Lookalike audiences explained: your source list decides everything

Aug 22
3 min read

Updated: 4 days ago

Introduction

The idea is straightforward. Give the platform a list of your customers, and it finds more people who resemble them.

What gets overlooked is that the output is entirely determined by the input. A lookalike built from the wrong list finds more of the wrong people, efficiently and at scale.

1. Lookalike audiences explained: the platform copies whatever you give it

You supply a source audience. The platform identifies patterns in it and builds a much larger group sharing those characteristics.

The mechanism is indifferent to whether the source was any good. Seed it with everyone who ever visited your website — including people who bounced in three seconds — and it will faithfully find more people who behave like bouncers.

Every decision worth making in this feature is a decision about the source list.

2. Small and clean beats large and mixed

A common instinct is to make the source as big as possible. It is the wrong instinct.

A few hundred actual paying customers is a better seed than fifty thousand website visitors, because the pattern in the first group is "people who bought" and in the second it is "people who visited".

Platforms usually need a minimum in the low hundreds. Meet that with your best data rather than padding it to look substantial.

3. Seed from your highest-value customers

Take the logic one step further. Not all customers are equally worth copying.

If you can identify the top fifth by lifetime spend, or those who bought more than once, use only those as the source. You are asking for more of a specific kind of person, so specify the kind you want.

This single change is the most reliable improvement available in the feature, and it costs nothing but a spreadsheet sort.

4. Percentage size trades reach against similarity

Most platforms let you choose how broad the resulting audience is, expressed as a percentage of a country's population.

A 1% audience is the closest match and the smallest. A 10% audience is far larger and much less similar to your source.

Start narrow. Broaden only when the narrow version has exhausted its results, and treat each percentage as a separate audience with its own performance rather than assuming bigger is a scaling step.

5. Layer a location filter over it

Lookalikes ignore your service area unless you tell them not to.

For any business serving a defined geography, always constrain the audience by location as well. Otherwise a well-built lookalike spends efficiently on people who cannot buy from you — which looks like cheap clicks and delivers nothing.

Same for any other hard qualifier: age restrictions, language, device where relevant.

6. Exclude the people you already have

Without exclusions, your lookalike will show ads to your existing customers, because they resemble themselves most of all.

Add your customer list and recent purchasers as exclusions. This is prospecting; existing customers should be reached through retention activity that costs nothing per contact.

Also exclude recent enquirers who are already in your sales process, or they will see prospecting ads while mid-conversation with you.

7. Refresh the source, and mind the permissions

A source list built two years ago describes a customer base you may no longer have.

Update it quarterly, and rebuild the audience when you do. If your offer or pricing changed, the old customers may be the wrong template entirely.

Also confirm you are permitted to upload customer data for advertising. That depends on what people agreed to when they gave you their details, and it is worth checking rather than assuming.

8. Test it against a plain interest audience

Do not adopt this because it sounds sophisticated. Compare it.

Run the lookalike against a straightforward audience built from location plus one or two relevant interests, with the same creative, offer and budget. Judge on cost per enquiry.

Lookalikes usually win when the source list is genuinely good and lose when it was assembled from weak signals. That comparison tells you which situation you are in, which is more useful than any general claim about the feature.

Conclusion

The source list is the whole decision. Seed from paying customers rather than visitors, and preferably from your highest-value ones.

Start at the narrowest percentage, always layer location and other hard qualifiers on top, exclude existing customers and live enquiries, refresh the source quarterly, and test it head to head against a simple interest audience on cost per enquiry.