Amazon retail media attribution is last-touch within a lookback window: the most recent qualifying click, or in some placements a view, gets full credit for the sale. Sponsored ads and DSP measure different windows and interactions, so their numbers disagree by design. Treat attributed sales as a comparison tool, not as truth.
The short version
- One mechanism underneath everything. A sale is matched backwards to the most recent qualifying ad interaction inside a time window.
- Click-through and view-through are different claims. A click shows intent. A view shows an impression happened nearby. Never average them.
- Windows differ by ad product. Which is why the same week produces different attributed totals in different consoles.
- Attribution measures credit, not cause. It cannot say whether the sale would have happened without the ad.
- Use it relatively. Attribution is excellent for ranking campaigns against each other and poor at grading advertising as a whole.
The mechanism, in plain terms
Every attribution model answers one accounting question: when a sale happens, which touchpoint gets to claim it? Amazon's retail media answers with recency. The system looks back from the purchase, finds the most recent qualifying interaction with your ad inside the lookback period, and assigns the whole sale to it. Nothing is split, nothing is weighted; last touch takes all.
Three consequences follow directly from that design.
- Recency wins credit regardless of influence. A shopper who saw your product in a creator's video, searched your brand, clicked a branded ad, and bought, registers as a branded-ad conversion. The ad that closed gets the credit of the ad that opened.
- View-through inflates quietly. Display placements can claim sales from shoppers who merely scrolled past an impression and later bought. Some of those shoppers were already on their way to buy.
- Consoles disagree honestly. Sponsored ads and DSP measure different interaction types over different windows, so their totals are not comparable and reconciling them to the cent is a fool's errand. Check the current window definitions in your own console before comparing anything; Amazon revises them.
Symptom, cause, response
| Symptom in your reports | Likely cause | What to do |
|---|---|---|
| DSP efficiency looks impossibly good | View-through credit on shoppers already converting | Judge DSP on click-through and on total account lift, not blended returns |
| Attributed sales rise while total sales are flat | Ads reclaiming credit for organic demand | Watch paid share of total revenue, not attributed revenue alone |
| Branded campaigns show stellar returns | Last-touch harvesting demand created elsewhere | Grade branded spend on incremental logic, not console returns |
| Numbers shift days after the period ends | Late attribution inside the lookback window | Let periods settle before judging them; compare like-aged data |
| Two consoles disagree on the same week | Different windows and interaction rules | Pick one source of truth per decision and state it on the report |
How we actually use attribution numbers
Flapen's position, across about 70 managed brands, is that attribution is a ranking instrument. Inside one campaign type, with one window, attributed performance ranks keywords, targets, and creatives reliably, because the measurement bias applies equally to everything being compared. Across campaign types, or as a verdict on whether advertising works, it misleads.
So we anchor decisions in numbers attribution cannot flatter: total sales, paid share of revenue, conversion rate, and profit after all costs. It mirrors how we make product decisions generally, on 90 or more data points spanning market size, growth trajectory, return rate, segment dynamics, and the rating gap, rather than on any single seductive metric. An attributed return figure is one input with a known bias, and known biases are manageable.
One practical note: attribution disputes often dissolve when the funnel underneath improves. Raising baseline conversion through listing optimization lifts every channel's measured performance at once, which is a strong hint about where the real leverage was.
What the attribution console will not tell you
It will not tell you what would have happened without the ad, and that counterfactual is the only question that decides whether spend was worth it. The console reports credited sales, and credited is a bookkeeping word, not a causal one. The practical test for any agency or tool reporting your advertising: ask them where the console overstates. Anyone who runs retail media seriously can name view-through inflation and branded harvesting without hesitating. Anyone who answers that the numbers speak for themselves is reading you the receipt, not the reality.
Related answers
- Compare Amazon PPC vs DSP performance metrics
- Amazon advertising KPIs that matter
- Which metrics predict Buy Box win rate
- Best audit for Amazon PPC structure and neg match
- Amazon creative services: the complete guide
For an attribution-aware read of your own ad account, request the free 48-hour audit at Flapen.
