The most dangerous Google Ads account I see is not the one that is failing. It is the one hitting its target every month while the business has not grown in a year.
Everyone in the room is happy with the dashboard. ROAS on target, CPA on target, conversions up. And then the CFO asks why revenue is flat, and nobody has a good answer.
The target field is an instruction
Smart Bidding does exactly what it is told. It finds the auctions most likely to deliver the target CPA or target ROAS you set, and it avoids the rest.
That behaviour got stricter this year. Google said its August 17 bidding update pulls budget-constrained campaigns using target CPA and target ROAS closer to their stated targets, instead of letting them outperform. In plain terms, the number in that field is no longer a suggestion. It is an order, and the system now follows it more faithfully than before.
If the order is outdated, the algorithm will execute an outdated strategy with perfect discipline. That is the problem in one sentence.
Where the target came from
Ask any team where their current target came from. The answers are almost always the same.
Last year's performance. An industry benchmark. A number that felt safe when the account was set up.
Almost nobody revisits it when the business changes. Cost of goods moves, shipping rates move, product mix moves, discounts change. The target stays.
The arithmetic is unforgiving. A 4.0x ROAS on a product line with a 20% margin loses money on every sale once you count the media cost. The same 4.0x on a 60% margin line is very healthy. The ad account reports both as identical wins.
At difrnt., the first thing we ask in any paid media audit is who set the target and what margin it assumed. More often than not, nobody in the room can connect the number to current profit. I wrote about why ROAS is the wrong metric for AI-run auctions, and this is the operational reality behind that argument.
The easiest customer is the one who was coming anyway
Smart Bidding favours the people most likely to convert. Those are returning customers, people who already know the brand and people searching your name. They convert faster and cheaper than cold prospects, so the budget drifts toward them.
The result looks brilliant on paper. Branded search and retargeting post the best ROAS in the portfolio. But a large share of that revenue would have arrived anyway, through organic results, direct visits or an email.
When those campaigns take full credit, reported ROAS climbs while the customer base does not expand. You are paying to intercept your own customers on the way to your own checkout.
The opposite move is the one that grows a business, and it looks worse on the dashboard. Strict efficiency targets push the algorithm out of the more expensive auctions, which is often where the new customers are. Loosen the target and efficiency drops while reach grows. That trade is a business decision, and the platform will never make it for you.
Leads are not revenue
Lead generation and B2B accounts have their own version of this. The account optimises for form fills, booked calls or downloads. Those are reasonable signals, but they are not sales.
Over time, the algorithm learns to find people who fill in forms, not people who buy. Lead quality drifts, the conversion count rises and the sales team quietly stops calling the leads back.
The fix is a feedback loop. If closed deals and their value never make it back into Google Ads, the system has no way to know which leads mattered. It can only count submissions. Importing offline conversions is not glamorous work, but it is the difference between optimising for activity and optimising for revenue.
Some of the leak is not in the account at all. Leads sit untouched for three days, orders arrive for products that are out of stock, onboarding stalls. Google Ads can only deliver demand to the front door, and no bid strategy can fix what happens after it.
The questions for the next budget meeting
I would bring a short set of questions to the next review, and I would invite finance into the room.
When was the current target last set, and which margin did it assume? What share of conversions comes from new customers versus returning ones, and is that share growing or shrinking? How much reported revenue sits in branded campaigns, and when non-brand spend rose in the past, did total revenue rise with it?
Then the harder ones. Is the primary conversion a real sale or something that might become one? Are campaigns that hit their targets losing impression share because of the target itself? And how fast does the business follow up on what the account delivers?
None of these questions appear in the platform's recommendations tab. The platform reports whether it hit the instruction you gave it. It does not ask whether the instruction still makes sense. If you are testing broader automation at the same time, the discipline I described in running proper Google Ads AI Max experiments applies here too: change one variable, measure the business result, not the platform one.
The algorithm is not lying to you. It is answering the question you asked. Make sure it is still the right question.