Why don\u2019t my Amazon and Walmart numbers match?

Amazon Ads reports "Total Sales" - attributed sales within a 14-day (or configurable) lookback window, including sales of other products in the same ad-clicked brand. Walmart Connect reports Gross Merchandise Value, a broader measure that includes both attributed and some halo sales. Target Roundel largely reports at the campaign-rollup level, without exposing the same SKU-level attribution the other two provide.

None of these three terms mean the same thing, and none of them mean "incremental revenue." They are each the retailer’s own accounting of activity near an ad impression - which is a different question from "would this sale have happened anyway?"

Platform-reported revenue answers: "How much activity happened near this ad?" Incremental revenue answers a harder question: "How much of that activity would not have happened without the ad?" A shopper who already had your product in their cart and then saw your ad is counted as a conversion by the platform - but the ad didn’t cause anything.

The gap between the two numbers is usually large and rarely disclosed. It is the reason two people looking at "the same" campaign can walk away with completely different conclusions about whether it worked.

Start by normalizing to a single unit: net sales dollars, not GMV, not "Total Sales," not units. Pull each retailer’s raw sales export for the same date range and the same SKUs, and reconcile against your own shipment or order data where possible - retailer-reported figures should never be taken as ground truth on their own.

Next, build a simple pre/post or geo-holdout comparison: pause spend in a subset of markets or weeks, and compare sales in exposed vs. unexposed periods, adjusting for seasonality using the same weeks in the prior year. This is the manual, free version of what a geo-lift experiment automates - imperfect, but far better than trusting platform attribution outright.

Finally, keep a running log of promotions, price changes, and stockouts alongside your spend data. Most "mystery" swings in retail-media performance trace back to one of these three, not the media itself.

The manual approach works for one brand, one retailer, one quarter. It breaks down fast once you’re running Amazon, Walmart, and Target simultaneously across dozens of SKUs and promotional calendars that don’t line up - the spreadsheet reconciliation that took an afternoon becomes a part-time job.

A dedicated measurement layer does the same reconciliation and holdout logic continuously, across retailers, with quarterly geo-lift experiments replacing the manual pre/post estimate. That is the specific gap DataSivio’s Retail Media Module is built to close - covered on the Analyze page.