The message usually arrives on a Monday. "Something's wrong — the ads have stopped working." No numbers, no window, no definition of "working." And whatever you say in the next hour sets the tone for the rest of the relationship.
Before you explain anything, establish two things: whether performance actually fell, and whether the fall is a cliff or a slide. A cliff almost always means something changed — a tag, a page, a policy, a setting, a budget. A slide almost always means saturation, competition, or fatigue. Those two families have completely different fixes, and guessing between them in front of a client is how a fixable problem becomes a lost account.
The temptation is to reassure first and investigate later. Resist it. A confident wrong diagnosis on Monday is much harder to walk back on Thursday than an honest "give me until this afternoon."
Step one: confirm the drop is real
A meaningful share of "the ads stopped working" reports are measurement failures, not performance failures. The campaign is still producing business; the dashboard has stopped seeing it. Checking this first costs an hour and occasionally saves a month of pointless optimisation.
Three checks, in order of how often they pay off:
- Are the conversion events still firing? A site migration, a new consent banner, a tag manager edit, a theme update, or a redesigned checkout can silently break tracking. The signature is unmistakable: reported conversions fall off a shelf on a specific date while spend, clicks, and on-site behaviour look normal.
- Does the platform agree with the business? Compare reported conversions against the client's actual orders, bookings, or CRM entries for the same period. If the CRM is steady and the ad account is empty, you have a tracking problem, not a media problem.
- Did the reporting window or attribution setting change? Attribution windows, view-through settings, and modelled conversions all move numbers without anything real changing. Ad platforms revise these behaviours regularly, so check the platform's current documentation rather than assuming the setting you configured last year still means the same thing.
This is also where you find the opposite problem: performance that never was. If a channel's reported returns were always inflated by taking credit for buyers who would have purchased anyway, the "drop" may be the moment the accounting got more honest. Our marketing analytics and ROI guide covers measuring incrementality — the sales your advertising actually caused — which is the only way to settle that argument with evidence instead of opinion.
Step two: read the shape of the decline
Plot the metric over ninety days, not seven. The shape tells you which family of causes to investigate, and it narrows the search dramatically.
A cliff — performance falls on one identifiable day
Something changed. Human, mechanical, or external, but a discrete event. Candidates worth checking in the first pass: an ad or account disapproval, a landing page or site change, a payment or billing failure, a budget or bid edit, a campaign restructure that reset learning, a tracking break, or a competitor's promotion entering the auction. Find the date first, then look for what shares it.
A slide — performance decays gradually over weeks
This is the exhaustion family. Creative fatigue, where the same audience has seen the same assets too many times and stops responding. Audience saturation, where you've reached most of the people worth reaching in a finite market. Or steadily rising competition pushing costs up while your conversion rate stays flat. Slides rarely have a single cause you can point at, which makes them harder to explain and harder to fix — but they are also predictable and preventable.
A sawtooth — results swing wildly week to week
Often nothing is wrong. Low conversion volume produces noisy weekly numbers, and a client comparing last week against a good week will read random variation as collapse. The honest answer here is uncomfortable but correct: the account does not generate enough conversions per week to support week-level conclusions. Move the reporting cadence to a window that carries enough volume to mean something.
Step three: build a change log before you build a theory
The most useful artefact in this situation is boring: a dated list of everything that changed in and around the campaign. Ad edits, budget changes, new creative, audience changes, landing page deployments, pricing updates, stock outages, promotions ending, the client's own email or organic activity, seasonality, and platform announcements.
Overlay that list on the performance chart. The cause is very often sitting on the date of the drop, and it is very often something nobody thought to mention — a developer pushed a change, a promotion ended, a product went out of stock, the client's sales team stopped answering the phone at 5pm.
This matters beyond the diagnosis. Walking a client through a dated change log turns a defensive conversation into a joint investigation. You are not making a case; you are showing your work.
Step four: answer the client before you have the full answer
You will usually have to say something before the diagnosis is complete. What works:
- Restate the problem in numbers. "Cost per lead moved from roughly X to roughly Y starting the week of the 14th" is a shared fact. "Things are down" is a mood.
