Amazon DSP Lookalike Audiences: How to Scale New Customer Acquisition

The Growth Ceiling That Lookalike Audiences Break Through

Every growing Amazon brand hits the same wall eventually. Your retargeting audiences are performing well, your ASIN viewers are converting, your cart abandoners are being recovered — but those audiences are finite. There are only so many people who have already visited your product pages. At some point, you have to go find new customers, and that’s where most brands run into trouble.

Cold audience prospecting on Amazon — running in-market or lifestyle segments at scale — works, but it’s broad by definition. You’re reaching a category of shoppers, not specifically people who look like your best buyers. The ROAS from cold prospecting tends to disappoint brands expecting retargeting-level returns, and many pull back before the campaigns have a chance to prove themselves.

Lookalike audiences are the middle ground that most brands overlook. They let you move beyond your existing warm audiences without going entirely cold. Instead of targeting everyone in a category, you’re targeting people whose behavior on Amazon closely mirrors your actual customers. The result is a prospecting audience that’s genuinely more qualified than in-market segments alone — and it scales in a way that retargeting never can.

I’ve worked with brands at different stages of building lookalike strategies on Amazon DSP, and the learning curve is real but manageable. This article covers everything you need to know — how the system actually works, how to build a seed audience worth modeling, the different similarity tiers available, and the AMC-powered approach that gives you the most control.

What Amazon DSP Lookalike Audiences Actually Are

A lookalike audience starts with a seed — a group of your existing high-value customers — and uses Amazon’s machine learning to find other Amazon users whose browsing, searching, and purchasing behavior closely resembles that seed group.

Amazon’s system analyzes thousands of behavioral signals across its ecosystem to build a profile of who your best customers are and what makes them distinctive. It then finds other shoppers who share those patterns — category interests, purchase frequency, spending level, content preferences, life stage signals — and makes them available as a targetable audience in DSP.

What makes Amazon’s lookalike modeling different from Meta’s or Google’s is the signal type. Meta builds lookalikes from social graph and interest data. Google builds them from search behavior. Amazon builds them from what people actually buy — the most commercially predictive signal available in digital advertising.

The performance gap is meaningful. AMC-powered lookalike audiences consistently deliver 3 to 5 times better ROAS compared to standard in-market campaigns, with conversion rates of 23% versus 12% for untargeted prospecting.

Two Routes to Lookalike Audiences: DSP vs AMC

There are two ways to build lookalike audiences on Amazon, and they’re not equivalent. Understanding the difference changes how you approach the whole strategy.

The short version: DSP’s basic lookalike is a starting point, fine for exploration. AMC lookalikes are the real tool for brands serious about new customer acquisition. The rest of this article focuses primarily on the AMC approach, because that’s where the meaningful performance difference lies.

The Seed Audience: The Most Important Variable in Your Lookalike Strategy

Everything in a lookalike campaign traces back to the seed. A mediocre seed produces a mediocre lookalike. A precisely defined seed of genuinely high-value customers produces a lookalike that finds people with real commercial potential.

This is where most brands get it wrong — they take the easy route of seeding from ‘all purchasers in the past 90 days’ and then wonder why the lookalike doesn’t outperform their in-market campaigns. The problem is that ‘all purchasers’ includes one-time bargain buyers, coupon hunters, and people who bought once during a sale and never came back. A lookalike built from that pool finds more people who buy once and disappear — which is not growth, it’s churn.

What makes a strong seed audience

• Repeat purchasers: Customers who bought from your brand 2 or more times within 6-12 months show brand loyalty, not just category interest. Lookalikes built from repeat buyers find people with a similar pattern of intentional brand preference.
• High order value: Filtering for customers whose average order value is above a threshold (say, the top 25% of your buyer pool) finds people who spend more per transaction — a useful signal if your goal is to acquire customers with high revenue potential.
• Subscribe and Save subscribers: For consumable products, S&S subscribers represent your highest-LTV customers. Their long-term commitment to your brand is exactly the behavioral pattern you want to model.
• NTB customers who repurchased: Customers who came in as new-to-brand buyers and then made a second purchase are your best acquisition success stories. A lookalike built from them finds people likely to follow the same journey.
• High-margin SKU purchasers: If you have a specific product line with higher margins or strategic importance, building a seed from buyers of that SKU gives you a lookalike oriented toward your most profitable product.

One real example of why seed quality matters: One client discovered their ‘VIP customer’ seed included wholesale buyers who were skewing the behavioral profile. After cleaning the seed to remove those accounts, lookalike ROAS improved 40%. The signal was wrong, not the system.

