Performance dropped: a systematic method for finding out why.

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

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, structural work — 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.

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 outperforms improvisation every single time.