Performance Max suddenly performing badly: a diagnosis in fixed order

Touch the campaign last

When Performance Max suddenly performs worse, the reflex is to turn something in the campaign: lower the target ROAS, rebuild asset groups, move budget. I think that is the wrong first move. The causes that do not resolve themselves lie outside the campaign, and every intervention in the campaign creates a new adjustment period that masks exactly those causes.

According to Google, Smart Bidding needs one to two conversion cycles to adjust to a new target. After that you should evaluate over two conversion cycles or two weeks, whichever is longer. If you turn the target twice in one week, you then see a mix of your own change and the actual cause and can no longer separate the two.

This gives my order of checks. It follows two questions: what does not heal by itself, and what distorts the signal the algorithm bids on?

1. Are conversions still arriving correctly?

Tracking comes first because wrong data steers every bidding decision. A campaign with a halved conversion signal bids less because it believes clicks are worth less. In the account this looks like a performance drop but is a measurement drop.

Google itself says that sudden changes in tag firing directly affect delivery. So first check whether the number of conversions in the account matches the number of orders in the shop system, split by tag. A shop update, a new cookie banner or a rebuilt checkout is enough for a break. How to reconcile this systematically is covered in checking conversion data.

Once the break is found, repairing is not enough. The wrong days stay in the bidding data. For this there are data exclusions: they also apply to Performance Max and change only what Smart Bidding calculates with, not your reporting. According to Google, at least 90 percent of the affected clicks should be excluded, and the period must cover the conversion delay. If a week or more was affected, the fluctuations can still last another one to two conversion cycles.

2. Are bestsellers missing from the inventory?

Performance Max cannot promote a disapproved product, however well the campaign is set up. In Merchant Center the statuses are "Under review", "Processing", "Approved", "Limited" and "Not approved". Disapprovals arise from policy or data quality violations, and the diagnosis is in the "Needs attention" section.

The total number of disapproved products says little. Count revenue, not items: if five products with a third of the previous PMax revenue are disapproved or limited in important countries, that explains the drop better than any campaign setting. Sort the product list by revenue in the weeks before the drop and check the status of the top rows. What makes a feed robust against such failures is covered in what makes a good product feed.

3. Has the shop become more expensive than the market?

A campaign cannot optimise away a price disadvantage. If a competitor offers the same item cheaper, click-through rate and conversion rate fall, and the bidding system withdraws from auctions. In the account this looks like a campaign problem.

The Pricing tab in Merchant Center Analytics compares your prices with a benchmark of other retailers. It needs valid GTINs and shows only current prices. A price history only exists if you save the data regularly through the Merchant API or BigQuery. The BigQuery export provides a daily benchmark per country and product for this: the click-weighted average price of all retailers advertising the same product. Without this history it is hard to prove afterwards that the market moved and not your campaign. I would therefore set it up before the next drop comes. How price gap and reach relate is explained in price and reach in Shopping.

For the decision, contribution margin counts here, not ROAS. A price cut can raise ROAS and still eat up the profit per order. That is a decision for purchasing and management, not a campaign setting.

4. Has demand fallen?

According to Google, performance fluctuations are sometimes caused by external market shifts and not by campaign settings. Google points to Google Trends for this. In addition I would look at the same period of the previous year, provided the range is comparable.

If demand falls, the problem resolves with demand. A campaign that spends less when interest declines behaves correctly. Rebuilding the campaign changes nothing about falling demand, it only costs convergence time that you lack when demand returns.

5. Only now: what has changed in the campaign?

If the four outer points are clean, the change history is next. It shows changes from the last two years on a timeline next to impressions, clicks, conversions and cost and names in the user column who made the change, including API tools. Read it like this: fix the day of the drop and look for the last change before it. If it falls within one conversion cycle before the drop, it is the first suspect.

Two limits belong here. According to Google, account settings or changes made by Google staff during consultations are sometimes missing. Changes in Merchant Center and in the shop are not listed there anyway, which is why they come earlier in this order.

A target change is typical. According to Google, a lowered target ROAS or pausing strong ads can shift budget into other channels such as Discover. Whether that has happened is shown by the channel report. For context: budget, target, creative and targeting changes never trigger the "Learning" status, according to Google. The absence of the status therefore does not mean the campaign is not currently re-steering.

When waiting is right

Waiting is a decision, not helplessness. My rule: waiting is right if tracking, feed and price are clean and either demand has fallen or your last own change is younger than two conversion cycles. Otherwise waiting does not help, because a tracking break, disapproved bestsellers or a price disadvantage do not resolve themselves.

Suppose a hypothetical shop has a conversion cycle of ten days, so about ten days pass between click and the bulk of orders. On Monday the target ROAS was changed, and on Thursday the week looks weak. According to Google, the adjustment takes up to twenty days, and evaluation follows over twenty days. If you intervene again on Thursday, you reset the clock. Google expressly advises avoiding several target changes within one conversion cycle.

On top of that comes the delay itself: the last days always look worse in the account than they will end up, because orders are still coming in. More on this in conversion delay in Google Ads. For the retailer, what counts during the wait is contribution margin over the whole period, not the daily ROAS.

When the order does not apply

The order assumes that no obvious cause is known. If someone halved the budget yesterday and revenue falls today, you need no diagnosis. The same goes for very small campaigns: according to Google, Performance Max needs at least 15 conversions in 30 days for stable performance. Below that, "suddenly bad" is often just random fluctuation, and a single weak day justifies no check, let alone a rebuild.

Check catalogues such as the one from Optmyzr are useful as a reference if the five steps produce no finding. As a starting point I prefer a fixed order.