The Five Similarity Tiers: Balancing Precision vs Reach

When building lookalike audiences through AMC, you can choose how closely the modeled audience should resemble your seed. Amazon offers five tiers, each representing a trade-off between similarity (and therefore conversion likelihood) and audience size (and therefore reach).

Most brands should start with the Balanced tier for their primary prospecting campaigns and run Most Similar as a separate high-precision ad group with a higher bid. This gives you two useful data points: how the lookalike performs when tightly matched, and how it performs when given room to scale.

A critical setup note: Each tier should exclude all tighter tiers to prevent overlap. If you run Most Similar and Balanced in the same campaign without exclusions, you’ll see 30-40% audience overlap — meaning the same users are being reached twice, your frequency data is distorted, and your budget is wasted.

Step-by-Step: Building a Lookalike Audience in AMC

Here’s the practical workflow for building your first AMC-powered lookalike audience, using the no-code interface that became available in October 2024.

1. Access Amazon Marketing Cloud. AMC is available to brands using Amazon DSP. Log in through your DSP account and navigate to the AMC console.
2. Open the High Value Audiences (HVA) solution. This is the no-code entry point. It lets you analyze customer spending percentiles and build audience segments without writing SQL.
3. Define your seed filters. Use the HVA interface to filter your customer base by purchase frequency, order value, time since first purchase, and SKU. For example: customers with 2 or more orders, average order value over $50, first order more than 90 days ago. This filters out recent one-time buyers and finds your genuinely loyal segment.
4. Review the audience size. AMC requires a minimum of 500 user IDs for a lookalike seed. If your filtered segment is smaller than that, broaden your criteria slightly — reduce the repeat purchase threshold to 2+ orders within 12 months rather than 2+ within 6 months, for example.
5. Create the lookalike audience. Select ‘Create Lookalike Audience’ from the seed and choose your similarity tier. Start with Balanced for your primary campaign.
6. Push the audience to DSP. AMC audiences typically appear in your DSP account within 48 hours, ready to use in campaign targeting.
7. Build your campaign with exclusions. When setting up the DSP campaign, exclude your existing customers (past purchasers), current ASIN retargeting audiences, and any other segments you’re already reaching. The lookalike is for acquisition — you don’t want to pay prospecting CPMs to reach people who are already your customers.
8. Set a 60-90 day measurement window. Lookalike campaigns need time to optimize. Week 1-4 is the learning phase. Weeks 5-8 produce first meaningful data. Weeks 9-12 are when scaling decisions become reliable.

Creative Strategy for Lookalike Audiences: Not the Same as Retargeting

This is a detail that gets overlooked far too often. Lookalike audiences are cold — these people have never interacted with your brand. Running the same creative you use for ASIN retargeting (which assumes the person already knows your product) on a lookalike audience (which assumes they don’t) is one of the most common and costly creative mistakes in DSP.

One consistent pattern across well-run lookalike campaigns: creative that leads with a specific benefit or proof point (a meaningful review excerpt, a before/after, a comparison to the category norm) outperforms generic lifestyle imagery on lookalike audiences. The behavioral match means the audience has category interest — your creative just needs to give them a reason to pick your brand over the alternative.

Measuring Lookalike Performance: The Right Metrics

Lookalike campaigns sit between retargeting (which measures direct ROAS well) and broad awareness (which measures only impressions). The right metrics reflect that middle position — you want acquisition metrics, not just efficiency metrics.

Why NTB rate matters more than ROAS for lookalike campaigns: The goal of a lookalike campaign is customer acquisition, not conversion efficiency. A lookalike campaign with 2x ROAS and 80% NTB rate is doing its job — it’s finding new buyers. A campaign with 4x ROAS and 30% NTB rate is mostly reaching people who would have bought anyway, which means it’s wasting prospecting budget on near-retargeting work.

Using AMC Lookalike Audiences Beyond DSP

One of the most significant developments in Amazon’s advertising ecosystem over the past year is that AMC audiences are no longer limited to DSP campaigns. As of late 2024, AMC audiences can be applied directly in Sponsored Ads campaigns — which opens up a whole new dimension for lookalike strategy.

How this works in Sponsored Products and Sponsored Brands

In Sponsored Products and Sponsored Brands, AMC lookalike audiences work as bid modifiers — not as exclusive targeting. This means when someone from your lookalike audience searches your targeted keyword, your bid increases by a set percentage. Everyone else sees your base bid.

In practice, this lets you apply purchase-data-backed lookalike intelligence to keyword-based campaigns. If your lookalike audience searches for your category keyword, your ad is more likely to win that auction — and you’re not paying extra unless that person actually sees and clicks the ad.

