Enter a daily budget and your cost per result to see whether an ad set can actually reach the events it needs to leave the learning phase, and what it would take if not.
Events in a 7 day window
20 of 50
Days to 50 events
17.5
Stays in learning
Needs about $250 a day
At $100 a day and $35 per result, this ad set produces about 2.9 events a day. Meta exits the learning phase at roughly 50 events in a rolling seven day window, so anything slower than that leaves delivery unstable and the numbers hard to trust.
Significant edits restart this clock. Changing creative, audience, optimisation event or making a large budget change puts the ad set back into learning, so batch those edits rather than trickling them.
Fewer, better fed ad sets, launched in one pass
If the answer is fewer ad sets with more budget each, the work becomes deciding what goes in them. Volume Creatives builds and publishes the whole batch with naming, UTMs and enhancement opt-out already applied.
Enter the daily budget for one ad set
Use the ad set level budget. If the campaign runs on campaign budget optimisation, divide the campaign budget by the number of ad sets for a rough figure, then remember Meta will not actually split it evenly.
Enter your cost per optimisation event
This is the cost of whatever the ad set optimises for: purchase, lead, add to cart, landing page view. Pull the trailing 30 day figure for that event from Ads Manager. The event matters more than the budget, because 50 purchases cost far more than 50 page views.
Read events per week and days to 50
The bar shows how many events the budget buys in a seven day window against the 50 Meta looks for. Days to 50 tells you how long it would take at this pace. Anything over seven days means the ad set will likely show learning limited.
Act on the verdict
If it says stays in learning, the tool gives the daily budget that would fix it. Your options are to raise budget to that figure, optimise for a cheaper event, or consolidate ad sets so fewer of them share the money.
You enter
You get
$120 ÷ $30
4 × 7, or 56% of the threshold
50 ÷ 4
Needs about $214 a day (50 × $30 ÷ 7)
At $120 a day this ad set buys 28 purchases a week, just over half of what it needs, and would take nearly two weeks to reach 50 in total, which is outside the rolling window. Raising the budget to $214 fixes it. So does switching the optimisation event to add to cart at, say, $8 each, which turns the same $120 into 15 events a day and 105 a week. Which fix is right depends on whether add to cart predicts purchase in your account.
| Events per day | Events in 7 days | Days to 50 | Daily budget at $30 per event | Verdict |
|---|---|---|---|---|
| 2 | 14 | 25.0 | $60 | Learning limited |
| 4 | 28 | 12.5 | $120 | Learning limited |
| 7.2 | 50 | 7.0 | $214 | Exits, no margin |
| 10 | 70 | 5.0 | $300 | Exits comfortably |
| 15 | 105 | 3.3 | $450 | Exits in half a week |
Events in 7 days = events per day × 7. Days to 50 = 50 ÷ events per day. The budget column assumes a $30 cost per event; scale it to your own cost. About 7.2 events a day is the pace that exits learning inside the seven day window.
When an ad set is new or significantly edited, Meta's delivery system explores to work out who responds. Results during this period are more volatile and usually more expensive, which is why decisions made inside it are unreliable. The ad set leaves the phase once it accumulates roughly 50 optimisation events in a rolling seven day window. Below that pace it can sit in learning limited indefinitely.
Events per day = daily budget ÷ cost per event. Events in the window = events per day × 7. Days to 50 = 50 ÷ events per day. Budget needed = cost per event × 50 ÷ 7. That last one is the number to remember: whatever your cost per result is, the daily budget that exits learning is about 7.14 times it. A $20 lead needs about $143 a day. A $60 purchase needs about $429. Most accounts that live in learning limited are simply running ad sets at a third or a quarter of that figure and hoping.
Fifty purchases is a very different budget from fifty add to cart events or fifty landing page views. If purchase volume is too low to reach the threshold, optimising for an earlier event is a legitimate way to get out of learning, with the trade-off that you are optimising for something less directly tied to revenue. Make that trade deliberately rather than discovering it. A useful check: if your add to cart to purchase rate is stable and above roughly 20%, cart optimisation tends to hold up. If carts are cheap because the traffic is curious rather than ready to buy, you will exit learning and watch purchases fall.
Significant edits reset learning: changing creative, changing targeting, changing the optimisation event or attribution setting, and large budget or bid changes. Pausing for seven days or longer also resets it. Renames and small budget adjustments generally do not. When you have to make several learning resetting edits, batch them into one session so you pay the cost once instead of every few days. This is also why duplicating a winning ad set to scale it is usually better than editing the original: the winner keeps its history while the copy learns. See why an ad set goes back into learning and how to duplicate without resetting it.
Raise the budget to the figure the tool gives. Consolidate: two ad sets at $100 each that both sit in learning limited become one ad set at $200 that exits, and Meta's own guidance leans toward fewer, broader ad sets for exactly this reason. Or move the optimisation event one step up the funnel. Cutting the cost per event through better creative is the fourth option and the only one that improves economics rather than just pace, which is why the combination calculator and a proper creative testing cadence matter more than any budget trick.
The 50 event figure is a reference point, not a contract, and Meta's behaviour varies by account and objective. The calculation here also assumes your cost per result is roughly stable, which it will not be in the first days, when it typically runs 20 to 50% above where it settles. Campaign budget optimisation makes the per ad set budget a moving target, so the input is an estimate under CBO. Use the tool to answer a structural question, whether this budget can ever support this many ad sets, rather than as a forecast. Pair it with the budget calculator when planning a test, and the ROAS calculator to make sure the cost per event you are budgeting for is one you can afford.
Meta's documentation says an ad set typically exits the learning phase after about 50 optimisation events within seven days of its last significant edit. That is 50 of whatever event the ad set optimises for, so 50 purchases for a purchase campaign or 50 leads for a lead campaign. It is a rolling window, so the pace has to be sustained, not hit once.
Learning limited is the status Meta shows when it predicts the ad set will not reach 50 events in the seven day window at its current budget, audience and cost. Ads still deliver, but the system stops actively trying to exit learning and performance is usually less stable and more expensive. It is a diagnosis, not a penalty: fix the pace and the status clears.
Multiply your cost per optimisation event by 50 and divide by 7. At a $30 cost per purchase that is about $214 a day per ad set. At a $5 cost per lead it is about $36 a day. The calculator does this for you and shows how far short a given budget falls.
Meta treats these as significant edits: changing targeting, changing placements, changing the optimisation event or attribution setting, adding new creative or editing existing ads, switching bid strategy, and large budget changes. Pausing an ad set for seven days or more also resets it. Small budget nudges, renames and schedule tweaks generally do not.
Meta has not published a fixed percentage, so treat this as a rule of thumb: changes of roughly 20% or less, made no more than once a day, usually do not restart learning. Doubling the budget in one step often does. When you need to scale a winner, step it up gradually or duplicate the ad set at the higher budget and keep the original running.
Sometimes, and it is a real trade. Optimising for add to cart instead of purchase might take an ad set from 15 to 60 weekly events and exit learning, but Meta will now find people who add to cart rather than people who buy, and those are not the same audience. It works best when the cheaper event correlates tightly with the one you care about. Check the downstream purchase rate after a week rather than assuming.
Not automatically. Plenty of small accounts run profitably in learning limited for months because the ad set still delivers and the CPA is fine. The real cost is unpredictability: results swing more day to day, and scaling is harder because the system never stabilised. If your numbers are good, leave it. If you are trying to scale, fix the pace first.
Connect your ad account, drop in the creatives, and watch the batch land paused. Nothing spends until you say so.