“Learning limited” might be the most ignored warning in Ads Manager. It sits next to half the ad sets in a typical account, week after week, while the budget keeps going out. Most people treat it as background noise. It isn’t. It’s the platform telling you the account is built in a way it can’t learn from. On a budget of $5,000 to $30,000 a month, that’s almost always a structure problem, not a creative one.
The good news: it’s one of the cheapest problems in paid media to fix. You rarely need more budget. You usually need fewer places to spend it.
Why this matters
What the learning phase actually is
Meta’s delivery system enters the learning phase whenever you create a new ad or ad set, or make a significant edit to one, and uses that time to work out how the ad set should deliver (Meta). An ad set usually exits after about 50 results in the week after its last significant edit (Meta). Until then, delivery is less stable and the cost per result is usually higher (Meta). When an ad set isn’t getting enough results to get there, the status changes to “Learning limited” (Meta).
The detail that matters most: those 50 results are counted per ad set. Not per campaign, and not per account.
The maths most accounts fail
Take an account spending $30,000 a month on Meta. That’s roughly $6,900 a week. Spread across 15 ad sets, each one gets about $460 a week. At $40 per purchase, that’s 11 or 12 purchases per ad set, per week. Nowhere near 50. Every one of those ad sets will sit in learning more or less permanently.
Put the same budget into three ad sets and each gets about $2,300 a week, or roughly 57 purchases at the same cost. Same spend. Completely different machine.
The quick version of the sum: multiply your cost per result by 50 to get the weekly budget one ad set needs, then divide your weekly budget by that number. The answer is how many ad sets your budget can actually feed. A $20 cost per lead, for example, means about $1,000 a week for a single ad set (I Can Learn Ads).
Meta says so itself
This isn’t an agency theory. Meta’s own guidance warns that when you create many ads and ad sets, its delivery system learns less about each one, and that combining similar ad sets also combines what it learns (Meta). It also tells advertisers to set budgets large enough to get enough results (Meta).
Most guides on the topic land in the same place, and they’re right: consolidate, give each ad set enough budget, and stop editing it every other day (AdLibrary, Modern Marketing Institute, ZenWeb).
Why accounts end up fragmented
In our experience, fragmentation rarely happens on purpose. It comes from building the account for the report rather than for the algorithm: one ad set per audience, per product range, per age band, so every line can be reported on. It looks organised. It learns like a small account.
It also costs twice. Once in higher costs while learning, and again in decisions made on numbers that swing from day to day. Those swings make it tempting to pause the “losers” and tweak the rest, and every edit made during learning resets it (Meta).
We saw the other side of this with an Australian retailer. We restructured its Meta campaigns into lean, focused units with cleaner optimisation signals, and across Q4 2024 its ROAS went from 1.94x to 2.32x while spend fell by about 7% (Dadek Digital). Structure was one of four changes, alongside a continuous creative testing engine and spend decisions tied to POAS, so we won’t pretend it did the job alone. But none of the others work on an account that never stops learning.
What people overestimate
That 20% is a Meta rule
Most guides say a budget change of more than 20% resets learning (Modern Marketing Institute, ZenWeb). Another puts the line at about 25% (AdLibrary). Meta itself publishes no percentage. It says frequent budget changes can cause an ad set to re-enter learning, as can edits that meaningfully change how an ad set might perform (Meta). Treat 20% as a cautious habit, not a law.
That Google works the same way
This is the big one. A lot of Google Ads advice copies Meta’s rules: change your target CPA, target ROAS or budget by too much and you’re back in learning (Coinis, Ian Redford). Google’s own help centre now says the opposite. Changes to budgets, targets, creative or targeting never trigger a “Learning” status or remove historical data (Google). The status only appears after a new or reactivated bid strategy, a change to the campaigns, ad groups or keywords inside it, or a significant change to the conversion goal (Google). It never lasts more than seven days, and ads keep serving normally the whole time (Google).
That doesn’t make daily target changes a good idea. Google still advises against several target changes within one conversion cycle, and says performance takes one to two conversion cycles to settle on a new goal (Google). But on Google, the label is a seven-day notice, not a trap. If a Google campaign feels stuck, look at tight targets, small budgets and thin conversion data, which Google itself names as the things that slow a campaign down (Google). The fix overlaps with Meta’s: Google recommends consolidating ad groups and campaigns so its bidding learns from larger pools of data (Google).
That “don’t touch it” is the strategy
The opposite mistake is freezing the account. Meta is clear that you shouldn’t try to avoid the learning phase completely, because testing new creative and new approaches is how performance improves over time (Meta). So the answer isn’t banning changes. It’s planning them: batch your edits into one change, give each change a clean week, and test new creative in its own small unit so your proven ad sets keep learning.
That the label is the score
“Learning” is a delivery status, not a business result. An ad set that’s still learning but sitting near your target is often worth leaving alone, while one that’s out of learning and still above your break-even cost is the real problem (I Can Learn Ads). And the platform can only count the results it can see. If tracking is missing part of your sales, an ad set can look stuck while the business is doing fine. Check the platform’s numbers against your own, and judge the account on what the business made, before you rebuild anything.
Where to start this week
Open Ads Manager, add the “Last significant edit” and Results columns, and look at each ad set since its last significant edit (Meta). Count how many are short of 50. Then do the sum: your cost per result times 50, divided into your weekly budget. If you’re running more ad sets than your budget can feed, merge the similar ones, make all your changes in one go, and give the account two clean weeks before you judge it. On Google, stop waiting on the label and look at your targets, your budget and your conversion data instead.
Dadek Digital builds accounts around the number of ad sets a budget can actually feed, and judges them on the business’s own numbers rather than the platform’s labels. The full structure, for ecommerce and for lead generation, is laid out step by step in our playbook, The Dadek Digital Pyramid.