AMC Custom Audiences: 8 Recipes and How to Use Them

AMC custom audiences let you target shoppers by what they actually did with your ads and products, not by a segment Amazon defined for everyone.

This guide is for brands and agencies already running Sponsored Ads or DSP. It covers what the audiences are in 2026, where they run, Amazon's size and refresh rules, eight recipes we build in real accounts, a worked example and how to prove the audience earned its budget.

What Are AMC Custom Audiences?

An AMC custom audience is a group of shoppers you define with a query in Amazon Marketing Cloud (AMC), Amazon's clean room of event-level ad and shopping data. Amazon sends the list straight to your ad accounts; you never see the people in it. Our Amazon advertising reports guide covers AMC's analytics side.

The definition can use any signal in your AMC instance. Amazon lists DSP and sponsored ads traffic, ad-attributed conversion events, your own first-party tables and, for DSP audiences, select paid feature tables.

That is what makes them different from Amazon's standard audiences. An in-market segment is the same for every brand that buys it. "Added my 32 oz bottle to cart in the last 14 days and didn't buy" exists only for you.

AMC has offered audiences for DSP since May 2023. Three things changed since, and most guides on page one still describe the old version.

What Changed in 2025 and 2026

1. Sponsored Ads Advertisers Got Direct Access

On September 16, 2025, Amazon opened AMC to sponsored ads advertisers inside the ads console, under Measurement & Reporting. No partner or account manager has to provision it any more.

The same release added no-code templates and AI assistance. For a seller with no DSP spend, this is the first time AMC audiences have been a few clicks away.

2. SQL Became Optional

The AI audience generator (July 2025) writes audience SQL from a plain-English prompt. At unBoxed 2025, Ads Agent in AMC extended that to analytics queries, opened from "Ask AI for AMC help".

In March 2026 Amazon added right-click edit, explain and fix in the query editor, with a side-by-side diff before you accept a change. You still need to read the SQL. The AI writes it faster; it doesn't know your replenishment cycle.

3. Longer Lookbacks

AMC's ad traffic lookback grew from 13 to 25 months, in open beta in the US and Canada from November 2025 and in more countries from Q1 2026.

The paid Amazon Retail Purchases dataset goes further, with five years of store purchases. It costs $500 a month for measurement but is free for building audiences, which Amazon now supports in all AMC marketplaces, though UK and EU purchases are left out of the audiences.

At unBoxed on September 29, 2026, the ads console and DSP became one platform, Amazon Ads Agent. Amazon also announced an AMC Expert Tool that lets outside AI agents generate AMC queries. Audiences themselves work the same way.

Rule-Based vs Lookalike Audiences

Every AMC audience is one of two types. The choice decides who ends up in it.

A rule-based audience is precise. Your query returns a list of user IDs, and that list is the audience.

A lookalike starts from the same kind of query, used as a seed. You then pick how far Amazon should reach, from Most similar to Most broad. Amazon's lookalike guide estimates 500,000 to 1 million people at Most similar and 3 to 10 million at Most broad in Europe and the Americas.

Three Ways to Build Either Type

1. SQL in the query editor. Full control. The final SELECT must return only `user_id`, from the tables ending `_for_audiences`.

2. No-code templates. The use case library, formerly the instructional query library, holds Amazon-written queries. Filter by "No code" for template audiences that only ask for campaigns or ASINs. The high-value audiences solution is the point-and-click version for top spenders.

3. AI generation. "Generate SQL with AI" in the editor, or Ads Agent chat. Check the date range and the exclusions before you submit.

Where You Can Use AMC Audiences

AMC audiences work in Amazon DSP and in sponsored ads. Since unBoxed 2024, the sponsored ads side has three uses.

Amazon DSP campaigns are now Display, Video and Audio (DVA) campaigns, and their old line items are ad groups. Our Amazon DSP targeting guide covers how AMC audiences sit next to the other DSP options.

Bid boosting is the use most sellers miss. It doesn't create new traffic; it raises your bid on searches you already target when the shopper is in your audience. Our guide to audience bid boosting covers the setup.

When you create an audience you pick one advertiser: a DSP advertiser or a sponsored ads account. An audience built for sponsored ads is available across Sponsored Products, Sponsored Brands and display ads.

One limit to plan around: paid feature tables such as Retail Purchases only feed DSP audiences. A five-year lapsed-buyer audience can't become a Sponsored Products bid boost.

Size Minimums, Refresh Rates and Other Rules

These rules come from Amazon's rule-based audience guide and support pages. They cause most failed audiences we see.

• Size. Rule-based audiences need at least 2,000 people, and Amazon advises aiming for 2,500. Lookalike seeds need 500 to 500,000. Count first with an analytics query using `COUNT(DISTINCT user_id)`.

• Refresh. Audiences refresh daily by default, and you can set an interval of up to 21 days. Each refresh reruns the query and overwrites the list.

