Definition
A lookalike audience is an ad audience generated by Meta, Google, LinkedIn or similar platforms by identifying users who resemble a seed audience (existing customers, high-LTV users, converters) on dimensions the platform models.
Lookalike audiences trade audience size for similarity. A 1% lookalike on Meta finds the ~2-3 million US users most similar to your seed; a 10% lookalike finds ~25 million users somewhat similar. Smaller, tighter lookalikes outperform; larger, looser ones reach more people but convert worse.
The seed quality determines everything. A lookalike of 200 random signups produces a noisy audience. A lookalike of 5,000 highest-LTV customers produces gold. Most lookalike campaigns fail because the seed was wrong — not because the technique was flawed.
Origin
Facebook launched Lookalike Audiences in March 2013. Google Customer Match and similar audiences followed; LinkedIn introduced Lookalike Audiences in 2019. Recent privacy changes (Apple ATT, Chrome Privacy Sandbox) have weakened the underlying signals.
How it works
- Build the seed audience — high-LTV customers, recent converters, top-tier purchasers.
- Upload to the platform (Meta Custom Audience, LinkedIn Matched Audience, Google Customer Match).
- Generate the lookalike at 1% (smallest, most similar).
- Layer with demographic / interest filters if appropriate.
- Test against your standard prospecting audience for 2–4 weeks.
- Scale by widening (1%, 3%, 5%) only if narrower lookalikes plateau.
When to use it
Use when
- When you have a high-quality seed audience (1,000+ records minimum, 5,000+ ideal).
- For prospecting (top-funnel acquisition).
- When existing prospecting audiences plateau.
Skip when
- With small, noisy seed audiences. Garbage in, garbage out.
- For retargeting. Lookalikes are prospecting tools, not retargeting tools.
Key metrics
- CPA (cost per acquisition) vs prospecting baseline.
- Frequency (how often the same person sees the ad).
- Conversion rate vs interest-based audiences.
- ROAS at scale.
Examples
- 1% lookalike of our 5,000 highest-LTV customers cut CPA 38% vs interest targeting.
- Their lookalike performed worse than broad targeting because the seed was 200 trial signups, not paid customers.
- iOS 14 changes hurt lookalikes; not killed, just narrowed signal availability.
In practice at Makreate
Makreate's paid acquisition work for B2B and DTC clients tests lookalikes against broad and interest-based prospecting in every account. A recent ecommerce client had been using interest-based targeting at $42 CPA. We pulled their last 6 months of paid customers ($120 LTV+), built a 1% lookalike, and tested. Lookalike came in at $26 CPA and held at scale through $80K spend. The seed-quality discipline mattered more than the lookalike technique itself — interest targeting still works for cold prospecting; lookalikes work when the seed is sharp.
Advertising →Common mistakes
- Using small or low-quality seeds. Below 1,000 records the patterns are noise.
- Using all customers as the seed. Use only the highest-value cohort.
- Over-stacking lookalikes with demographic filters. Filters often kill the signal the lookalike found.
- Forgetting to refresh seeds. Quarterly updates prevent drift.
Frequently asked
Minimum seed audience size?
Meta requires 100; Google requires 1,000; in practice, 5,000+ produces meaningfully better lookalikes than the platform minimums.
1% vs 5% vs 10%?
1% is tightest, smallest, highest performance. 5% widens reach with some performance dilution. 10% is more like 'generally similar' than 'looks like'. Start at 1%.
How often should I refresh?
Quarterly. Customer behaviour shifts; seeds go stale. Most lookalike accounts under-refresh.