How many conversions does a Meta ad set need per week?

Meta says an ad set is likely to become learning limited if it is unlikely to receive around 50 optimization events in the week following your last significant edit. Structuring ad sets to achieve at least 50 events over a 7-day period is Meta's own guidance for getting through the learning phase.

Why shouldn't I optimize a new Meta campaign for purchases?

Because purchases are the rarest event in the funnel. A new account will not produce 50 purchases a week, so the ad set stays learning limited and delivery is unstable and expensive. The algorithm also has almost no examples of what your buyer looks like, so it cannot find more of them.

What does learning limited mean on Meta?

It means the ad set is unlikely to get the roughly 50 optimization events per week it needs for delivery to stabilize. Meta lists the usual causes as small audience size, low budget, low bid or cost control, high auction overlap, an infrequent optimization event, or running too many ads at once.

Don't Optimize a New Meta Campaign for Purchases

Q: What should a new Meta campaign optimize for instead?

An event that happens often enough to teach the system: a lead, an add to cart, or a genuinely meaningful button click. Meta's own remedy for a learning limited ad set is to choose an optimization event that occurs more frequently, and it gives moving from purchases to add to cart as the example. Move deeper once volume supports it.

A new Meta account launches, the campaign is set to optimize for purchases because purchases are the goal, and the results are terrible and stay terrible. The usual diagnosis is the creative or the audience. Often it is neither. The campaign was asked a question it had no way to answer.

The mechanic

Meta ad sets go through a learning phase while delivery calibrates. To get through it, Meta says an ad set needs roughly 50 optimization events per ad set per week. Fall short and the ad set is flagged learning limited, and delivery stays unstable and expensive.

Now count the purchases a new account produces in a week. It is almost never near 50. So the ad set sits in learning limited indefinitely, and the algorithm is working with almost no examples of what one of your buyers actually looks like.

You asked it to find buyers using almost no examples of a buyer.

Why it fails, in plain terms

Imagine handing someone a box of product and telling them to go find the people who will buy it, with a rule that they may not approach anyone who won't. They have no idea who those people are yet. They freeze.

Tell them instead to hand it to anyone, and by the end of the day they will have noticed who actually took it. Now there is a pattern, and tomorrow they can go find more of those people.

That is the difference between a pixel with data and a pixel without it. The optimization event is not just a reporting choice, it is the teaching signal.

What to do instead

Start on the event that happens often enough to teach the system, then move deeper once volume supports it.

Pick a frequent, meaningful event. A lead, an add to cart, or a click that genuinely indicates intent. Frequent enough to clear 50 a week, meaningful enough that optimizing toward it isn't garbage.
Let it collect real examples. The point of the early phase is not cheap conversions, it is a pixel that has seen who converts.
Then move down the funnel. Once purchase volume can realistically support around 50 a week per ad set, switch. Expect the learning phase again, because changing the optimization event is a significant edit.

This is not a workaround. Meta's own remedy for a learning limited ad set is to choose an optimization event that occurs more frequently, and the example it gives is moving from purchases to add to cart. The platform is telling you to do this.

One caveat worth stating

A more frequent event is only useful if it correlates with money. Optimizing toward a shallow event that anyone triggers, such as a page view, just teaches the system to find people who browse. That is the same failure in the opposite direction, and it is exactly what happens when a bid strategy quietly ignores your conversion goal. Pick the shallowest event that still means something.

The takeaways

Meta needs roughly 50 optimization events per ad set per week to exit the learning phase.
A new account cannot produce 50 purchases a week, so the ad set stays learning limited.
Start on a frequent, meaningful event so the pixel collects real examples of your buyer.
Move to purchases once volume supports it, and expect a fresh learning phase when you switch.

Sourced from the field, checked against the docs. The approach and the analogy are our own, taught on mentoring calls; no client data is used. The platform facts are Meta's: an ad set becomes learning limited when it is unlikely to receive around 50 optimization events in the week after the last significant edit, and Meta advises structuring ad sets to achieve at least 50 events over 7 days (About the Learning Phase, About learning limited). Meta's own listed remedy is to "change your optimisation event... choose an optimisation event that occurs more frequently. For example, move from purchases to add to basket."

Optimize a new Meta campaign for purchases and you've made the algorithm blind.

The mechanic

Why it fails, in plain terms

What to do instead

One caveat worth stating

The takeaways

Is an ad set of yours stuck in learning limited?

Optimize a new Meta campaign for purchases and you've made the algorithm blind.

The mechanic

Why it fails, in plain terms

What to do instead

One caveat worth stating

The takeaways

Related reading

Is an ad set of yours stuck in learning limited?