• Unused audiences slow down. A DSP audience not attached to an ad group drops to a two-day refresh within 14 days, seven days within 30, and stops within 100.

• Rolling dates. Set "Auto adjust date" to Yes, or the window stays fixed and the refresh slows to weekly after 30 days. Queries running longer than four hours also drop to a two-day refresh.

• Timing. A new audience can take 48 hours to reach your account. Build event audiences at least a week early.

• No edits. Once an audience is live you can't change it, only copy it. Failed audiences can be edited.

• No export. You never see the user list. It goes to DSP or sponsored ads, and nowhere else.

Our rule of thumb: if a 14-day window gives you fewer than 2,500 people, widen it to 30 days before you loosen the logic. A looser definition changes who is in the audience; a longer window only changes how many.

How to Build an AMC Custom Audience

1. Start With the Decision, Not the Data

Write down what you will do differently for this audience: exclude it, bid more, show a different creative. If nothing changes, don't build it.

2. Write the Logic in One Sentence

"Shoppers who added any of our ASINs to cart in the last 14 days and didn't buy any of them." Include, exclude, window. Every recipe below follows that shape.

3. Count It

Run the same logic as a measurement query with `COUNT(DISTINCT user_id)`. Under 2,500, widen the window or add ASINs.

4. Create It

In AMC, open the Query editor, click +, choose Audience, then set the advertiser, country, type, name and "Auto adjust date". Paste or generate the query, set the date range and create it.

5. Activate It

Attach it to a DSP ad group, a display ads campaign or a bid boost within two weeks, or its refresh starts to slow.

6. Hold Some Out

Keep 10% to 20% of the audience unexposed, so you can measure what it caused. The measurement section shows how.

8 AMC Audience Recipes With the Logic Spelled Out

These are the eight we build most. The windows are our starting points, not Amazon rules; adjust them to your category's purchase cycle.

AMC's conversion tables hold ad-attributed events: carts and purchases from shoppers who saw or clicked your ads. Brands with little ad reach will find these audiences smaller than their real customer base.

1. Cart Abandoners

• Include: added any of your ASINs to cart (`shoppingCart` events) in the last 14 days.

• Exclude: purchased any of those ASINs in the same 14 days.

• Use it: DSP or display ads retargeting, plus a Sponsored Products bid boost on your branded and category terms.

This is the highest-intent audience AMC can build. Keep the window short; a cart from six weeks ago is a weak signal. Our AMC audience playbook for Prime Day uses the same build for deal events.

2. High-Value Purchasers

• Include: purchasers in the top 20% by product sales over 12 months.

• Exclude: one-time buyers of a single low-priced ASIN.

• Use it: as a lookalike seed for prospecting, and as an exclusion from prospecting lines.

The rule-based version protects your margin. The lookalike version is where the growth comes from.

3. Lapsed Buyers

• Include: bought between 60 and 180 days ago.

• Exclude: bought in the last 60 days.

• Use it: DSP or display ads with a "time to restock" message.

Set the windows from your replenishment cycle. For a 30-day product, lapsed starts around day 45; for a 90-day product, around day 120.

4. Cross-Sell

• Include: bought product A in the last 90 days.

• Exclude: ever bought product B in the lookback.

• Use it: DSP or display ads featuring B, and a bid boost on B's keywords.

Pick the pair from AMC's purchase overlap analysis, not from your catalogue. The pairs shoppers actually buy together are often not the ones you'd guess.

5. Exposed but Not Converted

• Include: saw your streaming TV, online video or Sponsored Brands video ads in the last 30 days.

• Exclude: viewed a detail page or purchased in the same window.

• Use it: display ads and a Sponsored Products bid boost to turn awareness into a visit.

This is the audience Amazon itself highlights for bid boosting. It's how upper-funnel spend starts paying back in search.

6. Category Conquest

• Include: shoppers who clicked your ads on competitor keywords or competitor product pages, or who were reached by DSP ad groups targeting your category's in-market segment, in the last 30 days.

• Exclude: anyone who bought your brand in 12 months.

• Use it: DSP or display ads with your comparison message.

AMC can't show you who bought a competitor. It can show who engaged with your conquest ads, so run category and competitor targeting first, then build this from the clicks.

7. Subscribe & Save Candidates

• Include: bought the same replenishable ASIN at least twice in 180 days.

• Exclude: shoppers with a first Subscribe & Save order, which AMC counts among its conversion events.

• Use it: DSP or display ads that sell the subscription discount.

A shopper who has bought twice has already shown the habit. Converting them to a subscription makes the next orders automatic, with no further ad spend.

8. New-to-Brand Lookalikes

• Include (seed): new-to-brand purchasers in the last 90 to 180 days.

• Exclude: all purchasers in 12 months, applied on the ad group.

• Use it: DSP prospecting at Balanced or Similar reach.

