Meta Ads learning phase: the minimum budget to exit learning limited

Quick answer

A simple planning benchmark for the Meta Ads learning phase: target CPA times 50, divided by 7. That buys roughly 50 optimization events per ad set in a week. With a $40 target CPA, that is about $286 a day per ad set. If your budget cannot support that volume, run fewer ad sets.

Where does the CPA x 50 / 7 formula come from?

To exit learning, an ad set needs about 50 optimization events within a 7-day window. If each event costs your target CPA, you have to spend 50 CPAs' worth in 7 days. Divide by 7 and you have the daily budget.

The detail that trips up most people: the 50 events are per ad set. A campaign with three ad sets needs 50 in each, not 50 combined.

Meta has tested a lower threshold. Jon Loomer reported in July 2024 that for a few weeks his accounts showed a learning phase of 10 optimized events in three days, then went back to 50 events in seven days with no explanation, and that Meta's documentation still said 50. Plan for 50 until your own Ads Manager tells you otherwise.

Is the 50 events rule official, and is the budget formula?

The threshold is Meta's: about 50 optimization events within 7 days of launch or of the last significant edit. The CPA x 50 / 7 budget is not. It is arithmetic on top of that threshold, and it assumes your actual cost per result matches the target. We have not found a minimum budget for leaving learning in Meta's documentation, so treat any such number, ours included, as a planning estimate.

What is the minimum budget for each target CPA?

How does the formula work for lead ads?

Swap CPA for your target cost per lead. At a $15 target CPL, $15 x 50 / 7 is about $107 a day per ad set optimizing for leads. A deeper, rarer optimization event needs more budget to reach the same 50 a week. Form setup, testing and lead quality are in Meta Lead Ads and instant forms.

How long does the Meta Ads learning phase last?

There is no fixed number of days. Learning lasts until the ad set reaches about 50 optimization events within 7 days of launch or of its last significant edit. With enough budget, that can happen within the first week; without it, the ad set never gets there and shows learning limited.

Significant edits reset the count. Meta's own wording, quoted by Jon Loomer, is that learning happens on new ads and ad sets and after significant edits. His list of edits that trigger it covers targeting changes, creative changes, a different optimization event, adding new ads, a different bid strategy and restarting an ad set paused for 7 days or more; budget and bid amount changes may or may not trigger it, and Meta does not say how big a change has to be. Batch your edits instead of making a small one every day.

How do you know an ad set is still in learning?

Check the Delivery column in Ads Manager. It shows Learning while the ad set is collecting optimization events and Learning limited when it is not on track to reach them. Judge cost per result only after the status leaves Learning, since Meta describes performance during learning as less stable.

What if the client does not have that budget?

• Run fewer ad sets. With budget for one ad set, run one ad set.
• Optimize for a more frequent event. If purchases are rare, optimizing for add to cart or initiate checkout gets you to 50 events faster, at the cost of optimizing toward a less qualified signal.
• Broaden the audience. An audience that is too small caps how many people can complete the event.
• Accept learning limited with your eyes open. Sometimes it is all the budget allows; just do not expect stable costs.

Is the 2x CPA test floor the same as CPA x 50 / 7?

No. They answer different questions. Roughly 2x target CPA per day is a working starting point for an ABO test, meant to collect initial creative signal; it does not get the ad set out of learning. CPA x 50 / 7 estimates the budget needed for roughly 50 events a week, assuming actual CPA matches the target. See Meta Ads creative testing for the test structure.

Does a budget increase get you out of learning?

More budget helps if budget was the constraint, but a big jump can itself reset learning. Our own approach is to raise budgets in steps of up to 35%; that lowers the risk but does not guarantee learning will not reset. Review delivery after each change. See how to scale Meta Ads campaigns and CBO vs. ABO.

If you automate scaling in Ads Editor, two settings work in the same direction. Auto-Scale by Performance lets you set a minimum interval between changes, and the Minimum results to scale field (3 by default) blocks any increase until the campaign has that many results in the day. The Help Center suggests starting with the Conservative preset to watch the behavior before speeding up (checked on October 31, 2026).

A tight bid can also keep an ad set from getting the conversions it needs to exit learning. If you use a cap, size it from margin as shown in bid cap vs cost cap in Meta Ads.

Does Google Ads have the same learning math?

Similar idea, different timing. Google Ads Help says Smart Bidding performance usually converges on a new target within 1 to 2 conversion cycles. The Learning status comes from structural changes: a new or reactivated bid strategy, campaigns, ad groups or keywords added or removed, or a significant change to the conversion goals. Budgets, targets, creatives and targeting settings do not trigger it. See how to manage Google Ads for multiple clients.

To change a bid strategy in Google Ads without risking the whole campaign, test it on a copy first: see Google Ads experiments.