First: is it real?
Before investigating delivery, establish that the conversions actually fell rather than the reporting of them. Three checks take minutes. Reporting lag: Meta attributes conversions to the day of the ad interaction, and conversions keep arriving for days after, so the most recent days always look worse than they will once the window fills. Compare like with like by excluding the most recent days or by reading a window that has closed. Measurement: open Events Manager and check that the optimisation event is still arriving at a normal rate, with its parameters, from both browser and server where both are set up. A site release, a consent banner update, a tag container publish or a platform upgrade can stop the pixel without anyone in marketing knowing. If events have stopped, the campaign has lost both its reporting and its optimisation signal, and the drop in reported results is the symptom of a measurement break, not a delivery problem. The pixel setup tutorial has the testing sequence to confirm each event. Attribution setting: check whether the ad set's attribution setting or the reporting column's comparison window changed. A move from a longer to a shorter window, or a reporting view set to a different AttributionAssigning conversion credit to touchpoints. setting from the one results were judged on, shrinks the number without any change in behaviour. Only when all three are clear is the drop a delivery problem.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Conversions fell but clicks and spend are normal | Events Manager: is the optimisation event still arriving from the site? | Pixel or server event broken by a site, tag or consent change | Repair the tag or integration per the pixel setup tutorial; expect a recovery period while delivery re-learns |
| Results fell on the same day something was edited | Change history for the ad set and campaign | A significant edit restarted the learning phase | Stop editing; wait for learning to complete; see the learning phase guide |
| Cost per thousand rose sharply, results fell in proportion | Compare cost per thousand by placement and over the same period last year | Auction shift: seasonal competition, a competitor's launch, or a placement change | Accept the new price, adjust the target or bid strategy, widen placements; see the scaling guide for bid controls |
| Click-through fell gradually on a long-running ad while frequency rose | Ad-level trend of frequency, click-through and cost per result against the ad's first week | Creative wear-out or audience saturation | Add new creative concepts or refresh the audience; see the frequency and fatigue guide |
| Spend dropped to a fraction of budget or stopped | Ad status, account quality page, payment settings, spending limits | Rejected ad, restricted account, failed payment or a reached spending limit | Resolve the rejection or restriction, fix payment, raise or remove the limit |
| Clicks are normal but conversion rate on the site collapsed | Load the landing page on a phone; test checkout; check stock and prices | Landing page or checkout broken, slow, out of stock, or changed | Fix the site; pause only the affected ads; re-check events after the fix |
| Results fell only in one placement, age group or country | Breakdowns by placement, age, region and device | A single segment saturated, a placement-specific creative problem, or a regional event | Exclude or re-creative the segment rather than editing the whole ad set |
| Reported results look worse but the business's own numbers are fine | Compare Meta's count with the shop or CRM over a closed window | Reporting lag or an attribution setting change, not a real drop | Wait for the window to close; restore the comparison setting; nothing to fix in delivery |
Measurement breaks
The most common cause of an apparent collapse is that the events stopped. Look at Events Manager for the dataset the ad set uses: the event count over time, the connection methods reporting, and the diagnostics tab. A browser event that fell while a server event held up points at the site or the consent layer; both falling together points at a platform or integration change. Check the site's change log and ask the developers what shipped. The Pixel Helper browser extension on the live site shows whether the base script loads and whether the event fires on the page it should. If a consent platform was updated, test the accepted branch; if a tag container was published, open its version history. Repair is usually quick, but delivery has been optimising on a starved signal for as long as the break lasted, and results take time to recover after the fix. Do not change the campaign in the meantime; the campaign was never the problem.
Learning resets
Meta documents that significant edits to an ad set restart the learning phase, during which delivery is less stable and cost per acquisition (CPACost of each conversion.) is typically higher. The edits listed include changes to targeting, placements, the optimisation event, creative, bid strategy and large budget changes, and a pause of extended length. Open change history for the ad set and campaign and line up every edit against the day results fell. An edit a day or two before a drop is the likely cause, including edits made by a colleague, an automated rule or a partner. The fix is patience: let learning complete, which Meta ties to a number of optimisation events in a set period, and judge the ad set only after it exits. The learning phase guide covers what counts as significant and which edits are safe. The trap here is compounding: an advertiser who sees the post-edit dip and edits again to fix it restarts learning again, and the ad set never stabilises.
