Smart Bidding Does What You Told It. That Is the Problem.

Smart bidding gets blamed for a lot of outcomes it was explicitly instructed to produce. An account sets a target ROAS of 800 percent because that was last year's best month, volume dries up, and the algorithm takes the blame for obeying. Another sets a target CPA at half the true break-even out of ambition, the system dutifully finds only the cheapest and worst conversions, and the algorithm takes the blame again. The machine did what it was told. The telling was the mistake.

It helps to be literal about what a bid target actually does, because the interface hides it. A target is not a goal the system tries hard to reach. It is a filter on which auctions you are permitted to win. Set an aggressive CPA target and you have not asked the algorithm to work harder; you have instructed it to abandon every auction where the likely converter costs more than your number, which is frequently every auction containing your best customers. The expensive prospects, the high-intent competitive queries, the buyers researching a considered purchase, all of them live in auctions your target just forbade. Volume does not dry up because the algorithm failed. It dries up because you fenced it out of the market where your volume lives.

Targets come from economics or they come from politics

A bid target is the single most consequential number in the account, the dial that decides which auctions you are allowed to win, and it deserves to come from economics rather than aspiration. The right question is not what CPA would look good on a slide. It is what a customer is worth, at margin, over a realistic horizon, and what acquisition cost that worth can carry. That calculation takes an afternoon with someone from finance, and in most accounts I audit it has never been performed; the target in the system traces back to a predecessor's spreadsheet, a benchmark from a conference deck, or a number the room felt comfortable with. Set the target from the math and the algorithm becomes your buying agent, executing your economics at a scale no human trader could. Set it from politics and the algorithm becomes a mirror for the politics, which is how plans built before anyone knew the margin fail in automated form, faster and at scale.

Here is the part that never makes it into the platform's best-practices documentation: the target is also where the incentive conflict lives. Google earns more when you win more auctions, so every native recommendation, raise your target, remove the constraint, switch to maximize conversions, points the same direction. Some of those recommendations are right for you. All of them are right for Google. The only defense is owning the math yourself, because the counterparty offering to set your number has a position in the trade.

Two habits, one diagnostic

Two operating habits separate the accounts that make this work. First, they move targets deliberately and gently, because a target is a constraint on an auction system, and yanking it twenty percent overnight tells the system to abandon its learned territory and start over; small steps, ten percent or less, held two weeks or more, long enough to read the response before the next move. Second, they revisit quarterly, since break-even math drifts with pricing, product mix, and season, and a target set in January quietly becomes fiction by August. The quarterly review is thirty minutes: has the margin changed, has the customer value changed, does the number in the machine still describe the business. Most quarters the answer is no change. The quarters where it is not pay for the habit ten times over.

And when volume collapses or quality craters, run the diagnostic in the right order: check the instruction before blaming the instrument. Pull the target, trace it to its source, and ask whether anyone can produce the economics behind it. In my experience the ratio runs about nine to one: for every genuine algorithm failure, nine targets were set by someone who wanted a number rather than calculated one. Smart bidding is a very fast, very literal employee who never questions an order. That is enormously valuable, and it means every order you give is a specification, not a wish. Write them accordingly.