Learning Limited, Limited Delivery, stuck in review, not spending: what each status means

During learning, Meta is building a model of who converts for you. It spends deliberately unevenly, sampling across placements, times and audience segments to find out where your conversions come from. That exploration is why early costs look bad and swing about: you are paying for information rather than for efficiency.

The phase ends when the model has seen enough conversions to predict with confidence. It needs a meaningful number of them inside a recent window, and the size of that number is what the budget guide works through, because it is really a spending question rather than a status question.

Two things follow. Exploration costs money that does not look like performance, so judging a campaign mid-learning tells you almost nothing. And anything that invalidates the model sends it back to exploring, which is why edits are expensive in a way that is not obvious from the interface.

Learning Limited means the ad set is not getting enough conversion volume for the system to optimise reliably, and Meta has given up waiting. It is not a warning that you are about to fail. It is a statement that the ad set has already stopped improving, and will keep delivering at whatever efficiency it reached.

Fix it by increasing the signal or simplifying the structure, in this order:

1. Count conversions per week for the ad set, not the campaign. This is the number the status is about.
2. Consolidate. Fragmented structures divide signal, and several starved ad sets are worse than one fed one.
3. Consider a shallower optimisation event for a while, so events arrive often enough to learn from.

What does not work is waiting. Learning Limited is a stable state, not a transitional one, and an ad set can sit in it indefinitely while spending your budget at a cost it will never improve on.

It is worth knowing that the threshold is not really a cliff. Meta's own analysis frames it as how much of your spend sits in the learning phase rather than whether a single ad set crossed a line: advertisers with roughly 20 per cent of spend in learning see about 17 per cent more conversions and a 15 per cent lower cost per acquisition than advertisers with roughly 80 per cent of spend stuck there. So the goal is not to hit a magic number once, it is to keep most of your account out of learning most of the time.

Learning requires stable conditions as well as volume. If either is missing the phase never completes, and the two causes look identical from the outside while needing opposite fixes.

If volume is the cause, conversions are arriving too slowly for the window to fill, and the answer is the consolidation and event-depth work above.

If instability is the cause, conversions are arriving but something keeps resetting the clock. Check your edit history before changing anything else, against the list further down this page of edits that actually reset learning. Budget adjustments are not on that list, despite being the change people reach for most readily. New creative, an added ad, a new audience or a changed optimisation event are.

1. Check weekly conversion volume for the ad set.
2. Check the edit history for frequent changes.
3. Confirm the optimisation event is achievable at your volume. An event too deep in the funnel fires too rarely to learn from.

Limited Delivery means the campaign cannot reach its delivery potential. Meta uses this status when something structural stops the algorithm finding enough conversion events to optimise effectively, so it is closer to a configuration complaint than a performance one.

1. Calculate the budget floor for your target cost per result and compare it with what the ad set actually has. The budget guide has the formula.
2. Check whether too many ad sets are dividing the budget.
3. Check the conversion event fires often enough to be worth optimising toward.
4. Check the audience size estimate in the ad set. Very small audiences cannot sustain delivery.
5. Check whether a bid cap or cost cap is set below what the auction currently charges.

That last one is the quiet cause. A cap set below market price is not a discount, it is an instruction to stop bidding, and the campaign obediently stops. Bid strategies covers when a cap is the right tool.

An active campaign that is not spending means Meta cannot win auctions for it. The campaign exists and is technically eligible, so the interface shows nothing obviously wrong. Five things are worth ruling out, in this order.

1. Read the delivery status at ad set level, not campaign level. The specific reason is shown there and people miss it because they are looking one layer up.
2. Confirm every ad is approved and active. One approved ad in a paused state delivers nothing.
3. Check audience size. Below roughly 50,000 people, delivery becomes unreliable.
4. Check for a bid cap or cost cap below market price.
5. Confirm the schedule, and check billing. A failed payment stops everything and does not always announce itself clearly.

If billing turns out to be the cause, fix it quickly rather than waiting for a retry, because an unresolved failure can escalate beyond a pause. The billing section covers it.

Most ads complete review within 24 hours. A review extending beyond 48 hours indicates one of three things: a policy flag that has routed it to a human, a high-volume queue period, or an account history that triggers extra scrutiny.

1. Check the exact status. In Review and Pending Review are different states.
2. Note how long it has actually been. Under 24 hours is normal and there is nothing to do.
3. Check whether the campaign is in a Special Ad Category, which adds review time.
4. Check the account's recent disapproval history. Meta names account history as one of the three things that extends a review.
5. After 48 hours, contact support with the specific ad ID rather than resubmitting.

Editing a campaign resets or disrupts learning, and during learning CPMs are elevated and delivery is unstable. The larger the edit, the larger the disruption. What you are seeing is the exploration cost being paid again, not a new problem.

1. Check the edit history for what actually changed and when.
2. Compare CPM for the week before the edit with the week after, rather than against yesterday.
3. Check whether delivery now shows Learning or Learning Limited.
4. Count the days since the edit. Learning typically resolves in about seven.
5. Make no further edits while it settles, however tempting.

The trap is reacting to the disruption with another edit, which restarts it again. Each intervention feels justified, because costs genuinely are bad at the moment it is made, and the result is an account that never settles.

What is the difference between Learning Limited and Limited Delivery?

Learning Limited is about signal: the ad set is not getting enough conversions to optimise, so it has stopped trying to improve. Limited Delivery is about configuration: something is preventing the campaign spending what it could. They often share a root cause in a budget spread too thin, which is why they get confused, but the fixes differ.

How long does the learning phase take?

It is measured in conversions rather than days, so it takes as long as your account needs to produce them. A high-volume account can clear it in a couple of days; a low-volume one may never clear it. This is why a fixed answer in days is always wrong.

How long should I leave a new campaign before judging it?

Do not panic-edit in the first six hours, and do not make several small changes a day, because that guarantees you never leave learning. But do not sit and bleed money on an obvious loser either. Judge early on leading indicators rather than final purchases: hold rate on video, click-through, add to cart, and the trend in cost per result. Set a rule against spend: if an ad has spent two or three times your target cost per acquisition with no result and no leading signal, it is not shy, it is failing, and you should cut it.

Will my costs come back down after learning?

Usually, and that is the point of tolerating the early numbers. If they do not settle within about a week of the phase completing, the problem is not learning and it is worth reading why CPA rises instead.

Can I edit a campaign without resetting learning?

Yes, and the dividing line is more specific than most advice admits. Changing targeting, changing creative, changing the optimisation event, changing your bid strategy, adding a new ad, or pausing for seven days or more all restart learning. Changing the budget amount, the spend cap, the bid control amount or the ad set spending limit does not. So raising a budget is safe in learning terms, and adding one more ad to a working ad set is not,.

My ad set says it is out of learning but performance is still bad.

Then learning was never the problem. Exiting the phase means the model is confident, not that the campaign is profitable, and a confident model can be confident about a bad offer. Check what you can afford to pay with the break-even calculator before rebuilding the campaign.