Amazon DSP and the Retail Data Moat

Context

Every demand-side platform buys ads. The interesting question about any DSP is not how it bids — most of the bidding tech has converged — but what it knows. The Trade Desk knows the open web. DV360 knows Google’s audiences and YouTube. Amazon DSP knows something different and, for a certain kind of advertiser, more valuable: what people actually bought.

I own the data behind a DSP and an ad exchange, so I spend my days looking at ad events — impressions, clicks, renders — and the eternal problem is connecting those events to an outcome that a CFO cares about. That connection is exactly where Amazon has a structural advantage nobody else can copy by trying harder.

What I noticed

Amazon DSP’s moat is not the buying interface. It is closed-loop purchase data. When Amazon shows you an ad and you later buy the product, Amazon sees both events, because Amazon owns the impression and owns the transaction. That closes a loop that Meta and Google structurally cannot close on their own platforms: they can prove you saw an ad and clicked, but they do not run the store, so they cannot see the purchase without the advertiser handing them conversion data and trusting the match. Amazon does not need to be handed the sale. It rang up the sale.

That is the whole game. The single hardest thing in advertising measurement is tying exposure to revenue with a straight face. Amazon starts with the revenue.

The product thinking

The mechanism that turns this data advantage into a product is the Amazon Marketing Cloud, AMC — a clean room where an advertiser can join Amazon’s pseudonymised ad signals with their own data and run analysis without either side seeing the other’s raw records. This is what lets a brand move past “we got these impressions” to “this campaign drove this many incremental purchases, de-duplicated across touchpoints.” Standard channel reports credit each channel in isolation; a clean room lets you ask the harder, more honest question of what actually moved the outcome.

The strategic bet is clear and, I think, correct: become the outcome-proving DSP for commerce. In a market where every other platform is fighting over attention and reach, Amazon competes on proof. For a performance advertiser — especially anyone selling physical products — a DSP that can demonstrate incremental sales rather than modelled conversions is worth a premium. That is a genuinely differentiated wedge, and it is downstream of owning the transaction, which is not a thing a competitor can build.

Where the moat thins

Here is the honest limit, and it is where I would push back if I were reviewing this roadmap. The purchase-data edge is strongest inside Amazon’s own loop and weakens as it travels off-platform.

Amazon has been expanding AMC and the DSP to measure and buy beyond Amazon-owned properties, and in 2026 it has been widening access — for example, reportedly letting Sponsored Ads advertisers reach AMC without a full DSP requirement, and positioning AMC to bring in off-Amazon touchpoints and third-party data sources (vendor descriptions — verify current terms before relying on them, as these access rules are moving quickly). That is the right direction. But two forces cap how far the advantage extends.

First, a clean room aggregates. The privacy guarantee that makes AMC usable is the same constraint that blurs it: you get de-duplicated, aggregated answers, not the raw, person-level closed loop you have on Amazon’s own inventory. The further a campaign runs from Amazon-owned surfaces, the more the measurement leans on matching and modelling rather than the direct, we-saw-both-events certainty that is the whole point.

Second, signal loss. Off-platform, Amazon faces the same identity erosion everyone else does — cookie deprecation, platform signal restrictions, walled gardens that do not share. The closed loop is airtight when Amazon owns both the impression and the sale. Off Amazon, it owns neither cleanly, and the proof degrades toward the same modelled attribution its competitors already offer.

The PM lesson

Amazon DSP is a beautiful example of a moat that is real but bounded. The data advantage is not a feature you can out-engineer — it comes from owning the transaction — which is why it is durable inside the loop. But durability inside the loop is not the same as portability across the open web. The temptation for the product team is to sell the incrementality story everywhere; the discipline is to be honest about where the closed loop actually closes.

If I were the PM, I would lean hardest into the on-Amazon and adjacent-commerce surfaces where the proof is strongest, and treat off-Amazon as a reach extension measured with appropriate humility — not as territory where the purchase-data moat magically follows. The pitch that wins is “we can prove what you sold.” The pitch to avoid is pretending that proof is equally strong the moment the ad leaves Amazon’s own shelves.