What is Smart Bidding?

Smart Bidding is Google Ads' label for the group of automated bid strategies that use its per-auction conversion prediction: maximise conversions and maximise conversion value, each optionally carrying a target cost per acquisition or target return on ad spend.

There is no control anywhere called Smart Bidding. An advertiser picks a specific strategy and, if appropriate, attaches a target. The label exists to distinguish those strategies from the automated ones that optimise toward clicks or impressions, which use no conversion prediction at all.

That distinction is the useful content of the term. Automation is not one thing: a strategy maximising clicks within a budget is automated and needs no conversion data whatsoever, while a strategy targeting a return on ad spend cannot function without a substantial history of valued conversions. Grouping the second kind under one name makes it possible to say something true about all of them at once.

At each auction the model estimates the probability that this impression leads to the nominated conversion, and where value-based, the value that conversion is likely to carry. It then bids in proportion to that estimate and the target.

The signals behind the estimate are the platform's own and are not fully enumerated publicly. The categories are unsurprising: characteristics of the query or context, the device and operating system, location and time, the person's prior interaction with the advertiser, and patterns learned across far more data than any single account contains.

Two consequences follow. The first is that the estimate improves with data, which is why volume matters so much. The second is that the advertiser cannot inspect or override an individual bid, which is the control being traded away.

The conversion. The model optimises toward whatever event the account nominated, faithfully and without judgement. An account whose conversion is a newsletter signup will get newsletter signups. Moving the optimisation event closer to revenue is usually the largest single improvement available, and volume is the limiting factor.

The value. For the value-based strategies, the number attached to each conversion is what gets maximised. A fixed value on every conversion turns a value strategy into a count strategy. Revenue works where margins are even. Where they are not, revenue-valued bidding reliably finds high-revenue, low-margin orders, and sending profit instead changes what the campaign buys.

The target. A target has to come from what a conversion is worth: contribution margin multiplied by the rate at which the conversion becomes a customer. A target from anywhere else is a number with no relationship to the business, and when it is set below what the market supplies the result is not cheaper conversions but far fewer of them.

The word "smart" does a certain amount of work that the underlying system does not.

It does not mean the strategy knows what the business wants. It knows what event was nominated and what target was set.

It does not mean the strategy is robust to bad inputs. A wrong conversion definition, a stale value feed or a broken tag are all pursued efficiently.

It does not mean it outperforms in every case. On thin conversion volume, on brand-defence campaigns where the clicks were always going to convert, and in the weeks after a structural rebuild, it can and does underperform a plain manual bid.

And it does not mean the decisions have been taken away from the advertiser. The three inputs above are exactly the decisions that matter, and they all remain.

Get measurement right before anything else. This is the input with the largest downside and the least glamour.

Choose the deepest optimisation event that clears the volume threshold, and revisit the choice as the account grows rather than setting it once.

Derive the target from the business and write down the derivation, so that the next person to look at the campaign can see where the number came from and does not adjust it because it seemed high.

Give changes room. A new target, a large budget move or a new creative set puts the model into a period of unstable delivery, and reading that period as a verdict is the most common way accounts talk themselves out of automated bidding.

Watch the impression share columns rather than bids, because they are what now tells you whether the constraint is money or competitiveness.

And keep a view of cost that is not the platform's own. A strategy optimising against attributed conversions will look better in the platform's report than in the business's accounts, and the gap between those two is worth tracking for its own sake.

Beyond the three inputs, an account can improve the prediction without touching the bidding at all, and this is where most of the remaining headroom sits.

Conversion volume is the first. Consolidating several thin campaigns into one gives the model a larger pool of events to learn from, and the usual objection — that consolidation loses control — is answered by the observation that there was little control to lose once the bidding was automated.

Conversion quality is the second. Sending back a qualification signal — the lead answered, the order shipped, the customer did not refund — lets the model learn on outcomes rather than on form fills. This is the same move as optimising deeper, applied after the fact, and it is available even where the deeper event is too rare to optimise toward directly.

Consistency is the third. A model learns from history, and an account that restructures every month presents a history that describes campaigns which no longer exist.

The practical question is rarely whether conversion-based automated bidding is good in the abstract. It is whether a specific campaign has the evidence to support it.

A campaign with conversion volume comfortably above the platform's threshold, honest tracking and a target derived from contribution is very likely better off on one of these strategies than on any manual arrangement.

A campaign below that threshold is choosing between a model working on almost nothing and a bid that at least stays where it was put, and the second is frequently the better of two imperfect options until volume arrives.

A campaign whose job is defensive rather than optimised sits outside the comparison entirely.