The DSP Retargeting Overlap Problem
A beauty brand runs Amazon DSP retargeting to site visitors and to 'similar to' audiences, plus a Sponsored Products campaign on branded terms. Monthly DSP spend is $30,000; SP branded spend is $8,000. The brand's analytics team reports that reported DSP-attributed conversions and SP-attributed conversions together exceed total actual orders by about 18%. The DSP retargeting line shows a 2.9 ROAS, but the team suspects it is claiming credit for orders that would have happened anyway via SP branded search. Leadership wants to cut DSP because 'SP is cheaper and the DSP ROAS looks inflated.'
1. Explain what is likely causing the 18% overcount and why the DSP ROAS may be misleading.
Read the worked answer
The overcount is almost certainly attribution overlap and double-counting, not real incremental orders. Two mechanisms:
- Cross-channel overlap: the same shopper sees a DSP retargeting ad and later converts through an SP branded click. Both systems can claim the order under their respective attribution windows, so the sum exceeds actual orders.
- Retargeting selects for high-intent users who were going to buy anyway. A 2.9 ROAS on site-visitor retargeting largely captures existing demand, not demand the DSP created. That inflates apparent DSP performance relative to its true incremental contribution.
So the DSP ROAS is 'last-touch friendly' and the SP branded ROAS benefits from the same shoppers. The 18% gap is the tell that the two channels are not independent. Cutting DSP purely on the inflated comparison would be a mistake without an incrementality read.
2. How would you measure the true incremental value of the DSP retargeting before deciding whether to cut it?
Read the worked answer
Move from attributed ROAS to incremental measurement:
- Run a holdout/control group: suppress DSP retargeting for a randomized slice of the audience and compare conversion rates against the exposed group over the same period. The lift is the incremental effect.
- Use Amazon Marketing Cloud (AMC) to de-duplicate paths - measure how many converters touched DSP only vs. SP only vs. both, and compute path-level contribution rather than summing channel reports.
- Apply a consistent attribution window across DSP and SP so the comparison is apples-to-apples.
- Segment retargeting by recency: recent site visitors are more likely to convert anyway; the incremental value is likely concentrated in lower-recency or similar-to audiences.
Decision rule: if the holdout shows meaningful lift, keep or reshape DSP; if lift is near zero, reallocate toward upper-funnel or prospecting where incrementality is higher.
3. Leadership wants to cut DSP because SP looks cheaper. How do you respond, and what reallocation (if any) would you recommend?
Read the worked answer
Push back on the premise gently: SP branded is cheap because it captures existing demand on your own brand terms - it is not a substitute for demand creation. Cutting DSP entirely may shrink the pool that SP branded later harvests. But the inflated DSP ROAS does justify reshaping, not blindly defending, DSP.
Recommended path:
- Do not cut DSP on the strength of a non-incremental ROAS comparison.
- Run the holdout/AMC analysis first; make the cut decision on incremental lift, not attributed ROAS.
- If retargeting lift is weak, shift DSP budget from broad site-visitor retargeting toward prospecting/similar-to and upper-funnel audiences where incrementality is more likely, while keeping a lean retargeting layer to close the loop.
- Keep SP branded protected; it is efficient and complements DSP.
- Set a test budget and a decision date so leadership sees a measured answer, not an opinion.