Smart Bidding: what the machine optimises

Smart Bidding sets a bid for every individual auction using signals you cannot see, aiming at a goal you set. Understanding what it is optimising — and what it does when it cannot — is the difference between steering it and being surprised by it.

The strategies

• Maximize conversions — as many as possible for the budget. No efficiency constraint.
• Maximize conversion value — as much value as possible for the budget.
• Target CPA (tCPA) — as many conversions as possible at a cost per acquisition.
• Target ROAS (tROAS) — as much value as possible at a return on ad spend.

The targeted strategies add a constraint. Without a target, the system spends the budget and takes whatever efficiency results.

The target is a target, not a floor

This is the part that changed, and it is the most consequential PPC story of 2026.

Historically, budget-limited campaigns often overperformed their targets — a 400% tROAS routinely delivering far more, because a constrained budget meant only the very best auctions were bought. Many accounts exploited this deliberately: set a low target, let Google overshoot.

Google announced on 15 June 2026 that this was ending, and from 17 August 2026 bids on budget-limited campaigns optimise more consistently toward the stated target. Google's own example: a campaign with a $10 target CPA delivering an actual $5 CPA will now deliver closer to $10.

Campaigns that are not budget-constrained are unaffected — they already scaled in line with their target.

The practical fallout was immediate and is visible in the daily feed: practitioners reporting sharply higher CPCs, ROAS collapsing toward target, and a threshold effect where raising a target slightly above recent actuals throttles delivery. Every target set under the old overshoot behaviour is now mis-set, in the unprofitable direction.

The response is not a trick. Re-derive targets from the margin you actually need rather than from a historical number that used to be beaten.

Learning periods and data

Bid strategies need conversion volume. Changing strategy, target or structure triggers a learning period during which performance is unstable. Change one thing, then wait — the common failure is stacking three changes in a week and being unable to attribute anything.

What trips people up

• Setting a target from last month's actual rather than from required margin.
• Adjusting targets weekly. You are permanently in learning.
• Assuming automation removes the need for negatives and structure. It increases it — the machine spends faster in whatever direction you pointed it.

You have got this when

You can explain why a campaign that used to beat its target now merely meets it, and what you would change in response.