.

What the brands with real incremental lift had in common

Nine of the 80 accounts had reported-to-incremental ratios under 2:1. They did three things differently.

First, every retargeting line item had a documented exclusion list. Recent purchasers, current cart-holders within 24 hours, and brand-search clickers from the last 7 days were all excluded. Spend went only to shoppers in genuine consideration limbo.

Second, frequency caps were aggressive. Median cap across the high-incremental cohort was 4 impressions per 7 days, with 3 of the 9 capping at 3 per 7. The accounts with bloated reported ROAS averaged 11 impressions per 14 days.

Third, they ran quarterly holdouts. Not as a one-time exercise. As a permanent line item in the media plan. 20% holdout, four weeks, every quarter. The holdout cost is the price of knowing whether the rest of the budget is working.

If you are a brand that has outgrown Sponsored Brands and is being pitched DSP as the next layer, read our breakdown of Amazon DSP for the brand that’s outgrown Sponsored Brands before you greenlight the budget. The platform works. The default reporting does not.

The Monday checklist if you run DSP retargeting

Six actions, in priority order. Do them this week.

1. Pull the last 90 days of DSP retargeting line items. For each, calculate spend, reported sales, and reported ROAS. This is your baseline.

2. Add an exclusion for purchasers within the last 60 days to every retargeting line item that does not have one. Expect reported ROAS to drop 15-30%. The drop is not a regression. It is reality showing up.

3. Set frequency caps to 4 impressions per 7 days as a starting point. Tighten from there if your category is high-consideration.

4. Schedule a 4-week, 20% audience holdout starting next pay period. Treat it as the cost of knowing. Holdouts do not need executive approval, they need a calendar invite.

5. Reconcile DSP and Sponsored Products attribution. If both are claiming the same conversion, build a deduplication report. Most brands find 8-15% of attributed DSP sales were actually SP-closed.

6. After the holdout, recalculate ROAS as (exposed group sales − holdout group sales × scale factor) / spend. That is your real number. Use it for the budget conversation.

The reason this matters is not philosophical. It is that DSP retargeting budget is currently being pitched to founder-led brands as the obvious next step after Sponsored Brands maxes out. Some of that budget is incremental. A lot of it is not. The brands that will win the next two years are the ones who can tell which is which before they sign the IO.

Want a no-charge incrementality audit on your DSP retargeting? Book a 30-minute strategy call and we’ll show you the gap on your own data.