Lookalike audiences have been a cornerstone of mobile app user acquisition on Meta for years, but their role has shifted significantly in the post-ATT landscape. At RocketShip HQ, we've managed over $100M in mobile ad spend and seen firsthand how lookalikes have evolved from a targeting silver bullet into one tool in a broader toolkit. In this guide, you'll learn how to build high-quality seed audiences from your best users, choose the right lookalike percentages for your goals, and combine lookalikes with broad targeting to maximize performance in a privacy-first world. Whether you're spending $500/day or $50,000/day, getting your lookalike strategy right can mean the difference between a 20% improvement in ROAS and wasted budget.
Prerequisites: You need an active Meta Ads Manager account with a connected app, a working MMP (such as AppsFlyer or Adjust) firing events to Meta, at least 100 users in your seed audience (1,000+ is strongly recommended), and a basic understanding of privacy-first attribution and measurement in the post-ATT environment.
Page Contents
Step 1: Define Your Highest-Value User Events for Seed Audiences
The quality of your lookalike audience is entirely determined by the quality of your seed. Before touching Ads Manager, identify the specific in-app events that correlate with long-term value. This isn't always your purchase event. For subscription apps, it might be 'started_trial' or 'completed_onboarding.' For gaming apps, it could be 'reached_level_10' or 'made_second_purchase.'
Analyze your LTV curve
Pull cohort data from your MMP to identify which early events (Day 0 to Day 7) most strongly predict Day 30 or Day 90 LTV. At RocketShip HQ, we've found that second-purchase events typically produce seeds that outperform first-purchase seeds by 15-30% in downstream ROAS.
Filter for recency
Use a 30 to 90 day window for your seed audience. Users from 180+ days ago may no longer represent your current best customers, especially if your product has evolved.
Set a minimum value threshold
If you have revenue data, filter your seed to users above your median LTV. A seed of 500 high-LTV users will outperform a seed of 5,000 mixed-quality users almost every time.
We've tested this across dozens of apps: seeds built from 'repeat action' events (second purchase, 7-day retention, multiple sessions) consistently outperform seeds from single top-of-funnel events. The algorithm needs a clear signal of what 'valuable' looks like.
Step 2: Build Your Seed Audiences in Meta
Once you've identified your high-value events, create Custom Audiences in Meta Ads Manager. You have several options: app activity audiences, customer list uploads, or pixel/SDK-based event audiences. Each has trade-offs in the post-ATT world, particularly around match rates.
Use customer list uploads for maximum control
Export your high-value user list (emails, phone numbers, device IDs where available) from your MMP or CRM. Upload this directly to Meta. Match rates typically range from 30-60% depending on data quality, but the precision of your seed is worth it.
Create app activity audiences as a secondary option
In Ads Manager, go to Audiences > Create Audience > Custom Audience > App Activity. Select your app and the specific event. Note that post-ATT, these audiences may be smaller than expected due to limited tracking on iOS.
Build multiple seeds for testing
Create 2-3 different seeds: one based on purchase behavior, one based on engagement depth, and one based on your highest-value cohort. You'll test these against each other to find what the algorithm responds to best.
Post-ATT, iOS app activity audiences have shrunk by 30-50% for many advertisers. This is why customer list uploads have become more valuable. They don't rely on in-app SDK tracking and give you a more complete picture of your best users.
Step 3: Choose the Right Lookalike Percentage for Your Goals
Meta lets you create lookalikes from 1% to 10% of a country's population. The percentage you choose directly impacts the trade-off between precision and scale. There is no universally 'right' answer. It depends on your budget, your market size, and your optimization maturity.
Start with 1-2% for efficiency testing
A 1% lookalike in the US represents roughly 2.3 million people. This is your most precise audience and typically delivers the lowest CPA. Use this when you're validating creative or optimizing for a specific down-funnel event.
Scale to 3-5% when you hit frequency walls
When your 1% lookalike starts showing frequency above 2.0 in a 7-day window or CPAs begin rising 15-20%, expand to 3-5%. This gives Meta more room to find users while maintaining reasonable targeting quality.
Use 5-10% as a bridge to broad
At 5-10%, your lookalike is essentially a lightly constrained broad audience. This can be useful as a transition step before going fully broad, especially if you're not yet comfortable removing all targeting.
Test stacked lookalikes
Create a combined audience of 1% purchase lookalike + 1% engagement lookalike + 1% trial-start lookalike. This 'stacked' approach gives the algorithm multiple signals of quality while expanding your addressable pool.
