Ross Walker is the Director of Retail Media at Acadia.
For most of the last decade, running Amazon Ads at scale meant accepting a certain amount of waste as the cost of doing business.
You picked your keywords. You picked your in-market audiences. You bid. And then you paid to reach every single person who met that criterion, whether they were your ideal customer or someone who would never buy from you twice.
Amazon Marketing Cloud changes that. Not because it gives you more data, but because it lets you get narrower. It sharpens advertising through two complementary levers: Insights, which will tell you where the opportunity is. And custom audiences, which will allow you to turn those insights into action.
It finally answers the question: How do we refine our targeting so we don’t need to try and boil the ocean, hitting every person in the market or all new entrants to the category?
It’s Not Just About What Someone Searches For
Here's the shift in practice: I'm not just looking for someone who's searching for dog food.
I'm looking for someone who's searching for dog food and shares the same demographic characteristics, buying habits, interests, life stage, and lifestyle as our most valuable customers.
That overlay is one of the highest-impact elements we've found in AMC. Not the queries. Not the dashboards. Taking a basic search or audience criterion and refining it down to the people who are statistically more likely to become valuable customers, using lookalike audiences built off actual purchaser data.
Same keyword. Same auction. Radically different economics.
The alternative is what everyone was doing before: treating a keyword as if it were an audience. It isn't. A category keyword is a room with a thousand people in it, and maybe eighty of them look anything like your best customers. AMC is how you find those eighty and pay to reach them specifically.
From Targeting to Sequencing
So we talked about the value of who the user is. The second unlock is based on what they've already done and what they've already been exposed to.
A person sees a Streaming TV ad. Later, they view the product detail page. Someone who has done both can be served a different, more specific message. Then, if that person searches a category keyword after all of that exposure, we can bid much higher for them, because their odds are dramatically better than a cold searcher's in the same auction.
That's a sequential funnel, and it only works when the signals connect. AMC isn't just about finding the right person. It's about recognizing that the right message depends on where that person is in the journey.
At awareness, the job is reaching new and adjacent audiences who haven't considered you yet: Streaming TV, Prime Video, Online Video.
At consideration, the job changes. Someone already exposed to the brand shouldn't get the same creative again. A second message should deepen intent and keep you top of mind while they research.
At conversion, you can get “aggressive”. Bid up in search and off-site display for shoppers who have visited the PDP or thrown off other strong signals.
And after purchase, the objective changes again: cross-sell, replenishment, subscription, loyalty.
The more signals you can connect, the more deliberately you can manage those transitions. Right message, right person, right moment. That's what turns a collection of campaigns into a connected customer journey.
And It Finally Lets You Defend the Upper Funnel
There's a second-order benefit here that I'd argue is just as valuable as the efficiency gain.
We can now say: here are the sales that came through from this specific subset of people who were exposed to those upper-funnel ads.
If you've ever had to justify a Streaming TV budget to a CFO, you know why that matters. The upper funnel has always been the first line item cut, because it was the hardest to defend.
Connecting exposure to downstream purchase gives you a real basis for valuing awareness spend instead of an argument about brand health. In our experience, brands that can measure the upper funnel fund it properly. Brands that can't, don't.
The 5-Year Lens
All of these levers get considerably more powerful when you take a longer view of your customer base.
Twelve months isn't a complete picture of customer value. In categories with long purchase cycles, a customer who hasn't bought in the past year may still be highly valuable. And a customer who purchased twice may be worth far less over time than someone who has bought consistently for years.
Amazon's five-year retail purchase lookback, extended from the old 13-month window, gives brands a much richer way to understand those differences. The longer window is aimed squarely at durable goods, replenishables, and seasonal items, and it supports custom audience creation alongside more meaningful metrics like customer lifetime value and custom new-to-brand definitions.
The practical applications go well beyond calculating LTV. You can isolate high-value repeat customers and protect them. Convert high-volume supply purchasers into full customers. Build precise win-back campaigns for lapsed buyers. Identify which products actually bring in new customers, and which consumers only ever buy around Prime Day or Black Friday/Cyber Monday.
That's a very different media strategy from treating "new-to-brand" as a single bucket.
The question stops being who bought from us recently? and becomes what kind of customer are they likely to become? Acquisition is only valuable when the customer is worth acquiring.
Brands using the longer dataset to refine media buys have reported meaningful efficiency gains, including a reported 26% reduction in CPM versus a standard 12-month view. I wouldn't promise anyone that number. The point isn't the specific figure. It's that better customer definition can materially change the economics of media buying.
The Result: A Scalpel, Not a Butcher’s Knife
Conversion rate goes up. Cost per acquisition goes down. ROAS improves. When you target a more refined subset of people, you're simply more efficient with every dollar.
Here's a case study from one of our clients:
The challenge: improve ROAS on high-CPC, non-branded category keyword campaigns in Sponsored Products.
The solution: we didn't change the keywords. We changed who we were willing to pay a premium for. We duplicated the campaign, dropped bids by 75%, and applied a +300% bid modifier for a purchaser lookalike AMC audience.
The results:
| Metric | Control Campaign | AMC Lookalike Campaign | Change |
|---|---|---|---|
| ACoS | 60.25% | 23.33% | -62% |
| ROAS | $1.66 | $4.29 | +158% |
| CPC | $2.83 | $1.38 | -51% |
| CVR | 21.98% | 28.04% | +27% |
| CPA | $12.92 | $4.98 | -61% |
Same keywords. Same category. Smaller subset of searchers, chosen because they look like the ideal customer. Every efficiency metric improved.
Amazon Is Moving Toward Broader Targeting
There's an apparent contradiction here that's actually the same idea from the other direction.
We're seeing a higher impression share going to broader ad types than there used to be. Of all available impressions, share for broad match and automatic campaigns appears to be climbing while exact-match share declines. Amazon seems to be giving more priority to the more open match types.
Why? Because Amazon can lean on its behavioral insights. Not just what someone typed into the search bar, but the other signals that suggest they're in-market or ready to buy.
And through that lens, we're finding the same theme: behavioral overlays on top of that targeting are more efficient. The match type gets looser. The audience definition gets tighter. Both trends point the same way, toward targeting based not just on what someone searches for, but on who they are and what they do.
The Barrier to Entry Dropped
If you’re reading this, you should know the timing here is unusually good. Amazon has been steadily making AMC more accessible: Ads Agent now works directly with AMC and can generate analytics SQL from natural-language requests, which takes a real bite out of the technical barrier that kept AMC in the hands of a few specialist teams.
More significantly: Amazon has temporarily removed the paywall on its first-party AMC paid-feature datasets, making premium signals available at no additional cost through December 31, 2026. Advertisers can subscribe through the Paid Features page and use those signals for both measurement queries and audience creation. Third-party datasets like Experian still require a paid subscription.
Please note: these features are only free through December 31. If you opt in and don't opt out before the end-of-year deadline, you'll be charged a fee to continue access into 2027. Set a calendar reminder now, and note when you opted in.
Access Isn’t the Same as Advantage
So the data is cheaper, and the queries are easier. But making the data available doesn't make the strategy obvious.
When brands get access to more granular attribution, one thing becomes apparent fast: the picture gets messy. More paths, more touchpoints, more variables, and more possible interpretations, several of which contradict each other.
For a lot of teams, that's more paralyzing than empowering, because seeing the path is only the first step.
The moat was never access to the data. It's the ability to translate the data into execution: which audiences to build, which bids to modify, which sequence to run, and what to stop doing.
That's the work. If you'd like a second set of eyes on where AMC could tighten your targeting, we're happy to take a look. Bring your account and your questions. The consult is free.
Read more from this author: