Attribution Window Whiplash
A retailer runs a DSP campaign alongside Sponsored Products. The DSP line item uses a 14-day view-through and 14-day click-through attribution window; Sponsored Products uses a 7-day window. The retailer's finance team reports that total attributed sales exceed actual sales by 22% for the month, and both channels claim credit for the same orders. The DSP reports a 6.0 ROAS, Sponsored Products reports 4.2, but the retailer's own incrementality test (geo holdout) suggests the DSP's true incremental ROAS is closer to 1.8. The marketing lead wants to shift all budget to DSP because 'it has the highest ROAS.'
1. Explain why total attributed sales exceed actual sales and why cross-channel ROAS comparisons are misleading here.
Read the worked answer
Two mechanisms cause the overcounting:
- Overlapping attribution windows and cross-channel double counting. The DSP's 14-day view-through window is longer than Sponsored Products' 7-day window, so the DSP can claim credit for conversions that occur after the SP window closes-and both channels can claim the same order if the user saw a DSP ad and clicked an SP ad. Summing channel-reported sales therefore exceeds actual sales (the 22% gap).
- View-through inflation. View-through conversions credit the DSP for users who may have converted anyway; the longer the window, the more credit is absorbed. This inflates DSP ROAS relative to a shorter-window channel. Comparing a 6.0 DSP ROAS (14-day view-through) to a 4.2 SP ROAS (7-day click) is apples-to-oranges: different windows and different credit rules. The geo holdout-which measures incremental lift against a control-is the more reliable signal, and it says DSP incremental ROAS is ~1.8.
2. Recommend how to allocate budget and how to report performance so finance and marketing agree.
Read the worked answer
Allocation: do not shift all budget to DSP based on the inflated 6.0. Use the incrementality result (~1.8 incremental ROAS) as the basis for comparing channels, and allocate toward the channel with the better incremental return, testing marginal shifts rather than an all-in move. Keep some DSP budget if it is incremental at the margin, but the 6.0 is not a reason to consolidate. Reporting:
- Align attribution windows across channels where possible, or report each channel's window explicitly and never sum them as if independent.
- Deduplicate cross-channel conversions using a unified attribution model or a single source of truth (e.g., MTA or a clean-room/last-touch deduped view), and reconcile to actual sales.
- Report incrementality separately from platform-reported ROAS, and use geo holdouts periodically to calibrate. Reasoning: platform ROAS with mismatched windows double counts and overstates; finance needs deduplicated, reconciled numbers, and budget should follow incremental return, not the highest headline ROAS.