The swing nobody can explain
An account that moves like that is usually not badly targeted. It is badly divided. Somewhere along the way it grew into twelve, twenty or forty campaigns, each built to isolate something: a single keyword, one match type, desktop kept away from mobile, a small radius with its own budget. Every split felt like control at the time.
Automated bidding does not read your account as a whole. It decides campaign by campaign, with only the conversions that campaign has, and that is where fragmentation stops being a reporting preference and starts costing money. A campaign with almost no conversion history cannot behave calmly, because it has nothing to be calm about.
Sixty conversions, twelve ways
Bidding algorithms typically need at least thirty conversions a month at the campaign level before they can evaluate contextual signals effectively. That floor is per campaign, not per account. Spread sixty monthly conversions across twelve thin campaigns and each one is working with five sales, which is nowhere near enough to tell a genuine trend from random variation.
That is why the behavior looks irrational. In a low volume campaign a single conversion makes the algorithm overbid wildly, and one quiet day chokes off traffic entirely. Those campaigns stay trapped in perpetual learning phases and never reach a stable baseline you could judge. And the dense cross negation keeping those keywords isolated blocks the long tail queries that convert best.
The splits still worth keeping
Consolidating is not flattening the account into one campaign. Features like AI Max already absorb much of the manual work those splits used to serve, so granular control matters less. A split still earns its place when it carries a real business difference: lead generation apart from ecommerce because the objective and the bidding differ, a service with its own profit margin that needs its own target cost per acquisition, a region that needs a dedicated budget.
Intent inside the funnel counts too. A demo request and a trial signup ask very different things of the person, and keeping them apart protects relevance. The test is short. If the split changes your strategy, your budget or what the customer experiences, it stays. If it only changes how the same journey gets labeled in a report, it merges.
What changes in the week
Whoever runs the account can do the first pass in an afternoon: list every campaign, mark the ones built around a single keyword, a duplicated match type or a device, and write down how many conversions each produced last month. The ones sitting at two or five are starving the system. Those merge into the campaign that shares their objective, their margin and their landing experience.
Start with one service line so you have the rest of the account to compare against. What you watch for in that first week is the daily numbers swinging less, and budget no longer sitting idle in a quiet segment while a busy one caps out early. You stop arguing over which keyword to pause because of one bad Tuesday and start talking about the offer and the ads, which is where the growth actually lives.