On August 17, Google flipped a quiet switch across Search, Shopping, PMax, Display and Demand Gen: if your Target CPA is $10 but Smart Bidding has been quietly delivering $5 for months, the system will now stop rewarding that overperformance and steer you back to the $10 you typed in. Google calls this "target enforcement." Advertisers are calling it losing half their volume with no warning. Google shipped a Bid Target Adjustment Tool back in July specifically to soften this landing, and most agency account chains never opened it.

Both sides are grabbing the wrong end of the story. The real question isn't why the algorithm changed. It's who set that number, when, and whether anyone in the building still remembers doing it.

Why did nobody touch that target in two years?

Because nobody owned it. One auditor's account-review data puts the share of Google Ads accounts running "limited by budget" at over 80%, meaning the system knows it could deliver more conversions than the budget allows, and it's been quietly finding efficiency within that ceiling instead. In most of those accounts, Target CPA is a field typed once at campaign launch and never reopened; the planner who set it may have left the agency two account managers ago. Decades of goal-setting research from Edwin Locke and Gary Latham converge on one point: a target's power to shape behavior depends on whether someone actually owns it. A number you didn't set, and haven't thought about since, doesn't motivate anything, it just sits there. A study published earlier this year gave this exact failure mode a name: optimized-but-unowned goals, whether AI-authored or simply typed once and forgotten, quietly lose their grip on outcomes instead of driving them. For two years, Smart Bidding was politely ignoring your stale $10 and chasing the real opportunity. It just stopped extending that courtesy, the mirror image of what happened at Wells Fargo, where an unowned "eight products per customer" target pushed people toward fraud to hit it. Here, an unowned number is doing the opposite: quietly costing you the upside you'd already earned.

Is this a bug fix or a revenue lever?

It can be both, the math doesn't care about intent. Take a campaign with a $10 Target CPA and a $5 real CPA: on a $50,000 monthly budget, you were getting 10,000 conversions. Pin that campaign to the literal $10 target and the same budget buys 5,000, a 50% volume cut with zero extra spend. Advertisers read that as a price hike because on the results sheet, it functions exactly like one: same dollars, half the return. Google's side has a fair point too, a system quietly overperforming its stated target produces unpredictable swings the moment a budget gets raised. The timing isn't neutral either: this lands in the same year Meta is projected to pass Google as the world's largest ad seller, and a bidding engine that behaves predictably is easier to defend to procurement and auditors than one that quietly beats its own stated numbers. Both things are true. The party actually losing is whoever hasn't opened that account in two years.

• Who last touched your Target CPA / Target ROAS, has it moved since the campaign launched?
• For campaigns flagged "limited by budget," how far below the written target is real CPA actually running?
• Did you run Google's Bid Target Adjustment Tool, rolled out in July for exactly this, or did August 17 pass without anyone noticing?

The algorithm never beat you. You beat yourself with a number you wrote two years ago and forgot, the algorithm is just collecting on it now.

What changes before your next budget cycle

Treat Target CPA and Target ROAS as live inputs reviewed monthly, not constants set once at launch and left alone, the same discipline you'd apply to channel mix in a media plan. Before you lock October's budget, put every major campaign's trailing-90-day real CPA next to its written target; if the gap is over 20%, updating the number is cheaper than letting the algorithm update it for you. If this audit isn't already on your quarterly calendar, that's not a tooling gap, it's an ownership gap. Assign October a name before November sends the bill, and make sure that name isn't someone who left the account three quarters ago.