Performance dropped: a systematic method for finding out why.
The instinct when numbers fall is to start adjusting the account. The discipline is to diagnose in the right order first, because half the time the account is not the problem.
Sooner or later every account produces the same Monday morning: conversions are down, ROAS has sagged, and someone important wants to know why. What happens next separates disciplined teams from anxious ones. The anxious response is to start changing things (bids, budgets, targets, structure), which contaminates the evidence and resets algorithmic learning before anyone knows what the problem was. The disciplined response is to diagnose before touching anything, in a fixed order that starts as far away from the ad account as possible.
The order matters because of a simple asymmetry: the causes people check last are the causes that turn out to be responsible most often. Teams instinctively start inside the account, because that is the thing they control. But the account is the least likely culprit for a sudden drop, and the most expensive place to make speculative changes.
First: rule out measurement
A large share of sudden performance drops are not performance drops at all. They are measurement breaks wearing the costume of one. A tag removed in a site release, a consent banner update that changed opt-in rates, a conversion action edited by someone with access, an attribution window quietly reclassifying when conversions get counted. Real demand does not usually fall off a cliff overnight. Tracking does.
- Check whether actual business outcomes (orders in the store, leads in the CRM) fell in step with reported conversions. If the business is fine and the dashboard is not, it is measurement
- Look for site deployments, tag manager changes, and consent banner updates on or just before the drop date
- Compare browser-reported and server-reported conversions if both exist; divergence points straight at the broken layer
- Review conversion action change history — settings edits leave fingerprints
Never optimize an account against a broken measurement layer. Every change made while tracking is wrong teaches the bidding algorithm something false, and you will pay for the lesson twice.
Second: check the market
If measurement is intact, look outward before looking inward. Demand itself moves: seasonality, weather, news cycles, payday timing. Competitors move too: a rival launching a sale or entering your auctions shows up as rising CPCs and falling impression share while your account sits untouched. Price and stock changes on your own site belong in this category as well; media buyers are routinely the last to learn that the hero product went out of stock or lost its promotion.
Third: platform mechanics
Next, the machinery between you and the customer. Platforms ship auction changes, policy updates, and delivery shifts constantly, and automated bidding occasionally walks itself into a corner: a learning phase reset by an earlier edit, a target that has drifted out of reach and throttled delivery, an ad disapproval that silently removed your best performer from rotation. This layer is checked third because it is noisier than the first two: platform fluctuations are constant, so they explain less than people want them to.
Fourth, and only now: the account
Only after the first three layers are cleared do we open the change history and look at what we did. Recent edits, budget moves, audience changes, new creative, and structural work, all mapped against the drop date. If the timeline lines up, revert the suspect change and observe, one variable at a time. The temptation to fix five things at once is exactly how accounts end up unable to attribute the recovery, which guarantees the same panic next quarter.
A worked example, with the numbers
The method reads as obvious and is almost never followed, so it is worth walking one all the way through. The figures below are illustrative rather than a specific client account, constructed to be typical of a mid-market ecommerce advertiser, but the shape of this diagnosis is one we have run many times and the resolution is the most common one.
Monday morning. Last week against the week before, on a store spending roughly $41,000 a week.
- Spend: $41,200 to $40,800. Flat.
- Impressions and clicks: 18,400 clicks to 18,100. Flat.
- Average CPC: $2.24 to $2.25. Flat.
- Reported conversions: 612 to 404. Down 34%.
- Reported revenue: $196,000 to $131,300. Down 33%.
- Reported ROAS: 4.76x to 3.22x.
The instinctive reading is that the account has stopped converting and the targets need tightening. Notice what the secondary numbers already rule out, though. Auction pressure raises CPC and costs you clicks; both are flat. A budget or delivery problem shows up in impressions; they are flat too. Traffic arrived in the same volume at the same price and produced a third fewer recorded conversions. That is not a demand pattern.
So layer one, measurement, which takes about ten minutes. Orders recorded in the commerce platform: 618 the week before, 601 last week. Down 3%. The business is essentially fine. The dashboard is not.
The single most useful number in this diagnosis is the coverage ratio: recorded conversions divided by real orders. It ran at 612 of 618, about 99%, then fell to 404 of 601, about 67%. A third of the account’s conversions stopped being seen. No campaign metric would have told you that, because every campaign metric is computed from the broken number.
From there the cause is a short search rather than an investigation. Site deployments in the window: one, on the Wednesday, a checkout redesign. The new flow completes on a different confirmation route, and the conversion tag was bound to the old URL pattern. Orders still routed through express checkout kept firing, which is why the number fell by a third rather than to zero, and a partial break is far more dangerous than a total one because it looks survivable.
Total diagnosis time, roughly an hour, because the first question was asked of the order table rather than of the campaign view. Nothing in the account was touched, which matters for the counterfactual below.
What the panic response would have cost
Work through the same Monday with the common response, which is to protect the ROAS number by tightening targets. Cutting the target by 30% to defend a metric that is only broken in the reporting has three separate costs, and they compound in a specific order.
