Lookalike Audience Strategies: Build High-Value Audiences

We'd spent $200,000 on Facebook ads targeting "people interested in fitness apps."

The results were terrible. CPA was 3x our target. Retention was abysmal. LTV was so low we were losing money on every install. I was about to shut down the channel entirely when our data scientist asked a simple question:

"What if instead of telling Facebook who we want, we showed them who we already have?"

She pulled our top 500 purchasers—the users who'd spent $100+ in our app—and created a 1% lookalike audience. Same ad creative. Same budget. Same everything except the targeting.

CPA dropped 67%. Retention doubled. LTV jumped 4x. We went from losing money on every user to printing it.

That was the day I understood: lookalike audiences aren't just another targeting option. They're the difference between guessing and knowing.

What's Actually Happening Under the Hood

When you create a lookalike, the platform analyzes your seed audience's characteristics—demographics, behaviors, interests, app usage patterns—and finds new users who share similar patterns but aren't yet customers.

The Process

1. You provide a seed audience (your customer list or pixel data)
2. The platform identifies common traits among those users
3. The algorithm finds similar users who aren't in your seed
4. You target the resulting audience

Simple in concept. Transformative in practice.

The Seed Selection Framework We Built

After that first accidental success, we got systematic about seed creation. Your seed audience determines everything—choose wisely.

High-Quality Seed Options (In Order of Value)

• High-LTV customers: Users with top 20% lifetime value. The absolute best seed
• Purchasers: Anyone who made at least one purchase. Second best
• Engaged users: D30+ retained users. Still valuable
• Subscription converters: Users who converted from free to paid
• Multi-purchasers: Repeat buyers. Often better than one-time purchasers

The Size Sweet Spot

• Minimum: 100-500 users (platform dependent). Below this, the model has nothing to work with
• Optimal: 1,000-10,000 users. Enough signal, not too much noise
• Diminishing returns: Above 50,000 users. You're diluting quality with quantity

The Golden Rule: Quality Over Quantity

When forced to choose between more users or better users, always choose better:

• 1,000 whales beats 10,000 installers
• 500 paid subscribers beats 5,000 trial starters
• Recent users beat historical users (behavior changes)

The Expansion Size Decision

Most platforms let you choose how closely matched your lookalike should be. This is one of the most misunderstood levers in digital marketing.

The Trade-offs

• 1% lookalike: Highest similarity to your seed, smallest reach, best quality
• 5% lookalike: Moderate similarity, moderate reach—the balanced option
• 10% lookalike: Lower similarity, largest reach, lower quality

When to Use Each

• 1-2%: High-value campaigns, premium pricing, limited budgets. Start here
• 3-5%: Balanced approach, most common starting point once 1% is exhausted
• 6-10%: Scale campaigns, broad awareness plays, large budgets that need volume

"Start narrow, then expand. It's much easier to scale from a winning 1% lookalike to 5% than to diagnose why a 10% audience is underperforming. Earn the right to go broad."

The Advanced Playbook

Layered Lookalikes

Don't just create one lookalike—create a hierarchy:

• Primary: High-LTV purchasers at 1% expansion. Your best audience
• Secondary: All purchasers at 3% expansion. Scale option
• Tertiary: Engaged non-purchasers at 5% expansion. Volume play

Bid highest on primary, scale into secondary, use tertiary for reach campaigns.

Value-Based Lookalikes

When available, use value data to weight your seed:

• Include purchase amount or LTV score with each user
• Platform weights users by their value in the model
• Resulting lookalike skews toward high-value patterns

This is the difference between "find people like our customers" and "find people like our best customers." Massive difference.

Exclusion Strategy

Always exclude existing customers from lookalikes. Otherwise you're paying to reach people who already converted:

• All app users
• Website visitors (if applicable)
• Previous converters at any stage

Platform-Specific Nuances

Meta (Facebook/Instagram)

• Use Advantage+ lookalikes for automated expansion—often outperforms manual
• Value-based lookalikes from purchase events are extremely powerful
• Minimum 100 users in seed, but aim for 1,000+

Google/YouTube

• Similar audiences are being deprecated—adjust strategy now
• Move to optimized targeting plus first-party data signals
• Customer Match for seed audiences is the path forward

TikTok

• Custom and lookalike audiences both available
• Pixel-based and customer list options work well
• App event optimization recommended for best results

Measuring What Matters

The Metrics We Track

• Quality signals: CVR, retention, LTV by lookalike source. Not just CPI
• Scale capacity: Reach, impressions, spend capacity before saturation
• Efficiency: CPA and ROAS compared to other targeting methods
• Decay rate: Performance degradation over time as audience saturates

The Testing Framework

• Test different seed definitions against each other
• Compare expansion sizes (1% vs 3% vs 5%)
• A/B test lookalike versus interest-based targeting
• Always measure downstream metrics, not just CPI

The Mistakes That Cost Us

Using All Installers as Seed

This was our first mistake. "All installers" creates a lookalike of "people who install apps"—which is everyone. No signal, no value. Use quality signals.

Not Refreshing Seeds

User behavior evolves. The people who were valuable six months ago might not represent today's best customers. Update your seed audiences quarterly at minimum.

Ignoring Platform Guidance

Each platform has optimal seed sizes and best practices. Facebook wants 1,000+ users. TikTok has different requirements. Follow their documentation—they know their algorithms better than you do.

Lookalike audiences are the closest thing to a silver bullet in mobile advertising. But they're only as good as the seeds you plant. Invest in understanding your best users, feed that signal to the platforms, and watch the algorithms do what they do best: find you more users just like them.