- Say what you have ruled out. Ruling out tracking, disapprovals, and billing in the first hour is real progress, and it demonstrates method rather than panic.
- Give the two most likely causes and how you'll tell them apart. Clients tolerate uncertainty far better than they tolerate vagueness.
- Commit to a time, not a result. "I'll have the diagnosis by Wednesday" is a promise you can keep. "It'll be back to normal next week" usually isn't.
- Name what you need from them. Site changes, pricing changes, stock, sales capacity, and competitor activity all live on the client's side of the wall, and you cannot see them from the ad account.
Two things to avoid. Do not blame the algorithm — it is technically sometimes true and it always sounds like an excuse. And do not fix five things at once to look busy: simultaneous changes to creative, targeting, and bids guarantee that nobody will ever know what worked, including you.
When the honest answer is uncomfortable
Sometimes the diagnosis is that the campaign didn't break — the conditions changed. The addressable audience is exhausted at the current offer. A competitor with deeper pockets entered the auction. The product's price moved and the economics no longer support the target cost per acquisition. Demand is seasonal and this is the trough.
These conversations are the ones that decide whether a relationship survives. The version that works states the mechanism plainly, names what can and cannot be changed with media alone, and puts a decision in front of the client: widen the audience, change the offer, accept a higher cost per acquisition, or reallocate the budget. What destroys trust is not bad news; it is bad news that arrives three months late, after the budget was spent optimising around a problem that media could never solve.
Preventing the next one
Most of these fires are preventable with unglamorous habits. Monitor tracking health as a metric in its own right so a broken tag surfaces in hours rather than at month end. Watch frequency and creative age so fatigue is scheduled for rather than discovered. Keep the change log continuously instead of reconstructing it under pressure. Agree in advance on the reporting window and the metric that defines success, so "working" has a shared definition before anyone needs to argue about it. And set a review cadence that matches the account's conversion volume, which stops weekly noise from generating monthly emergencies.
FAQ
Why did my ads stop working overnight?
A same-day collapse points to a discrete event rather than gradual decay. Check, in this order: whether conversion tracking is still firing, whether ads or the account were disapproved, whether billing failed, whether the landing page or site changed, and whether budgets, bids, or campaign structure were edited. Sudden means something changed; find the date and work outward from it.
My ads still get clicks but no conversions — what changed?
The ad is still doing its job and the destination or the measurement isn't. Verify the conversion event fires, then check the landing page for a change: a new layout, a slower page, a longer form, a price increase, an out-of-stock product, or a message that no longer matches the ad's promise. Click-through holding steady while conversions fall almost always points past the ad.
How long should ads run before I decide they've stopped working?
Long enough to accumulate a meaningful number of conversions, not a meaningful number of days. Accounts with few weekly conversions cannot support weekly judgements — the swings are noise. Choose a window that reliably contains enough conversions to notice a real change, and hold to it in both good weeks and bad.
Can ad fatigue really kill a campaign that worked for months?
Yes, and it is one of the most common causes of a gradual slide. As the same audience sees the same creative repeatedly, response rates fall and cost per result climbs, even though nothing in the setup changed. The tell is rising frequency alongside a falling click-through rate. Raising bids treats the symptom; refreshing what people see first treats the cause.
Should I rebuild the campaign or fix the existing one?
Fix first when the symptom points at something specific — a broken tag, a page change, tired creative. Rebuild only when the structure itself is the problem, and understand the cost: a rebuild resets learning, discards historical signal, and destroys the comparison you would have used to prove what actually went wrong.
Next step
The agencies that survive this conversation are not the ones with the fastest answer. They are the ones who can show, on a dated chart, what changed and when — and who separate a real performance drop from a reporting artefact before saying a word. Verify the measurement, read the shape of the decline, build the change log, then explain the mechanism honestly and change one thing at a time.
If you are on the client side of this conversation and no longer confident you're getting straight answers, find and compare agencies by specialty on AgencyList.
