Building Effective Lookalike Audiences from Customer Data

How to build high-converting lookalike audiences on Meta from your customer list, booking data, and pixel events.

Interest targeting is a blunt instrument. You are telling Meta to find people who have clicked "like" on a Facebook page related to food or fitness or whatever your category happens to be. That is a weak signal. For an established business with real customer data, you have something much more valuable: a record of who has already paid you money.

Lookalike audiences let you give Meta that record. The algorithm then finds people on its platform who behave similarly to your existing customers. That is a fundamentally better signal than interest targeting, and the results consistently reflect it.

Why Lookalikes Beat Interest Targeting for Established Businesses

Interest targeting works reasonably well when you have no customer data at all. You are starting from nothing, so a broad behavioural signal is better than nothing. But the moment you have 1,000 or more customer records, you have outgrown interest targeting as your primary audience strategy.

The reason is signal quality. Interest targeting tells Meta someone clicked a page about coffee. A lookalike built from your email list tells Meta someone spent real money with a business like yours. These are not equivalent signals. One predicts browsing behaviour. The other predicts buying behaviour.

For established Adelaide businesses with even modest customer lists, switching from interest-based audiences to lookalike audiences typically produces a lower cost per result within the first two to three weeks of testing.

The Three Best Source Audiences

Not all source audiences produce equally strong lookalikes. The quality of the lookalike depends entirely on the quality of the source.

The first and often strongest source is your customer email list. Upload a CSV of hashed customer emails directly to Meta's Custom Audiences tool. Meta matches these against its user database. The matched audience becomes your seed. For this to work well, you need at minimum 1,000 matched users from your uploaded list. Aim for 5,000 or more if possible. The larger and more specific the seed, the tighter the lookalike.

The second source is pixel-based purchase or booking events. If your pixel has been firing a Purchase or InitiateCheckout event consistently for six months or more, you can build a lookalike from people who completed that event. This source self-selects for the highest-intent users. The limitation is volume: if your site only processes 20 purchases per month, you will not have enough data to generate a meaningful lookalike quickly.

The third source is video engagement: specifically, people who watched at least 25 percent of one of your Reels or video ads. This audience tends to be larger than a customer email list for most small businesses, and it captures people who showed genuine interest without necessarily having converted. A top-25-percent video viewer lookalike works well as a cold audience layer alongside or instead of an email list lookalike.

Minimum Size Requirements

Meta requires a minimum of 100 matched users in a source audience to create a lookalike. In practice, 100 matched users will produce a weak lookalike with very limited usefulness.

The real minimum for a useful lookalike is 1,000 matched users. Below that, the algorithm does not have enough signal to identify meaningful patterns. At 1,000 matched users, you can generate a lookalike that will outperform interest targeting. At 5,000 or more, the lookalike becomes significantly more reliable.

If your email list has 800 records, do not conclude that lookalikes are not available to you. Upload the list anyway. If Meta matches 500 or more, test a lookalike with that seed and monitor results. It may still perform reasonably. The 1,000-match guideline is a threshold for reliability, not a hard cutoff for functionality.

1% vs 2-5% Lookalikes: When to Use Each

A 1% lookalike finds the users on Meta who are most similar to your source audience. It is a small, tight audience. In Australia, a 1% lookalike will typically be in the range of 150,000 to 250,000 people depending on the source.

A 2 to 5% lookalike expands the net. You are accepting a looser match in exchange for more reach. The cost per result is usually higher but reach is greater.

Use 1% lookalikes when you are running campaigns with limited budget and want the highest-probability audience. Use 2 to 3% lookalikes when your 1% audience has fatigued or when you are scaling spend and need more reach to sustain volume.

For most Adelaide SMBs spending under $100 per day, start with 1%. Layer in 2% or 3% once the 1% audience shows signs of fatigue, which shows up as rising frequency (above 2.5 over a 7-day window) and rising CPM.

Geographic Layering for Local Businesses

Lookalike audiences are generated nationally by default in Australia. That creates a problem for a local business. Your 1% lookalike of South Australian customers will include people in Queensland and Western Australia who will never visit your venue.

Always layer geographic targeting on top of lookalike audiences for local businesses. Set your location targeting to the Adelaide metro area or a specific radius around your venue. This combination: the behavioural precision of a lookalike, narrowed by geography, is your most efficient cold audience for a local hospitality or service business.

For An Nam Quan and other Adelaide venue clients, we set lookalike audiences against a 20km radius centred on the restaurant. This cut wasted impressions significantly and brought cost per result down by reducing the proportion of the audience that had zero chance of visiting.

Refreshing Source Audiences

Lookalike audiences are static. They are generated from the source audience at the point of creation. They do not automatically update when new customers join your list or new pixel events fire.

For customer email list lookalikes, refresh the source audience and regenerate the lookalike monthly. Upload your latest customer list, create a new custom audience, and create a new lookalike from it. This keeps the model current as your customer base grows and evolves.

For pixel-based lookalikes, the source audience (people who completed Purchase or Lead events) updates continuously as your pixel fires. However, the lookalike itself is still generated at a point in time. Recreate pixel-based lookalikes every one to two months for active campaigns.

The refresh discipline matters more as your business grows. A lookalike built from your customer data 12 months ago reflects who your customers were then, not who they are now.

What Poor Source Data Looks Like

Not all email lists make useful lookalikes. The source needs to be specific to be useful. A general newsletter list that includes people who signed up years ago for a discount and have not engaged since is a weaker seed than a list of customers who made a purchase in the last 6 months.

Broad, unfiltered lists produce broad, generic lookalikes. The algorithm finds similarities across too many dimensions and the resulting audience behaves like a loose interest audience rather than a tight behavioural match.

If your source list includes a mix of genuine customers, cold leads, and people from a list you purchased or imported from a third party, segment it. Create separate custom audiences for each. Use only the genuine customer segment as your lookalike seed.

Signs your lookalike is not performing as expected: CPM is similar to or higher than broad interest targeting, click-through rate is not meaningfully better, and cost per result is not improving over a two-week test period. In that case, review the source audience quality before assuming lookalikes do not work for your business.

Building the Stack Over Time

Lookalike audiences work best as part of a layered audience strategy. Cold lookalike audiences bring in new people. Warm retargeting audiences (website visitors, video viewers, page engagers) capture people who showed interest but did not convert. The two layers work together.

Start with one strong lookalike from your best source. Test it against your current cold audience (whether that is interest targeting or a broader geographic audience). Measure cost per result over two weeks with equal budget. Let the data tell you which performs better, not your intuition.

Once you have a performing lookalike, layer in the warm retargeting audiences as a separate campaign or ad set. This is the stack that consistently outperforms single-layer audience approaches for local businesses with real customer data to work from.