Can Meta learn from your bookings, or only your clicks?

The two floors, as published

• Meta: about 50 optimisation events per ad set in 7 days to leave the learning phase. Pixel, Conversions API and modelled events all count. Meta measures it per ad set, which is the part that decides everything below.
• Google: at least 15 conversions in the last 30 days to use Target CPA, and 30 in 30 days recommended for it to perform. Counted per campaign, or per portfolio strategy where campaigns share one, not per account. Strategy names changed in June 2026; the floors did not.

What the table does

Take your monthly count at each depth. Times the share that came through Meta, divided by 4.35 weeks, divided by the number of ad sets. That is what one Meta ad set sees in a week. The rest of the volume, divided by the number of Google campaigns (or portfolio strategies), is what one campaign sees in a month. We set each against its floor.

The standard advice, including Meta's own, is: if you cannot reach the floor on the deep event, optimise toward a shallower one. That buys people who do the shallow thing. Whether that is a good trade depends on what a form fill is worth against a visit that happened, and this page does not decide that. It says which events you can train the machine on at your volume, and what volume or structure would let you train it on the deeper one.

Worked examples, checked on the page

These are the tests the arithmetic runs against. This page computes the badge when it loads. Nobody typed it.

What it leaves out

• Meta's floor is a guideline, not a hard cut. Ad sets sometimes exit learning below it and sometimes stall above it.
• Modelled and deduplicated events change the count in ways only your Events Manager shows.
• Google's floor is for Target CPA. Maximise conversions runs with fewer; Target ROAS wants more.
• Seasonality: a floor cleared in May is not cleared in January at the same budget.