Meta lookalikes vs broad targeting: does the seed audience still matter?

Take a dental clinic running lead gen on a small budget. The old playbook was a 1% lookalike off your customer list and hope for the best. Now Advantage+ audience pushes most accounts toward broad targeting, and honestly the creative does most of the work. My view is that broad with a strong hook and a clear offer beats most lookalikes after the first few weeks. I'd still seed when the list is high-value, like repeat patients with a big average ticket, because that gives the algorithm a clearer picture early on. A seed built from every newsletter subscriber usually just reproduces your existing audience with less reach. Has anyone run a clean split on this at lower volume? Most of the tests I'd trust come from big accounts.

It comes down to how many conversions the account gets a week on the event you're optimizing for. A small clinic booking maybe 5 to 10 consults a week gives broad very little to learn from, so a seed built from recent paying patients hands Meta a signal much closer to what you actually want. Once you're in the dozens a week, broad usually catches up fast and the seed matters less. Your newsletter point holds either way, since that list mostly looks like people who already know you.

Agreed on the volume threshold, but at low volume the quality of the seed matters more than its size. A list of 40 recent paying patients often beats a 3,000-person list of newsletter readers, because the small list sits much closer to the actual conversion event. Rule of thumb: seed only with people who did the thing you're paying for, and skip the lookalike entirely if you can't name that group.

Quality over size, agreed, but a 40-person list may not clear the floor. Meta generally wants about 100 matched people in a source audience before it will build a lookalike, and match rates on patient phone and email lists tend to be lower than clinics expect. So 40 patients might become 25 matched people, and at that point you're not really running a lookalike.

The list is still useful, just for something else. You can use it to exclude existing patients from prospecting, so you stop paying to reach people already in your recall system, or as a small retargeting pool. For prospecting, broad targeting with a strong offer is probably the honest test until the list grows.

The match count is the number to watch, not the export. A clinic sees "40 patients" and assumes that's the seed, when Audience Manager might show only 25 matched. So check the matched count first, and only build a lookalike once it clears Meta's threshold. Below that, use the list for exclusions and a small retargeting pool, and let broad do the prospecting. A seed that barely qualifies gives you a lookalike that's mostly guesswork, which means the result tells you very little about whether lookalikes work for you.

If you're near the floor, the fix is often in the upload file. Meta matches on more than email and phone, so adding first name, last name, zip and country can lift the match rate a lot. Clinics tend to export only phone numbers from the practice system and then wonder why half the list doesn't match. It also pays to recheck the matched count after every upload, since phone numbers go stale and the match rate slowly drifts down.

Agreed on the upload fields, that's an easy win. What decides whether it's worth the effort is how many patients have a usable email on file, since email is usually the field doing most of the matching. Phone-only lists tend to match worse on Meta than clinics expect.

Email count isn't the only thing to check, recency matters too. A practice system often holds addresses for patients who last came in three or four years ago, and plenty of those have bounced or changed. Dead addresses drag down the match rate, and the ones that do match can skew toward an older group than your current patients.

For a seed list, filtering to a last visit within the past 12 to 18 months usually gives you a smaller list that matches better and actually describes who you're treating now. Are you filtering the export by last visit date, or taking everyone in the system?

Agreed on recency, and the filter should apply to the conversion event too, not just last visit. A patient who came in for a cleaning two months ago may be on the list, but they tell Meta nothing about who books consults. Filtering to people who booked a consult or treatment in the past 18 months gives you a tighter seed than a generic visit filter.