Seed it with new customers, not your loyal base, and run it as the prospecting layer of a full-funnel DSP plan. That teaches the model what a *new* buyer looks like, which is the person you want more of. Our new-to-brand metrics guide explains how Amazon defines one.

Worked Example: Cart Abandoners for a Water Bottle Brand

Here's the arithmetic on recipe 1 for an illustrative brand selling a $42 insulated bottle.

1. Size it. 6,000 shoppers added the bottle to cart in 14 days and 1,900 bought, so the audience is 4,100. That clears Amazon's 2,500 guidance.

2. Split it. Hold out 20% (820 people) and target 3,280 with display ads plus a 50% bid boost on branded search.

3. Spend. $2,400 over 14 days.

4. Read the attributed result. 9.0% of the targeted group bought: 295 orders, $12,390 in sales, an attributed ROAS of 5.2.

5. Read the holdout. 6.0% of the holdout bought anyway. The ads added 3.0 points.

6. Find the incremental value. 3,280 × 3.0% = 98 extra buyers, worth $4,133 at $42. Incremental ROAS is $4,133 ÷ $2,400 = 1.7, or $24 per extra buyer.

At a 35% margin, $4,133 of extra sales earns $1,447 of margin against $2,400 of spend. On attributed ROAS this audience looks like the best in the account; on incremental numbers it loses money.

The fix is usually cheaper delivery, not a different audience. A lower frequency cap and the bid boost alone (no display ads) cut the spend while keeping most of the lift. Our frequency capping guide has the caps we start from.

How to Measure the Uplift From AMC Audiences

1. Build the Holdout Into the Audience

In our builds we split the audience query into two audiences with a fixed rule on the user ID, such as a regular-expression match on its last character. Activate one and leave the other unused, so both fill the same way.

2. Compare Conversion Rates in AMC

After two to four weeks, run an analytics query comparing purchase rates for the two groups. Incremental share is the lift divided by the exposed rate; in the example, 3.0 ÷ 9.0 = 33%.

3. Check Bid Boosts Actually Fired

Amazon records when a bid boost applied in the `matched_behavior_segment_ids` column of AMC's sponsored ads traffic table. If the audience rarely matches your keywords, the boost isn't doing anything.

4. Judge Each Recipe on Its Job

Attributed numbers still matter for daily pacing. Use our Amazon attribution guide for the windows, and judge the quarter on TACOS, not ad ROAS alone.

Mistakes We See With AMC Audiences

• Building audiences nobody attaches. After 100 days unused they stop refreshing, and the ad group silently serves to nobody.

• Static date ranges. "Auto adjust date: No" freezes the window, so your "last 14 days" audience is soon last spring.

• No purchaser exclusions on prospecting. A lookalike line without one buys your existing customers at prospecting prices. Our guide to excluding audiences in DSP has the builds.

• Too many overlapping audiences. Five retargeting audiences built from the same carts compete for one shopper and inflate frequency.

• Trusting AI-generated SQL unread. Check the window, the ASIN list and the exclusion. The generator is good, but it doesn't know your catalogue.

• Judging on attributed ROAS. Retargeting audiences claim sales that would have happened anyway. The holdout tells you how many.

How We Build AMC Audiences in Adbrew

Everything above works in the AMC console. The work that doesn't scale is the rest: counting each audience, keeping windows rolling, attaching them before refresh slows, and measuring each against a holdout across many brands.

Adbrew is an agentic retail media management platform for Amazon and Walmart, and AMC sits in it next to Sponsored Ads, DSP and Marketing Stream:

• AMC without SQL. Adbrew can create your AMC instance or connect yours, then build cart abandoners, repeat viewers, high-value buyers and new-to-brand audiences from templates. Enterprise plans add custom queries.

• Activation in one place. Audiences go to DSP ad groups and to Sponsored Products and Sponsored Brands bid boosts, managed with the same rules you set for bids and budgets. Amazon DSP runs through Adbrew's seat with no minimum spend.

• Pre-built AMC reports. Path to conversion, overlap and new-to-brand reports refresh daily, so you pick cross-sell pairs and windows from your own data.

• Adbrew Intelligence. Ask "build a lapsed-buyer audience for my 60-day SKUs and compare it with last quarter's holdout", and the agent drafts it from your account. Through Adbrew MCP the same work runs from Claude or ChatGPT, and changes apply only after you approve them.

K2 Health used AMC through Adbrew and lifted conversions 20.28%. Brands on Adbrew cut ACOS by 46% on average.

Build the Audience Your Competitors Can't Buy

Amazon's standard segments are available to every brand in your category. The audiences in this guide exist only in your AMC instance, and that's where the edge is.

Want cart abandoners, lapsed buyers and new-to-brand lookalikes running on your account without writing SQL? Book a demo with our team and we'll build the first three with you. For the paid datasets behind recipes 2 and 3, see AMC paid features, and for creative and bidding on retargeting lines, our Amazon DSP retargeting guide.