How this works in Sponsored Display

In Sponsored Display, AMC lookalike audiences work as direct targeting — you can specifically target lookalike users with Sponsored Display ads both on Amazon and (in some configurations) off-Amazon. This is a simpler and less expensive entry point to lookalike advertising than full DSP campaigns for brands not yet running programmatic.

The Most Common Lookalike Mistakes and How to Avoid Them

1. Using a poorly defined seed

‘All purchasers’ is almost never the right seed. You want behavioral specificity — repeat buyers, high-order-value customers, Subscribe and Save subscribers. The more precisely you define who your best customer is, the better the model performs.

2. Not excluding existing audiences

If you don’t exclude your current customers, retargeting audiences, and existing warm prospects from your lookalike campaign, you’ll waste prospecting-level CPMs reaching people who should be in a much cheaper retargeting pool. Exclusions are not optional — they’re foundational to lookalike efficiency.

3. Running all similarity tiers simultaneously without exclusions

Running Most Similar and Balanced in the same campaign without excluding the tighter tier from the broader one creates 30-40% audience overlap. You pay twice to reach the same person, your frequency data becomes unreliable, and optimization signals get confused. Run each tier as a separate ad group and exclude tighter tiers from broader ones.

4. Judging the campaign at 30 days

Lookalike campaigns need 8-10 weeks to reach reliable performance. The first two weeks are volume validation (are ads delivering?). Weeks 3-6 produce initial signals. Weeks 7-10 produce actionable data. Brands that pull campaigns at 30 days because ROAS looks low are almost always cutting before the algorithm has found its rhythm.

5. Using retargeting creative on a cold lookalike audience

A lookalike audience has never heard of your brand. Creative that says ‘Still thinking it over?’ or assumes product familiarity will miss entirely. Cold audiences need brand introduction messaging, not conversion messaging.

6. Not testing multiple similarity tiers

Most brands launch with a single tier and optimize from there. Running Most Similar and Balanced as separate ad groups from day one gives you meaningful comparison data — you’ll often find that Most Similar converts better per impression while Balanced drives more total NTB volume. That trade-off should inform your long-term budget allocation.

A Practical Scaling Playbook: From First Lookalike to Full Acquisition Engine

Here’s the progression that works for most established Amazon brands moving from zero lookalike activity to a full acquisition system.

Phase 1 — Build and Test (Weeks 1-8)

• Define your seed audience using AMC HVA. Start with repeat purchasers, 2+ orders, past 12 months.
• Launch Balanced tier lookalike in DSP with a conservative budget ($5,000-10,000/month minimum to generate learning).
• Use brand introduction creative — not retargeting creative.
• Exclude all existing customers and warm audiences.
• Let it run for 6-8 weeks before drawing conclusions. Monitor NTB rate and DPVR, not just ROAS.

Phase 2 — Optimize and Expand (Weeks 8-16)

• Review NTB rate and cost per NTB purchase. If NTB rate is below 50%, check exclusions — you may be reaching existing customers.
• Add Most Similar tier as a separate higher-bid ad group (exclude Balanced from this group).
• Test new creative variants — at least two per tier. The one that performs at week 12 may not be the best at week 20.
• Use AMC to measure branded search lift. If branded search volume has increased, the lookalike campaign is building awareness that Sponsored Products will capture.

Phase 3 — Cross-Campaign Activation (Month 4+)

• Apply AMC lookalike audiences as bid modifiers in Sponsored Products. Boost bids by 15-25% for users in your lookalike pool who search your category keywords.
• Activate Sponsored Display targeting for the lookalike audience as a lighter-touch complement to DSP.
• Build a second seed from your lookalike-acquired customers (those who came in via the DSP campaign and converted). A lookalike of a lookalike often finds a highly refined acquisition audience.
• Use AMC to track 6-month CLV of customers acquired via lookalike vs other channels. This is the data that justifies increasing the lookalike budget.

Final Thoughts

Lookalike audiences are one of the most powerful tools in Amazon DSP, and they’re consistently underused by brands that are still thinking about prospecting as ‘in-market audiences at scale.’ The behavioral precision that Amazon’s first-party purchase data enables — modeling your best customers rather than just your category — is a genuinely different approach to new customer acquisition.

The learning curve is real. Seed quality matters enormously. Similarity tiers need to be set up with proper exclusions. Creative has to be appropriate for a cold audience. And the measurement window needs to be long enough to see the actual impact. But brands that get these pieces right consistently report that lookalike campaigns become the highest-quality source of new customers in their Amazon advertising portfolio.

If you want to talk through what a lookalike strategy would look like for your specific category and customer base, that’s exactly the kind of planning we do.

Frequently Asked Questions