How a lookalike audience works
You start with a source: a customer list you upload, people who triggered a particular event on your site, everyone who watched a video, everyone who engaged with the page. Meta looks for what the accounts in that source have in common across the signals it holds, then assembles a new audience of people in a country you choose who resemble them.
You also choose how tightly to match. A narrow setting produces a smaller audience that resembles the source most closely; a wider setting reaches much further and resembles it less. The important thing is that the output is only ever a copy of the input. The audience is built out of the pattern in your source, so the source, not the setting, is what decides whether it is any good.
Why lookalike audiences matter
They let you prospect using your own data instead of guessing at interests. Rather than assembling a list of hobbies and job titles you hope describe a buyer, you point at the people who already bought and ask for more like them. That is a stronger starting point in almost every account, and a much stronger one in markets where interest data is thin — accounts in Nepal carry far less of it than accounts in the UK or the United States, so behaviour-based sources tend to outperform interest lists here.
They also age. A source built from a business’s customers two years ago describes a business that has probably moved on: different products, different price point, different kind of buyer. The audience keeps describing the old one until you rebuild it.
Where lookalike audiences go wrong
Weak sources, most of the time. A list of every lead including the unqualified ones, or every site visitor including the accidental ones, produces an audience that resembles nothing in particular. Meta will build it happily and it will perform like broad targeting with extra steps.
Overlap is the second problem. Several lookalikes running at once, alongside a retargeting campaign, means your own ad sets compete for the same people and you bid against yourself. The third is expecting the audience to do the work of the offer. A precisely built lookalike shown a weak ad still fails, and a narrow lookalike is often so close to your existing customers that it is not prospecting at all.
How to get it right
Build from your best source rather than your biggest: buyers over enquiries, repeat buyers over one-off, high-value customers over the full list. Refresh it on a schedule so the audience keeps describing the business you run now, and exclude existing customers and your active retargeting pools so prospecting stays prospecting.
Test a narrow and a wide version instead of assuming the tight one wins, since a small audience exhausts quickly and pushes frequency up. And keep your written ideal customer profile honest against what the source list actually contains — where the two disagree, the list is usually right and the profile is wishful.