Lookalike Segments: When the Platform's Clone of Your Best Buyers Works and When It Doesn't
The platform cloned your best buyers and found you a million more of them. Half of them will never buy anything.
A lookalike segment is the platform's clone of your best buyers, built from a seed audience of your customers. It works when the seed is large, accurate, and made of your best buyers. It fails when the seed is small, stale, or made of everyone.
The seed is the segment
A lookalike is only as good as its seed. The platform builds the lookalike from the people in the seed, and if the seed is your best buyers, the lookalike will resemble your best buyers. If the seed is everyone who ever bought, the lookalike will resemble everyone who ever bought, including the one-time buyers who will never buy again.
The seed should be your best buyers, not all buyers. The customers who bought twice and spent the most, the customers with the highest lifetime value, the customers who bought in the last six months. Those are the people you want more of, and the lookalike should be built from them.
Meta recommends a seed of 1,000 to 5,000 people. A seed under 1,000 puts the lookalike in low-data mode, which means a smaller audience, higher CPMs, and less accurate targeting. The platform does not have enough data to build a good clone, and the clone it builds is a rough one. If your seed is under 1,000, the lookalike will work, but it will work poorly, and the money spent on it will be spent on a rough clone.
The match rate and the percentage
About 20 to 30 percent of a lookalike audience will not match to a real person, because the platform cannot find them or cannot confirm them. The audience you build is smaller than the audience you asked for, and the match rate is part of the cost. A lookalike of 1 million people is really a lookalike of 700,000 to 800,000 people, and the ones who did not match are the ones you cannot target.
The percentage is the other dial. A 1 percent lookalike is the closest 1 percent of the population to your seed, which is the most similar and the smallest. A larger percentage, such as 5 or 10 percent, is less similar and larger. The trade is between similarity and size. A 1 percent lookalike will convert better, because the people in it are more like your best buyers. A 10 percent lookalike will reach more people, because it is ten times larger, but the people in it are less like your best buyers.
The right percentage depends on the size of the seed and the size of the market. A large seed in a large market can support a 1 percent lookalike that is still big enough to spend against. A small seed in a small market needs a larger percentage to reach enough people. The percentage is not a setting. It is a decision about how much similarity you are willing to trade for reach.
Value-based lookalikes
A value-based lookalike uses a CLV column, which tells the platform not just who your customers are, but how much they are worth. The platform builds the lookalike from the highest-value customers, not from all customers, and the audience is more valuable as a result. A customer who spent $500 is worth more than a customer who spent $50, and the platform should know that.
The CLV export has minimums. Klaviyo needs at least 500 customers with orders, 180 days of history, and some customers with 3 or more orders. Below that, the model does not have enough data to be reliable. The export is worth the wait, because a value-based lookalike built from a thin export is worse than a regular lookalike built from a good seed.
The value-based lookalike is the version that works best, because it is the version that tells the platform what you actually want: more customers like the ones who spend the most, not more customers like the ones who spend the least. The platform will build what you ask for, and the CLV column is how you ask for the right thing.
When the clone fails
A lookalike fails when the seed is wrong. A seed of all buyers, including the one-time buyers, will produce a lookalike of people who buy once and never again. A seed that is stale, built from customers who bought two years ago and have not bought since, will produce a lookalike of people who resembled your customers two years ago. A seed that is too small will produce a lookalike that is a rough clone, and the rough clone will not convert.
The lookalike also fails when it is not refreshed. A seed that was built six months ago and never updated is a seed that reflects the customers you had six months ago, not the customers you have now. The sync should run weekly, in full, so the seed always reflects the current best buyers.
The test is the conversion rate. A lookalike that converts at the same rate as a cold audience is not a lookalike. It is a cold audience with a different name. A lookalike that converts better is working, and the improvement is the measure of how good the seed was. If the lookalike is not converting, the seed is the problem, not the platform.
Keep reading
Related guides on this site: how RFM segmentation scores recency, frequency, and monetary value lifecycle segments from new to lapsed to lost how to audit the segments nobody activates More from Creative that adapts to every customer segment.
Frequently asked questions
What is a lookalike segment?
A lookalike is an audience of people who resemble a seed audience of your customers, built by the platform. The platform finds people who share characteristics with the seed, and it targets them with your ads.
How large should the seed audience be?
Meta recommends a seed of 1,000 to 5,000 people. A seed under 1,000 puts the lookalike in low-data mode, which means a smaller audience, higher CPMs, and less accurate targeting.
What does a 1 percent lookalike mean?
A 1 percent lookalike is the closest 1 percent of the population to your seed, which is the most similar and the smallest. A larger percentage, such as 5 or 10 percent, is less similar and larger. The trade is between similarity and size.
How much of a lookalike audience actually matches?
About 20 to 30 percent of a lookalike audience will not match to a real person, because the platform cannot find them or cannot confirm them. The audience you build is smaller than the audience you asked for, and the match rate is part of the cost.
What is a value-based lookalike?
A value-based lookalike uses a CLV column, which tells the platform not just who your customers are, but how much they are worth. The platform builds the lookalike from the highest-value customers, not from all customers, and the audience is more valuable as a result.
When does a lookalike work best?
When the seed is your best buyers, not all buyers. A seed of everyone who ever bought includes the one-time buyers who will never buy again, and the lookalike will resemble them too. A seed of your best buyers, the ones who bought twice and spent the most, will resemble the people who buy twice and spend the most.