- Real revenue is lost immediately. Tighter targets restrict delivery, so a business genuinely running at around $196,000 a week gives up volume it was profitably buying. Even a modest 20% delivery reduction is a five-figure weekly cost against a problem that did not exist.
- The learning phase resets, so the account spends the next two to three weeks re-learning, during which every subsequent reading is noisier and harder to diagnose.
- The reporting still looks broken, because the tag is still broken. ROAS has not recovered, so the obvious inference is that the first cut was not deep enough. This is the second cut, and it is where accounts genuinely spiral.
The spiral is not caused by incompetence. It is caused by responding to a number before establishing what the number measures, and each response degrades the evidence available for the next one. Two weeks in, the change history is dense enough that the original break is no longer visible in it.
Reading the signature: same symptom, different cause
A 34% conversion drop is a symptom, not a diagnosis, and the secondary metrics carry the signature that identifies the layer. Take the same headline and change what sits underneath it.
- Conversions down, clicks and CPC flat, real orders flat: measurement. Check deployments, tag changes, consent banner updates, and conversion action edit history. This is the most common single cause of a sudden drop.
- Conversions down, clicks down, CPC up, impression share lost to rank: auction pressure. A competitor entered or raised bids. Your account did not change and mostly cannot fix this by changing.
- Conversions down, clicks flat, real orders down in step, on-site conversion rate down: the offer or the site. Stock outage on a hero product, a promotion that ended, a price change, a page performance regression. Media buyers are routinely the last to hear about all four.
- Conversions down, impressions down sharply, budget under-delivering: delivery mechanics. A learning reset from an earlier edit, a target drifted out of reach, an ad disapproval that quietly removed your best performer, or a policy action.
- Conversions down on one platform only, while the others hold: platform-specific. Almost always tracking for that platform, a disapproval, or a policy change, and almost never a real shift in demand, because demand does not respect platform boundaries.
Reading the signature before opening the account is the entire discipline, and it takes minutes. The reason teams skip it is not ignorance of the method. It is that someone senior is waiting, and producing a change feels more responsive than producing a diagnosis.
The mentality underneath the method
The sequence is fixed precisely so that judgment is spent on evidence, not on deciding where to look. It removes the panic from the room: everyone knows the next step is a check, not a gamble. And it protects the account from its own operators, because the most damaging thing that can happen after a drop is rarely the drop itself. It is the flurry of unrecorded, unmeasured changes made in response. Diagnose in order. Change one thing. Write down what you learned. That is the whole method, and it is more reliable than improvisation when the pressure is on.
It is also a fair question to ask of whoever runs your account, because the answer is diagnostic in itself. Ask what they checked first the last time performance fell. An answer that starts inside the campaign view tells you something about how the next drop will be handled, and it is one of the more revealing questions to put to a prospective agency or specialist.
Why did my Google Ads performance suddenly drop?
Work outward rather than inward, because the causes teams check last are the ones most often responsible. Start with measurement: compare recorded conversions against real orders or leads in your own system for the same period, since a large share of sudden drops are tracking breaks rather than performance changes. Then check the market for competitor and seasonal movement, then platform delivery mechanics, and only then your own recent account changes. Changing bids before completing that sequence contaminates the evidence and resets algorithmic learning.
How can I tell a tracking problem from a real performance problem?
Compare the reported conversions against the outcomes your business actually recorded, orders in the commerce platform or closed leads in the CRM, over the same dates. If the business held steady while reported conversions fell, it is measurement. The most useful ongoing version of this check is a coverage ratio: recorded conversions divided by real orders, tracked over time. That ratio is stable when tracking is healthy, so a step change in it identifies a break within a day rather than a quarter.
Should I pause campaigns when performance drops suddenly?
Not before you know what happened, and pausing is rarely the right first move even afterwards. If the drop is a tracking break, pausing sacrifices real revenue against a problem that exists only in the reporting. If it is competitor pressure, pausing concedes the auction. If it is a delivery or learning issue, pausing makes it worse by resetting learning again. Diagnosis takes hours; a poorly reasoned pause costs weeks of learning plus the revenue foregone during it.
How long should I wait before reacting to a performance drop?
Diagnose immediately, act deliberately. The diagnostic sequence costs an hour or two and should start the same day, because evidence degrades as more changes accumulate. Acting is different: a single bad day is usually noise, and daily conversion counts fluctuate enough that reacting to one is how accounts acquire a history of unattributable changes. Look at a rolling seven-day window against the prior period before making a change, unless the diagnosis has already found a specific break.
What should I check first when conversions fall?
Whether the conversions actually fell. Open your commerce platform or CRM and count the real orders or leads for the affected dates, then compare that with what the ad platform reported. This takes minutes, requires no access to the account, and resolves the most common cause outright. Every other check, competitor activity, seasonality, delivery mechanics, account changes, is only interpretable once you know whether you are diagnosing a business event or a measurement event.
Written by Sam Nouri, founder, adsrunner. If this resonated and you want to apply it to your own account, you can book a strategy call or run a free audit.
How we research, source figures, and handle corrections: editorial policy.