Here's a pattern we've seen repeatedly: as you scale spend past $5,000/day on a single lookalike, the performance gap between 1% and 5% narrows significantly. At high spend levels, the algorithm exhausts the 1% pool quickly and starts reaching into less optimal segments anyway. Test the wider percentages proactively rather than waiting for performance to degrade.
Step 4: Structure Your Campaign to Let the Algorithm Learn
Campaign structure matters as much as audience selection. Meta's algorithm needs sufficient conversion volume per ad set to optimize effectively. Post-ATT research across 15+ accounts shows that AEO campaigns need a minimum of 128 installs per day per campaign to exit the learning phase reliably. Fragmented structures with too many ad sets will starve the algorithm of data.
Consolidate ad sets aggressively
Rather than running separate ad sets for 1%, 3%, and 5% lookalikes simultaneously, test them sequentially or use a single consolidated ad set. Meta's Advantage+ audience tools now allow you to input a lookalike as a 'suggestion' while still letting the algorithm go broader if it finds better users.
Choose the right optimization event
For most apps, optimizing for installs (MAI) gives you more data volume, while AEO or Value Optimization (VO) gives you higher quality. If your seed is strong and your daily budget supports 50+ conversion events, go with AEO. Otherwise, start with MAI.
Set budgets to ensure learning phase exit
Each ad set needs approximately 50 optimization events per week to exit learning. If your target CPA is $10, that means a minimum budget of roughly $70/day per ad set. Plan accordingly.
One of the most common mistakes we see at RocketShip HQ: advertisers running 8-10 ad sets at $50/day each when they'd be better served by 2-3 ad sets at $150-200/day. Concentration of data is everything for algorithmic optimization.
Step 5: Test Lookalikes Against Broad Targeting Systematically
Here's the uncomfortable truth that many UA managers resist: in the post-ATT world, broad targeting (no audience selection at all) often matches or outperforms lookalike audiences, especially at scale. Hypercasual UA strategies have demonstrated that broad targeting with large audience pools can drive CPIs as low as $0.10-$0.12 when paired with strong creative. This principle increasingly applies to non-gaming apps too.
Run a controlled A/B test
Set up two campaigns with identical budgets, creatives, and optimization events. One targets your best lookalike, the other targets broad (age and geo only). Run for at least 7 days with sufficient budget to collect 200+ conversions per campaign.
Measure on blended CPA, not campaign-level CPA
Post-ATT attribution is noisy at the campaign level. Look at your overall blended CPA across both campaigns. If adding broad doesn't raise your blended CPA, that's a strong signal to allocate more budget to broad.
Document the crossover point
For most apps, there's a specific daily spend level where lookalikes stop outperforming broad. Find this number for your app. It's typically between $2,000 and $10,000/day depending on your geo and vertical.
In our experience managing campaigns across multiple verticals, approximately 60% of the time broad targeting matches lookalike performance when the creative is strong. The remaining 40% where lookalikes win tend to be niche apps with very specific user profiles (think: meditation apps for new mothers, or financial planning tools for retirees).
Step 6: Combine Lookalikes and Broad Using Advantage+ Audience
Meta's Advantage+ Audience (formerly detailed targeting expansion) lets you use your lookalike as an 'audience suggestion' rather than a hard constraint. This is the best of both worlds: you give the algorithm a starting signal of who your best users look like, while allowing it to go beyond that audience when it finds promising users outside the lookalike.
Set your lookalike as the audience suggestion
In the ad set, under Advantage+ Audience, add your 1% lookalike as the 'audience suggestion.' This tells Meta's algorithm to prioritize users similar to your seed but not restrict delivery to only those users.
Layer in age and geo constraints only
Remove interest-based and behavioral targeting layers. Keep only the constraints that make business sense: age minimums (for age-gated apps), country targeting, and language if relevant. Let the algorithm handle the rest.
Monitor audience expansion metrics
Check the 'Audience Segment' breakdown in your reporting to see how much of your spend is going to the suggested audience versus the expanded audience. If expanded is performing within 20% of your suggested audience CPA, that's healthy expansion.
We've seen Advantage+ Audience deliver 10-25% more volume at equivalent CPAs compared to hard-targeted lookalikes across multiple RocketShip HQ client accounts. The key is having strong creative. When your creative clearly communicates your value proposition, the algorithm doesn't need narrow targeting to find the right users.
