Lookalike seed on a small list: past customers or a wider high-intent event?

Which seed to build a lookalike from comes up a lot with small accounts. The textbook answer is your best customers, and that works when you have a few thousand purchasers. At 80 or so customers a month, the list is tiny and skews toward people who already knew the brand, so the lookalike ends up mirroring your existing fans instead of finding new buyers.

Take a dental clinic seeding from past patients. Meta will mostly find people who look like long-time patients, which might be fine for recall campaigns but is weak for new patient acquisition. I'd lean toward seeding from a broader high-intent event instead, like people who started the booking form or spent real time on the pricing page, and letting the creative do the qualifying. Has anyone running lookalikes off thin lists seen the customer seed beat the wider one?

Mira, with only 80 customers your real match rate is probably well below the raw count, and Meta needs roughly 100 matched people before it will even build a lookalike. Booking-form starters should clear that bar more easily, but they're noisier, so I'd run both at similar budgets and let the results decide instead of arguing it on theory.

Sasha, 100 is just the minimum Meta will accept for a source audience, not the point where lookalikes start working well. And since a booking-form start also fires on people who bail halfway, count unique people over the last 90 days before you trust that seed.

Fair point on the minimum, and checking the 90-day unique count is the right move. It does depend on volume, though. At a few hundred conversions a month, a broad event seed can work fine because the noise averages out. At 80 a month, I'd stick with the customer seed unless the event fires on a lot of real starts, since a thin event seed just mirrors the people who bounce.

Agreed on volume, but geography is what I'd check first. A 1% lookalike is the top 1% of people in the country it's built for, so if your ads are then limited to one city, the usable audience can shrink a lot. For a clinic that serves a single city, either seed from local customers only or widen the lookalike to 2 to 3%. Otherwise the seed itself is fine, but Meta has very little room to find anyone new.

Mira, the geography point is the one people miss, and the rule I'd use is simple: look at the reachable audience size after your location and age settings are applied, not the lookalike percentage you picked. Meta builds that percentage from the whole country, so a 1% lookalike can shrink to a few thousand people once you narrow it to one city.

If that number looks too small, widen the percentage or pool the seed over six months. The older patients are a bit less representative, but a seed of a few hundred people gives Meta far more to work with than 80.

Agreed, and I'd run that check before launch, not after. Set the city and age range on the ad set first, then look at the audience size Ads Manager shows for the lookalike. After that, exclude existing patients from prospecting. If you don't, a clinic seeded from its own patient list ends up paying to reach people who already booked there, which is the recall problem Mira described, just labeled as acquisition.

If you pool the seed over six months, excluding only that seed leaves every older patient sitting in the prospecting pool. Build the exclusion from the full all-time patient list and refresh it whenever you refresh the seed. Otherwise you're still paying to show clinic ads to someone who was in the chair in 2019.