Lookalike Audiences: Find Your Best Users at Scale

Our first lookalike campaign was a disaster. We uploaded our entire user base as a seed, created a 10% expansion, and wondered why performance tanked. Turns out, we'd essentially told the algorithm to find "people who might install our app"—not exactly groundbreaking targeting. The magic of lookalikes isn't in the technology; it's in the seed.

How Lookalikes Actually Work

Platforms analyze thousands of signals from your seed audience—demographics, behaviors, interests, device patterns, app usage. Then they find similar users who haven't converted yet. The algorithm's quality depends entirely on what you feed it.

The Seed Quality Hierarchy

• Tier 1 (Best): High-LTV purchasers, long-retained users
• Tier 2: Recent purchasers, engaged users
• Tier 3: All purchasers, subscribers
• Tier 4: All installers (avoid for quality campaigns)

Seed Selection Strategies

Value-Based Seeds

Create seeds based on user value, not just actions:

• Top 10% spenders (by LTV)
• Users who purchased 3+ times
• Subscribers retained 6+ months
• Users with D30 retention + purchase

Behavioral Seeds

Sometimes behavior predicts value better than transactions:

• Daily active users (DAU)
• Feature power users
• Social sharers
• Content creators within app

Expansion Size Strategy

The tradeoff: smaller expansion = more similar but limited reach. Larger expansion = more reach but diluted quality.

• 1% expansion: Highest quality, limited scale. Use for premium products.
• 1-3% expansion: Sweet spot for most apps
• 3-5% expansion: Good balance of quality and scale
• 5-10% expansion: Scale play, expect lower quality
• 10%+ expansion: Often no better than broad targeting

Layered Lookalike Strategy

Don't use single lookalikes. Build a portfolio:

1. Create multiple seeds (purchasers, retained users, high-engagers)
2. Test different expansion sizes for each
3. Exclude converting audiences from broader expansions
4. Refresh seeds monthly with new data

Common Lookalike Mistakes

• Stale seeds: Audiences from 2 years ago don't reflect current users
• Too broad: "All installers" teaches the algorithm nothing
• No exclusions: Targeting existing users wastes spend
• Single audience: One lookalike can't scale forever
• Ignoring platform nuances: Each platform's LAL works differently

Privacy Era Challenges

Lookalikes depend on matching—harder with ATT and privacy changes:

• Seed match rates have dropped 30-50%
• Use larger seeds to compensate for matching loss
• First-party data becomes more valuable
• Consider moving to broad + creative testing