First, rule out a stop and read a full week
Underspend and a stop look alike on a dashboard but have different causes. If the ad set spent normally for weeks and then fell away, or stopped entirely, start with the guide to troubleshooting performance drops, which covers rejected ads, restricted accounts, failed payments and reached spending limits. This page is for the other case: an ad set that has been live and approved since launch and has never used the money it was given. If the problem is the opposite, a day that went over budget, the sibling page on spending more than the daily budget applies instead, and the Meta Ads troubleshooting hub routes every other symptom.
With a stop ruled out, gather three readings before touching anything. The delivery column in Ads Manager shows whether the ad set is active, in learning or learning limited. The ad set's settings show the bid strategy, any cost per result goal or bid cap, the audience definition and the placement selection. And a seven-day view of spend against budget, by day, separates a genuine pattern from a slow Monday. Note the estimated audience size shown in the ad set editor as well; it is an estimate, but a very small figure is a clue in itself. Then use the table: each row names the reading that confirms the cause.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Spend low every day, results arriving at or below the goal | Bid strategy and the cost per result goal or bid cap against the cost actually achieved | Bid cap or cost cap below what results cost | Raise the goal to a figure based on the economics, or return to the highest-volume strategy |
| Spend flat while frequency climbs on a small reach | Estimated audience size; layers of interests, locations and exclusions in the ad set | Audience too small after layered targeting and exclusions | Remove layers, widen locations or ages, or move to a broader audience option |
| Delivery column says learning limited; few results per week | Optimisation event and how often it occurs per ad set | Optimisation event too rare for delivery to predict | Consolidate ad sets or optimise for a more frequent event further up the funnel |
| Several ad sets each spending a fraction, combined reach far below the sum | Audience overlap between active ad sets; auction overlap in the Inspect tool | Overlapping ad sets competing for the same people | Merge the ad sets or exclude one audience from the other |
| Spend capped on one or two surfaces; high cost per thousand | Placement setting; placement breakdown of spend and impressions | Placements restricted to a few surfaces | Return to Advantage+ placements, or add the surfaces the creative can carry |
Fix, verify and prevent each cause
A cost or bid ceiling the auction will not meet
A cost per result goal asks delivery to keep the average cost near a figure; a bid cap limits what it will offer in each auction. Both are guardrails, and both behave the same way when set too low: delivery declines the auctions it would have to overpay for and the budget sits idle. The give-away is that the results the ad set does buy look cheap, often under the goal, while spend stays far below budget. A ceiling copied from last quarter's average or from a different account is the usual origin, because the cost of the next result is always above the average one.
Fix: set the goal from what a result is worth to the business, as the bid strategy section of the scaling guide recommends, and raise it in one change rather than several; if volume matters more than a price guarantee, move back to the highest-volume strategy. Remember that changing the bid strategy or the goal is an edit Meta documents as restarting learning.
Verify: over the following week daily spend rises towards the budget and the cost per result (CPACost of each conversion.) settles near the new goal, not far beneath it.
Prevent: write the economic limit down before launch and derive every cap from it, never from a recent average. The guide to Meta bid strategies shows how to derive that limit from margin and why a new ad set is usually better started without a goal at all.
An audience squeezed until there is no room
Every layer of targeting removes people: a narrow location, a tight age band, an interest stacked on another interest, and then exclusions of visitors, engagers, leads and customers. Each layer looks reasonable on its own; together they can leave an audience that delivery exhausts in days. The pattern is reach that stops growing while FrequencyAverage times a person saw an ad. rises and cost per thousand impressions (CPMCost per 1,000 impressions.) climbs, because the same few people are bid for again and again.
Fix: remove the layer that cuts deepest, usually a stacked interest or a narrow geography, keep only the exclusions that protect against waste, and consider a broader audience option where the objective allows it. The guide to audience exclusions and overlap sets out which exclusions earn their place in each funnel stage.
Verify: the estimated audience grows in the editor, reach keeps rising across the week, frequency flattens and spend follows.
Prevent: check the estimated size after every layer is added during the build, not once the ad set is live.