Auction shifts
The auction is shared with every other advertiser wanting the same people, and its price moves without any change in your account. Cost per thousand impressions (CPMCost per 1,000 impressions.) is the indicator: if it rose sharply and results fell in proportion while click-through and conversion rate held, the campaign is paying more for the same behaviour. Compare the same weeks in the previous year for seasonality, check whether a placement was removed or a new one added, and look at the cost by placement and country breakdowns to see where the rise sits. Retail seasons, major events and competitor launches all show up this way. There is no fix in the sense of restoring the old price. The choices are to accept the new economics, to use a cost per result goal so delivery slows rather than overpays, to widen placements and audiences so delivery has cheaper options, or to reduce spend until the season passes. The CPA calculator makes it quick to work out what cost per result the new price implies at the current conversion rate. What moves cost per thousand in the first place, and how to read a cost change one link at a time, is set out in the guide to what drives the cost of Meta ads.
Fatigue
A gradual decline that was mistaken for a sudden one is usually fatigue: the winning ad wore out, or the audience was exhausted, and the drop became visible only when it crossed a threshold somebody was watching. The signature is at the ad level: FrequencyAverage times a person saw an ad. rising, click-through falling against the ad's own early weeks, cost per thousand drifting up, negative feedback growing, and sometimes a creative fatigue flag in the delivery column. Reach that has flattened points at saturation; reach still growing with falling click-through points at wear-out. Both are covered in detail, with the signs and the responses, in the frequency and fatigue guide. The fix is new creative, a refreshed or widened audience, or both, introduced alongside the incumbent rather than replacing it. The deeper fix is a creative pipeline, so that the next ad is ready before the current one tires.
Policy and account health
Delivery can be cut without any change in performance at all. Check ad status in Ads Manager for ads marked rejected or in review, and the account quality page for restrictions on the ad account, the Page or the Business portfolio. Meta's advertising standards are enforced by automated review as well as people, and an ad that ran for weeks can be rejected after a re-review or after a policy change. Also check the practical limits: a payment method that failed, an account spending limit that was reached, a campaign spending limit set long ago, or a scheduled end date. The symptom for all of these is spend falling to a fraction of budget, and the fix is administrative rather than optimisation: request review of a rejection with the ad edited if needed, resolve the restriction through the account quality page, update payment, and raise or remove the limit. Keep a note of which ads were rejected and why, so that future creative avoids the same trigger. The guide to ad review and rejected ads sets out that log, the choice between editing and appealing, and a pre-launch policy check.
Site changes
Finally, the website. If clicks are normal and the cost of a click is normal but conversion rate from click fell, the problem is after the click. Load the landing page on a phone over a mobile connection, walk the checkout or form, check that the products in the ads are in stock at the advertised price, and look for a redesign, a new pop-up, a broken redirect, a slower page or a changed form. Ask what shipped on the site in the days before the drop, and check the analytics tool for the same fall in on-site conversion across all channels; a drop that appears in organic and email traffic too is a site problem, not an ads problem. The fix belongs to whoever owns the site, and the ads' job is to stop sending people to a broken page until it is done. Once fixed, confirm the pixel events still fire on the changed pages, because a redesign that breaks checkout often breaks the tags with it.
Prevention
Most drops are cheaper to prevent than to diagnose, and prevention is mostly about seeing changes early.
- Alert on events, not only on results. A daily check that the optimisation event is arriving at a normal rate from Events Manager catches measurement breaks before they become performance drops.
- Be on the site's release list. Marketing should know when the website, the tag container or the consent banner is about to change, and should test the events after each release.
- Record every edit with a reason. Change history shows what; a note shows why, and both make the learning-reset check a minute's work.
- Set a review rhythm and keep to it. Daily for breakages, weekly for decisions; the optimisation hub sets it out. Reacting to single days causes more drops than it prevents.
- Keep a creative pipeline. Fatigue is predictable; a ready successor turns it from a drop into a handover.
- Know the seasonal auction. Note cost per thousand by week across the year so a seasonal rise is recognised rather than investigated.
- Audit account health monthly. Payment methods, spending limits, end dates, and the account quality page.
- Reconcile reporting regularly. A monthly comparison of Meta's results with the business's own numbers means an attribution or reporting change is spotted when it happens. The author's paid media analytics expertise page describes the reconciliation routine used across platforms.
When a drop has no single obvious cause, the Meta Ads account audit gives you a full review of structure, tracking, audiences and creative so nothing contributing to the decline is missed.
Common mistakes
- Editing before diagnosing. The reflex edit restarts learning and adds a second cause to the first.
- Reading yesterday. The most recent days are always incomplete.
- Looking only in Ads Manager. The cause is usually in Events Manager, the site, or the account settings.
- Blaming the algorithm. Delivery responds to signal, auction and creative; a drop has a cause in one of them.
- Pausing to think. An extended pause can itself restart learning.
- Fixing the measurement and expecting instant recovery. Delivery needs time to re-learn on restored signal.
- Not telling anyone. A drop caused by a site release will recur at the next release unless the process changes.
- Checking one cause and stopping. A measurement break and a tired ad can arrive in the same week; run the whole sequence even after the first finding, or the second cause surfaces as a new mystery a fortnight later.
