A lookalike audience is a targeting option on advertising platforms like Meta, Google, and LinkedIn that takes a source audience you provide (typically your customers, leads, or high-value converters) and finds new people who share similar characteristics. The platform analyses behavioural patterns, demographics, and interests from your seed list, then builds a much larger audience of people who statistically resemble them. It is not a guarantee of intent, but it is one of the most reliable ways to prospect beyond your existing reach.
Prospecting without lookalikes often means broad targeting and wasted spend. When you give the algorithm a strong signal of who your best customers actually are, you compress the learning phase and reduce your cost per acquisition. The difference between a well-built lookalike and a lazy interest-based audience can be 30-50% lower CPAs in the first few weeks alone. Get this wrong and you are essentially paying the platform to guess.
You upload or select a source audience, usually a customer list, website visitors, or people who completed a specific conversion event. The platform then profiles that group against its broader user base and scores millions of users on similarity. You choose a percentage range (typically 1-10%) that determines how closely the new audience must match your source; 1% is the closest match but smallest pool, 10% is broader but less precise. The algorithm refreshes the audience over time as your source data and platform signals evolve.
The most common mistake is building lookalikes from low-quality seed data. If your source audience is everyone who visited your homepage, the platform has almost nothing useful to model from. Use purchasers, repeat buyers, or high-LTV segments instead. Another frequent error is jumping straight to broad percentages (5-10%) before testing tight ones (1-2%) and understanding what works. Finally, too many advertisers set up a lookalike once and never update the source list, which means the model drifts further from your actual customer profile every month.
Straight answers to the questions we hear most often about lookalike audiences and how to use them properly.
Meta recommends a minimum of 100 people, but that is the floor, not the goal. In practice, source audiences of 1,000 to 5,000 high-quality records tend to produce the strongest results because the platform has enough data points to identify meaningful patterns rather than noise.
Start with 1%. It is the closest match to your source audience and typically delivers the lowest cost per result. Once you have proven performance at 1%, test 2-3% to scale. Jumping straight to 5% or 10% dilutes the signal before you know what signal actually works.
Both work. Email lists are matched against platform user profiles (match rates vary, typically 40-70% on Meta). Pixel-based sources like purchasers or add-to-cart events can be stronger because the platform already has rich behavioural data attached. The best approach is to test both and compare performance.
If you are using a customer list as your source, update it at least quarterly. If you are using a pixel-based event, the platform refreshes automatically, but you should still review performance monthly. Stale source data means the algorithm is modelling yesterday's customer, not the one you actually want tomorrow.
We audit your existing audience strategy, identify the highest-value seed segments from your data, and build a structured testing framework so you are not guessing at percentages or source lists. More importantly, we teach your team to do this independently. The goal is for you to manage and optimise lookalike campaigns with confidence long after our engagement ends.