An event too rare for delivery to learn
Delivery predicts who will take the optimisation event, and it needs a steady flow of those events to predict well. Meta's learning phase documentation describes the learning limited status for ad sets unlikely to gather enough of them; such an ad set keeps delivering, but cautiously, and caution looks like underspend. High-ticket purchases, qualified-lead events and budgets split across many ad sets all produce it.
Fix: consolidate ad sets that share an objective so their events pool together, or optimise for a more frequent step further up the funnel that still predicts value, as the learning phase guide explains.
Verify: the delivery column moves from learning limited to active over the following weeks, and spend becomes steadier day to day.
Prevent: before launching an ad set, estimate how many optimisation events its budget can buy in a week; if the answer is a handful, change the structure or the event before spending. Where the budget will stay small, the guide to running Meta Ads on a small budget turns that estimate into an operating plan with one ad set and an event the budget can feed.
Sibling ad sets reaching the same people
When two active ad sets are eligible for the same person, Meta enters only one of them in that auction. The loser is held back, so neither builds steady delivery and each spends a fraction of its budget. Duplicated ad sets, prospecting audiences that differ only slightly, and lookalikes of several tiers running side by side are the common sources.
Fix: compare the active audiences with the overlap tool, read the auction overlap figure in the Inspect tool where it appears, then merge ad sets that chase the same people or exclude one audience from the other, following the overlap rules in the exclusions and overlap guide. Where the campaign uses Advantage+ campaign budget, the help centre describes delivery moving spend between ad sets towards results, which makes consolidation easier to manage.
Verify: the surviving ad set spends more than the separate ones did together, and its combined reach no longer falls short of the sum of its parts.
Prevent: keep one place for each person in the funnel and re-run the overlap check whenever a new audience is added.
Placements trimmed to a few surfaces
The help centre on ad placements describes Advantage+ placements spreading budget across eligible surfaces, and notes that restricting placements reduces reach and can raise cost. An ad set limited to one feed or one format has a small auction pool; when that pool is thin on a given day, there is nowhere else for the money to go.
Fix: switch back to Advantage+ placements, or add every surface the creative can carry, and supply the vertical and square crops that the formats and placements guide describes so new surfaces receive a fitting version.
Verify: the placement breakdown shows spend arriving on the added surfaces, cost per thousand eases and total spend approaches the budget.
Prevent: restrict placements only for a stated reason, such as brand safety or a deliberate placement test, and record the reason in the ad set name.
Prevention: cost controls set from real costs
Underspend is a setup problem, so it is cheapest to prevent at launch. Before an ad set goes live, check five things against its budget. The bid strategy is the one the economics call for, and any goal or cap comes from the value of a result rather than a past average. The estimated audience is broad enough that the budget will not exhaust it in a few days. The optimisation event occurs often enough, at this budget, for delivery to learn. No other active ad set targets substantially the same people. Placements are open unless there is a written reason to close them.
After launch, read spend against budget weekly rather than daily, alongside the delivery status of every ad set. An ad set that has spent well short of its budget for a full week gets one change, chosen from the table above, and then a week to respond; stacking several edits at once restarts learning and hides which change mattered. Keep a short log of each cap, audience edit and placement change with its date, because the log is what turns the next underspend from a puzzle into a lookup. When the business wants more volume, treat it as a scaling decision with marginal cost in view, not as a reason to loosen every guardrail at once.
A furniture retailer's cost cap that strangled delivery
A furniture retailer launches a prospecting ad set with a £300 daily budget and a cost per result goal of £40, copied from the account's average cost per purchase the previous quarter. After seven days it has spent £610 in total, about £87 a day, and bought 17 purchases at £36 each. The audience is broad and placements are open, and the delivery column shows the ad set as active rather than learning limited, which points to the cap. The team works out that a sofa order is worth enough to pay up to £70 per purchase and still meet the margin target, and raises the goal to £65 in a single edit. Over the next week the ad set spends £1,940, about £277 a day, for 38 purchases at £51 each. Cost per purchase rose, as the scaling guide predicts for extra volume, but every purchase stays inside the margin and the budget is finally doing its job.
Common mistakes
Raising the budget when a cap is what binds. Copying a cost goal from the account average. Adding a fourth interest layer to "improve quality" and then wondering why reach stalls. Running near-identical ad sets side by side. Trimming placements by instinct. Making three edits on the same day and crediting the wrong one.
