Facebook Lookalike Audiences: Building the Ones That Actually Work

Lookalike audiences have been the backbone of Facebook prospecting for a decade. The mechanic is simple: you give Meta a seed audience of customers, and Meta finds users who look like them. But seed quality and audience size make the difference between a 4x ROAS lookalike and a 1.2x ROAS lookalike.

Here's how to build lookalikes that compound, not just check the box.

The seed is everything

Lookalike performance is bounded by seed quality. A great seed produces a great lookalike. A weak seed produces a noisy audience that costs you money.

What makes a great seed:

• High-intent. Customers, not pageviewers.
• Recent. Last 90 days, ideally last 30.
• Filtered for value. Top 25% by AOV or LTV, not all customers.
• At least 1,000 records. Below that, lookalike modeling is unreliable.
• Clean data. Email format consistent, customer info accurate.

A seed of "all website visitors past 180 days" produces a useless lookalike. A seed of "purchasers with AOV above $100, past 60 days" produces a great one.

Seed audiences worth building

Stack your account with these:

1. Top customer LTV (gold seed)

Top 10–25% of customers by lifetime value. Pull from your CRM or Shopify segments. Upload as Custom Audience, then build lookalike from it.

This is the highest-value seed. Lookalikes from it tend to acquire customers with similar spend patterns.

2. Recent purchasers (workhorse seed)

Past 60 or 90 days of purchasers. Larger volume than top-LTV. Good for volume scaling.

3. High-AOV purchasers

Customers who spent above your AOV threshold (e.g., 1.5x average) on first order. Useful for finding bigger-basket buyers.

4. Repeat customers

2+ purchases. Filters for the kind of customer who comes back — usually higher LTV.

5. High-engagement video viewers

Users who watched 75%+ of a key brand video. Mid-funnel signal. Good for warm prospecting.

6. Email subscribers who opened recently

Engaged subscribers from Klaviyo. Mid-intent, larger volume than purchasers.

Lookalike percentage sizing

Meta lets you choose 1% (closest match) up to 10% (broadest). The right size depends on your scale.

• 1%. Tightest match. Best for warm prospecting at small scale. Audience size = 2.3M in the US.
• 2–3%. Good balance. Most stores live here.
• 4–5%. Broader. Useful when 1–3% saturates.
• 6–10%. Often too broad to outperform interest targeting.

A common framework:

• Start with 1% lookalikes during testing.
• Layer 2–3% lookalikes as you scale.
• Test 5% lookalikes only after 1–3% are profitable and you need more volume.

Stacking lookalikes

You don't have to pick one. Stack 2–4 different lookalikes in a single ad set:

• Lookalike of top-LTV customers (1%).
• Lookalike of past-90-day purchasers (1%).
• Lookalike of repeat customers (1%).

This stacked audience:

• Combines complementary signals.
• Performs better than any single lookalike.
• Gives the algorithm more flexibility to find the highest-converting users.

Lookalikes vs Advantage+ (broad targeting)

Meta increasingly recommends Advantage+ Shopping Campaigns with broad targeting (no audience specified). For mid-size accounts ($10K+ monthly spend), broad often outperforms lookalikes because Meta's algorithm has gotten better at finding buyers without explicit audience guidance.

Reality:

• Advantage+ is dominant for accounts with strong Pixel data and sufficient scale.
• Lookalikes still work for new accounts, smaller spend, or accounts with limited Pixel volume.
• Best practice: run Advantage+ as your primary structure, with lookalike-targeted manual campaigns as the secondary structure.

Don't pick one or the other; let the algorithm tell you which works in your specific account.

Refreshing lookalikes

Lookalikes degrade. The customers you optimized for 6 months ago aren't the same as today's customers. Refresh quarterly:

• Rebuild seed audiences from latest data.
• Rebuild lookalikes (Meta will re-model).
• Test new lookalike vs existing lookalike for 7–14 days.
• Replace if new performs at parity or better.

Don't refresh weekly — lookalikes need 14+ days to perform.

Country-specific lookalikes

If you ship internationally, build separate lookalikes per country:

• Lookalike of past-90-day purchasers, US.
• Lookalike of past-90-day purchasers, UK.
• Lookalike of past-90-day purchasers, AU.

Performance varies by country because the underlying user base does. A US lookalike applied to UK won't perform as well as a UK-native lookalike.

Lookalike performance benchmarks

Rough benchmarks for e-commerce:

• 1% lookalike of top-LTV customers: 2.5–4x ROAS.
• 1% lookalike of recent purchasers: 2–3.5x ROAS.
• 3% lookalike of recent purchasers: 1.8–2.8x ROAS.
• 5% lookalike of broad audiences: 1.2–2x ROAS.

Wide range because category, AOV, and creative all impact final ROAS. Use these as directional, not absolute.

Common lookalike mistakes

• Using all-time customer list as seed. Diluted by churned, irrelevant customers.
• Building lookalikes too small (under 1,000 seed). Modeling noise overwhelms signal.
• Layering interest targeting on top of lookalikes. Defeats the purpose of letting Meta optimize. Use lookalikes alone or with Advantage+ Audience.
• Refreshing daily. Lookalikes need time to perform. Quarterly refresh is enough.
• Comparing 1% to 5% directly. They're meant for different jobs (precision vs scale).

Pairing lookalikes with creative

Lookalikes are an audience tool. Audiences don't drive performance — creative and offer do. A great creative on a 5% lookalike beats a mediocre creative on a 1% lookalike.

Don't blame the audience when the creative isn't working. Test and iterate creative first; lookalikes amplify what's already working.

A 30-day lookalike rollout

• Week 1: Build seed audiences from clean data.
• Week 2: Build lookalikes (1%, 3%) from each seed.
• Week 3: Launch test campaigns with 2-3 lookalike combinations vs Advantage+ broad.
• Week 4: Measure ROAS, CAC, scale winners, kill losers.

The right lookalike strategy is invisible — it just produces ROAS quietly month after month while the algorithm gets better at finding the customers your business actually wants.