As of 8 November 2026, advertisers on Amazon can build an audience from what people buy at other retailers. The data is not Amazon's: it comes from Attain, a purchase-data provider, and it plugs into Amazon Ads DVA+, the display, video and audio buying product that replaced the Amazon DSP name in September. Attain said the integration went live on 5 November. For consumer brands, the interesting question is not whether this is available, but what a rented purchase graph can and cannot tell you about a category.
What changed on 5 November
The mechanics are narrow and worth stating plainly. A brand can ask for people who bought a given category at a named chain, in a physical store, recently, and then pay to reach them on Amazon inventory including Prime Video and third-party supply. Attain was already inside Amazon at the far end of the campaign, as one of the first automated offline sales lift measurement partners of the former Amazon DSP. What changed is that the measurement panel now also picks who the campaign reaches.
Coverage of the launch is thin on specifics. The announcement names no app, gives no pricing, no segment count and no list of retailers in its taxonomy, and does not say whether transactions come from linked cards, scanned receipts, connected email or a mix. Attain's own description wobbles on refresh speed, calling the signals daily in the body and real-time in the company description. None of that makes the product unusable, but it does mean the first month is a test, not a rollout.
What a segment can and cannot say
The three variables
Retailer, category and channel are the only three knobs, and each one carries an assumption. Retailer-level targeting assumes the chain named in the segment is where the purchase happened, not where the shopper usually shops. Category assumes a taxonomy that may not match your own product hierarchy. Channel assumes the panel can tell an in-store purchase from an online one, which depends entirely on the capture method the vendor does not disclose.
Where the panel stops
A panel of roughly 14 million people is large in absolute terms and small relative to the population that shops at any major grocery chain. That matters for a simple reason: the smaller the overlap between your target category and the panel, the more the audience is a modelled projection rather than an observed list. Vendor evidence should be read with the same caution. A Trade Desk post on 5 October reported that 90% of campaigns beat Attain's own benchmarks for new-buyer return on ad spend by an average of 122%, but gave no study period, holdout design or campaign count.
Why this touches category growth
For a brand that sells across several retailers, the value is not extra reach. It is the ability to see whether a shopper who buys your category at one chain behaves the same way at another, which is the question most category growth plans quietly assume an answer to. If a shopper buys your category only in store at one chain and only online at another, one creative and one pack size will underperform in one of the two, and no amount of Amazon-side data would have shown it.
| Data source | Question it answers | Blind spot |
|---|---|---|
| Your own first-party sales | What sold, at what price, on which listing | Nothing about shoppers who bought elsewhere |
| Rented purchase graph | Whether a shopper also buys the category at another chain | Panel coverage and undisclosed capture method |
| Reviews and search data | Why the category is chosen or rejected | No link to actual purchase behaviour |
That table also explains why the three sources should not be blended into one score. They answer different questions at different levels of confidence, and a plan built on the weakest of the three will inherit its blind spot. Consumer demand shows up in the first-party data first, in the rented graph second, and in reviews last, so the sequencing of what you read is itself a signal.
A test plan before first-quarter budgets
The practical use of a new targeting input is not to expand a campaign but to falsify an assumption. Pick one category belief you have been acting on, design a segment that would behave differently if the belief were wrong, and run it small enough that a null result is still affordable.
None of this depends on the vendor's claims being wrong. It depends on the campaign being designed so that the answer is visible either way, which is the only way a rented data source becomes an input to category planning rather than another line in a media plan.
FAQ
Is this the same as Amazon's own shopper data?
No. Amazon's signals stop at the edge of its own store. This is a third-party panel covering purchases at other retailers, activated inside Amazon's buying tool.
Does the panel cover everyone who shops at those chains?
No. It covers people who opted in to the data provider's apps, which is a large but partial slice of any chain's shopper base.
Can it be used for products not sold on Amazon?
Yes, that is part of the pitch. Brands without an Amazon catalogue can still use the segments against Prime Video and third-party inventory.
What should a brand check first?
The capture method and the retailer list. Without them, a competitor-purchase segment is an assertion rather than a measurement.