Step 7: Refresh and Iterate Your Seed Audiences Regularly
Seed audiences decay over time. User behavior changes, your product evolves, and the Meta algorithm benefits from fresh signals. Treat your seed audiences as living assets that need regular maintenance, not set-and-forget targeting inputs.
Refresh seeds every 30-60 days
Update your customer lists and app activity audiences monthly. Remove users older than 90 days and add new high-value users. This keeps your seed reflective of your current best customers.
Create event-specific seasonal seeds
If your app has seasonal peaks (holiday shopping, New Year's resolutions, back-to-school), create dedicated seeds from users acquired during those periods. They'll be more relevant when the next seasonal window opens.
Test new seed definitions as your product evolves
If you launch a new feature or monetization model, the definition of 'high-value user' may change. Re-run your LTV correlation analysis quarterly to ensure your seed events still predict long-term value.
A simple but powerful practice: maintain a spreadsheet tracking each seed audience, its creation date, size, match rate, and the performance of its associated lookalike. This makes it easy to spot when a seed needs refreshing and identify which seed definitions work best.
Step 8: Decide When to Go Fully Broad and Retire Lookalikes
At a certain scale and creative maturity, lookalikes become training wheels. Concentrated data on fewer targeting approaches often outperforms spreading thin across multiple lookalike variations. The decision to go fully broad should be data-driven, not ideological. Here's how to know when you're ready.
Check your broad vs. lookalike performance gap
If your broad campaigns are within 10% of your lookalike campaigns on CPA over a 14-day window, you likely don't need lookalikes anymore. The algorithm has enough signal from your conversion data and creative to find users on its own.
Evaluate your daily conversion volume
If you're generating 200+ optimization events per day per campaign, Meta's algorithm has more than enough data to optimize without audience constraints. Lookalikes are most valuable when you're in the 20-100 daily conversions range.
Use lookalikes as a fallback, not a default
Even after going broad, keep your best-performing seed audiences ready. If you enter a new market, launch a new product line, or see broad performance degrade, lookalikes give you a quick lever to pull while you diagnose the issue.
The apps spending $50K+ per day on Meta almost universally run broad targeting as their primary approach, using lookalikes only for specific testing or new market entry. If you aspire to that scale, practice broad targeting early. Research confirms that self-attributing networks like Meta leverage historical purchase behavior far more powerfully than any audience-based targeting signal you can provide.
Common Mistakes to Avoid
- Using low-quality seed audiences: Building lookalikes from all installers instead of high-LTV users is the single most common mistake. Your seed should represent the users you want MORE of, not just everyone who downloaded your app. Filter ruthlessly for quality.
- Running too many lookalike ad sets simultaneously: Splitting your budget across 1%, 2%, 3%, 5%, and 10% lookalikes starves each ad set of conversion data. The algorithm can't optimize with 10 conversions per day. Consolidate into 2-3 ad sets maximum and test percentages sequentially.
- Never testing broad against lookalikes: Many UA managers assume lookalikes are always better and never run a clean test. Post-ATT, broad targeting with strong creative frequently matches or beats lookalike performance. You won't know until you test, and the results may surprise you.
- Ignoring seed audience decay: A seed built from users acquired 12 months ago represents a different user profile than your current best customers. Stale seeds lead to stale lookalikes. Refresh every 30-60 days as a non-negotiable habit.
- Over-restricting lookalikes with layered targeting: Adding interest targeting, behavioral filters, and narrow demographics on top of a lookalike audience shrinks your pool dramatically and prevents the algorithm from exploring. Use lookalikes OR detailed targeting, rarely both together.
Lookalike audiences remain a valuable tool in the mobile UA toolkit, but their role has shifted from primary targeting mechanism to algorithmic training signal. Start with the highest-quality seed audiences you can build, test systematically across lookalike percentages, and always benchmark against broad targeting. In the post-ATT world, the combination of strong creative and concentrated conversion data matters more than precise audience targeting. At RocketShip HQ, we've seen the most successful apps treat lookalikes as one phase in a maturity curve: essential when you're starting out and generating fewer than 100 daily conversions, progressively less critical as your creative and conversion volume improve. Master the fundamentals here, then let your data tell you when it's time to go broader.
Looking to scale your mobile app growth with performance creative that delivers results? Talk to RocketShip HQ to learn how our frameworks can work for your app.
Not ready yet? Get strategies and tips from the leading edge of mobile growth in a generative AI world: subscribe to our newsletter.
Related Reading
