Lookalike seed for a small store: purchasers or add-to-cart?

Lookalikes still get pitched as the fix for a small store, but I think the seed matters more than the percentage. If you upload around 80 purchasers from the last 90 days, the lookalike mostly learns your most loyal repeat buyers, and that pool is too small to tell Meta much. You end up aiming at people who would have bought anyway.

Using add-to-cart or initiate-checkout as the seed gives Meta more signal and a wider net, though you take on more window shoppers in exchange. For a store doing under a few hundred purchases a month, I'd lean toward the broader event and let the creative do the filtering. Has anyone seen the purchaser seed beat that on cost per purchase at low volume?

Eighty is thin, but I'd worry more about what you judge the seeds on. Rule of thumb: if you have under about 100 purchasers in your window, seed with add-to-cart, and compare the two on cost per purchase, not cost per add-to-cart, or the wider seed will look cheap for the wrong reason.

Comparing on cost per purchase is the right call. The answer really comes down to one number: your add-to-cart to purchase rate over the last 90 days. If a good share of carts end in a sale, the ATC seed is mostly people who got distracted, so it's a reasonable bet. If only one cart in twenty converts, the seed is mostly price comparers, and the purchaser seed might give you a cleaner signal even at 80 people. Check that rate before you pick.

Agreed on the ratio, but I'd sanity check the inputs before trusting it. Compare the ATC count in Events Manager with the cart events GA4 records over the same 90 days. If the Pixel double-fires on some browsers or misses ATC on iOS, the rate will look better or worse than it really is. Use the same attribution window on both sides too, or the purchase counts won't line up.

The 80-purchaser example may not even build. As far as I know, Meta wants at least 100 matched people in a source before it lets you create a lookalike, so at that volume a purchaser seed probably isn't an option. Check your real count against that before planning a test.