You have three hundred signed clients from the last ten years. A spreadsheet with names, phone numbers and emails. Someone tells you: "Give it to Meta, it will find people who look like them." The idea is appealing. It genuinely worked, at one point. Today, for a service business working within a 25-mile radius, it deserves a colder look.
This article does not show you where to click. It helps you decide whether a lookalike audience is worth building, from which list, at which percentage, and when it simply wastes your time.
What a lookalike audience really is
A lookalike audience starts from a source audience you provide: a client list, website visitors, people who filled in a form. Meta analyzes those profiles, then searches a given country for the users whose behavior most resembles them.
Two things to keep in mind:
- Similarity is calculated at country level, not at the level of your service area. Meta looks for the closest people across the whole country, then your location targeting cuts that population down.
- Meta is not looking for "homeowners who want a conservatory." It looks for people who behave on Facebook and Instagram like the people on your list. If your list is messy, so is the resemblance.
Everything else follows from these two points.
The source: signed clients, not leads
This is the decision that matters most, and most advertisers get it wrong. They upload the longest list they have at hand, often the export of their Meta lead forms.
The problem is simple. A lead list contains the curious, the fake numbers, the renters who daydreamed about a pool and the people who clicked by mistake. Asking Meta to find similar profiles means asking it to find more of them. You industrialize exactly what you wanted to filter out.
Here are the sources ranked from most to least reliable:
| Source | What Meta learns | Verdict |
|---|---|---|
| Signed clients, with project value | Who buys, and who buys big | The best, if the list is long enough |
| Signed clients, without value | Who buys | Very good |
| Kept appointments or quotes sent | Who goes all the way through the process | Good, especially when clients are too few |
| Lead form submissions | Who fills in a form | Avoid as a main source |
| Site visitors, page followers | Who is interested | Too broad for a high-ticket service |
If your client file includes the value of each project, Meta offers a value-based lookalike. It weights the resemblance toward your most profitable clients. For a conservatory installer whose projects range from 8,000 to 45,000 euros, that nuance changes the nature of the signal you send.
How many names do you need?
Meta accepts a source from 100 people in a single country. That is a technical minimum, not a useful threshold. With 100 clients, some of whom will have changed email or never had a Facebook account, the number of profiles actually matched can drop low. The resemblance then rests on too few cases.
In practice, a source of a few hundred matched clients starts to produce a usable signal. Meta itself states that a source between 1,000 and 50,000 people works better. Most local businesses will never reach that volume with clients alone. That is normal, and it is one reason the rest of this article matters.
Preparing the list
Three rules before uploading anything:
- Clean: remove duplicates, out-of-area clients and cancelled projects.
- Enrich: email, phone in international format, first name, last name, city, postcode. The more fields Meta has, the more profiles it matches.
- Check the legal basis: uploading a client file to Meta is processing personal data. The data is hashed before upload, but the duty to inform your clients and to have a legal basis stays with you. Check your privacy policy before uploading anything.
The percentage: 1%, 5% or 10%?
The percentage tells Meta what share of the country's population to keep, starting from the profiles closest to your source. 1% is the most similar, 10% widens considerably.
For a national business, the classic rule is to start at 1% and widen if volume runs short. For a local business, the math changes completely.
One percent of a country's users means several hundred thousand people spread across the whole territory. Then apply a 20-mile radius around a mid-sized town: only a small fraction of that audience remains, sometimes a few thousand people. Too few for delivery to go well, and the same people see your ad over and over.
So the decision runs in this order:
- Wide area (a whole county or region): start at 1% or 2%, where the resemblance is sharpest.
- Tight area (a 12 to 25-mile radius): go straight to a wider band, between 3% and 10%. The lost precision is offset by volume that is finally sufficient inside your area.
- Final size in the area too small whatever the percentage: a lookalike audience is not the right tool. See below.
Testing 1%, 3% and 5% side by side on a small local budget usually produces nothing readable. Each ad set gets too little delivery for a difference to show. Better to pick a band that fits your area and judge it over several weeks.
Why they work less well than before
A few years ago, a good lookalike audience was often the most profitable targeting in an account. Three changes have eroded that edge.
The signal got thinner. Since tracking restrictions on iPhone and the decline of third-party cookies, Meta recognizes less well who does what outside its apps. Client lists match less well than before, and the resemblance is computed on poorer material.
Advantage+ changed their status. With Advantage+ Audience, a lookalike is no longer a boundary but a suggestion. Meta starts with those profiles, then expands freely if it thinks it can find cheaper conversions elsewhere. The logic of strict controls versus suggestions is covered in the article on local targeting with Advantage+ or manual settings. Keep just one consequence in mind here: in an Advantage+ campaign, your lookalike sets a starting direction, it no longer holds the campaign in place.
The algorithm already does this work. Meta now learns directly from the conversions you send back. If your account passes reliable events, the algorithm builds its own form of resemblance continuously, from your real results. The lookalike becomes a shortcut to get started rather than a lasting advantage.
In practice, a lookalike audience keeps its value mostly when an account launches, while the algorithm knows nothing yet about your buyers, and in manually targeted campaigns, where it is still a real limit.
When not to use them
A lookalike audience is not a reflex for every campaign. You are better off without one in these situations:
- Your list has fewer than a few hundred actually matched clients. The signal is too weak to beat well-tuned location targeting.
- Your only available source is a list of unqualified leads. You would be asking Meta to reproduce your bad contacts.
- Your area is so tight that the final audience drops to a few thousand people. Broad targeting across the area, with an ad that qualifies well, gives delivery more room.
- Your clientele has changed. If you moved from small renovations to large projects, a client list from five years ago steers Meta toward a profile you no longer want.
- Your account already sends back qualified conversions in volume. The algorithm has a fresher signal than your list. The lookalike adds little.
In all these cases, the useful work lies elsewhere: an ad that clearly states who it is for, a form that filters, and kept appointments sent back to Meta so the algorithm learns from the right contacts.
The decision, in short
Build a lookalike audience if you have a clean list of signed clients, ideally with project values, and if your area stays wide enough for the final audience to keep some volume. Pick a low percentage over a large area, a wider one over a small area. Use it first when an account starts out, or in a manually targeted campaign.
Skip it if your only source is a lead export, if your list is too short, or if your account already learns from reliable conversions. The lookalike audience remains a useful tool, as long as you stop handing it the whole campaign.
To estimate what a contact may cost depending on these choices, see also the cost per lead benchmarks by industry.