Audience Extension: Reaching Amazon Audiences Off Amazon
Audience extension is Amazon DSP buying inventory on third-party exchanges — the open web, apps, publisher sites — while still targeting based on Amazon's own shopping and purchase-intent signals. It's how a brand reaches someone Amazon knows shops for a category, on a website that has nothing to do with Amazon at all. It behaves differently, in measurable ways, from inventory served on Amazon's own properties.
What this looks like in a real account
Why the term itself causes confusion
"Audience extension" is Amazon's own framing for a pattern that exists across most demand-side platforms in some form — the idea that an audience defined on one data source can be reached on inventory that source doesn't own. What's specific to Amazon's version is the underlying signal: shopping and purchase-intent data at a scale most other platforms simply don't have access to, built from an actual marketplace rather than inferred browsing behaviour. That's the reason audience-extended Amazon inventory is worth treating differently from a generic programmatic buy, even though the mechanics of buying exchange inventory are broadly similar across platforms.
The mechanism, in plain terms
Amazon DSP connects to ad exchanges and supply-side platforms to access inventory across the open web, in addition to Amazon's own owned properties like Prime Video, Twitch and thousands of third-party apps and websites in its direct network. The audience targeting is built from Amazon's shopping signals regardless of where the ad ultimately serves — a shopper who's browsed a category on Amazon can be reached on a completely unrelated news site or app, using that same underlying signal, because the audience definition travels with the buy even when the inventory itself doesn't belong to Amazon.
Why extended, off-Amazon inventory behaves differently, measurably
In our own DSP book, off-Amazon exchange inventory returned a higher ROAS than Amazon's own shopping surfaces — 6.05x against 4.90x — across 27 advertisers in a recent 31-day window. But new-to-brand share told the opposite story: exchange inventory converted at only 15.9% new-to-brand, against 42.0% on Amazon's own surfaces. The two supply sources are doing genuinely different jobs — extended, off-Amazon inventory tends to convert shoppers already close to a purchase decision efficiently, while Amazon's own shopping surfaces do more of the actual recruiting, at a lower return multiple. Judging both on ROAS alone quietly starves the placement doing the acquisition work.
A worked example of what this means for budget allocation
Take a $50,000 monthly DSP budget split proportionally to our own book's pattern — roughly 63% Amazon-owned ($31,500) and 37% extended exchange ($18,500). If a stakeholder unfamiliar with the placement-level breakdown sees only the blended ROAS and pushes to shift budget toward the higher-returning exchange inventory, the account's new-to-brand rate will fall as a mechanical result — not because targeting got worse, but because the mix shifted toward the placement type that structurally recruits fewer new customers. Understanding the audience-extension mechanism is what prevents that misread.
134 distinct places this can actually land
A single month of Amazon DSP delivery across our book touched 134 distinct supply sources in a recent 31-day period — Amazon's own shopping surfaces, its streaming inventory, and dozens of third-party exchanges. Most explainers of Amazon DSP describe two or three placements; the real spread is far wider, and the performance difference between individual supply sources was larger than the spread between different advertisers entirely. Buying audience-extended inventory without a supply-source-level report is effectively buying an average of things that behave nothing alike.
The common mistake, including ours
The mistake is treating "off-Amazon exchange inventory" as one uniform placement type rather than a category containing wide internal variance across 134 different sources. We've reported audience-extension performance as a single blended number in early client updates, before breaking it out by individual supply source revealed that a handful of exchanges were driving most of the return while others were essentially dead weight — a pattern the blended number completely hid.
What audience extension means for a Shopify-and-Amazon brand specifically
A brand also running link-out DSP campaigns to its own site can use audience-extended, off-Amazon inventory for either destination — the audience signal is the same regardless of where the click lands. That's worth knowing because it means the placement-level performance pattern described above (higher ROAS, lower new-to-brand on exchange inventory) is a real consideration for a link-out strategy too, not just for campaigns pointed back at Amazon. A link-out campaign built entirely on exchange inventory may convert efficiently but skew toward shoppers who were already likely to buy somewhere, rather than genuinely expanding the customer base the way Amazon-owned placements tend to.
How to review audience-extended inventory properly
Pull a supply-source-level breakdown, not just an Amazon-owned-versus-exchange split — the internal variance across the 134 sources matters as much as the two-way split does. Weight review time by spend concentration rather than an alphabetical list, since spend on DSP tends to concentrate heavily in a minority of line items. And always pair ROAS with new-to-brand rate for any audience-extension review, since the two metrics tell genuinely different, complementary stories about what that inventory is actually doing.
| Supply type | ROAS (27 advertisers, 31 days) | New-to-brand rate |
|---|---|---|
| Amazon-owned shopping surfaces | 4.90x | 42.0% |
| Off-Amazon exchange inventory | 6.05x | 15.9% |
| Total distinct supply sources touched | 134 in one month | Wide internal variance across all of them |
Which one you should actually pick
Any advertiser with DSP reporting access can pull a supply-source-level breakdown themselves — it's part of standard reporting, just often unread below the account-level blend, and no special tooling is required to look at it properly. reMKTR reviews audience-extension performance at the supply-source level on every managed account, having found the blended number hiding real variance before, as part of the discipline behind Full Circle's $500M+ in managed Amazon spend across 100+ brands.
Shortlist on the job, not the feature grid. Pull your search-term report for the last 90 days and total the spend against terms that produced no orders — 33.6% on the account above. Then ask each vendor on your list what they would do about it in week one, and see who answers with a process rather than a screenshot.
Common questions
What is audience extension on Amazon DSP?
Buying ad inventory on third-party exchanges and the open web while still targeting based on Amazon's own shopping and purchase-intent signals — reaching Amazon-known audiences on websites that have nothing to do with Amazon.
Does off-Amazon exchange inventory perform better than Amazon's own placements?
On ROAS, often yes in our data — but on new-to-brand rate, the opposite: Amazon's own shopping surfaces recruited more new customers in our sample. The two placement types do different jobs.
How many different supply sources does Amazon DSP actually reach?
134 distinct sources in one recent 31-day period across our book — far more than the two or three placements most explainers describe.
Should I judge audience-extended inventory purely on ROAS?
No — pair ROAS with new-to-brand rate, since a placement's return multiple alone doesn't tell you whether it's recruiting new customers or converting people already close to buying.
We show the method before the number.
Claim the free auditRead next
- Tinuiti Pricing: How the Quote Gets BuiltPricing · tinuiti pricing
- Criteo vs Amazon DSP vs reMKTR: Which Demand You BuyHead to head · criteo vs
- Pacvue Pricing: No Public Number — What to AskPricing · pacvue pricing
- Skai vs Pacvue: Contracts, Not Feature GridsHead to head · skai vs pacvue