What Smart Bidding optimises and what it needs
Google documents Smart Bidding as auction-time bidding that uses signals available at the moment of the auction — device, location, time, query, audience membership, browser, operating system and others — to predict the likelihood and value of a conversion and set a bid accordingly. The advertiser does not see the bid; the advertiser sees the outcome and sets the goal.
Three things have to be true for that to work. The conversions the campaign optimises towards must be tracked completely, because the model learns from what is reported and a conversion it never sees is one it will not pursue. The conversion actions included in the campaign's goals must be the ones that matter, because the model treats every included action as equally worth chasing unless values distinguish them. And for value-based strategies the values must reflect what the business actually earns, because a campaign told that every lead is worth the same will optimise for the cheapest leads.
Data volume matters less than it once did but still matters. A model with many recent conversions predicts better than one with few, and a campaign whose conversions are rare and irregular will hold a target loosely. Where volume is thin the options are to optimise towards a more frequent action that correlates with the real goal, to pool campaigns in a portfolio, or to accept wider variation. The guide to bidding with few conversions works through each of those options for an account whose conversions are rare whatever its budget. The bidding hub sets out how targets derive from margin; this guide assumes that arithmetic has been done. Whether Smart Bidding should also be given broad match keywords is a separate decision, tested in the broad match expansion test with the bid strategy held fixed.
The strategies
| Strategy | Optimises for | Target | Data needed | Best for |
|---|---|---|---|---|
| Maximise conversions | Most conversions within budget | None, or an optional target CPA | Conversion tracking | New campaigns, lead generation without a firm cost ceiling, establishing a baseline |
| Target CPA | Conversions at or near an average cost | Target CPA (set as an option on maximise conversions) | Conversion tracking and enough volume to learn | Lead generation and sales where each conversion is worth about the same |
| Maximise conversion value | Most conversion value within budget | None, or an optional target ROAS | Conversion values passed per conversion | E-commerce and any goal where conversions differ in worth |
| Target ROAS | Conversion value at or near a return on spend | Target ROAS (set as an option on maximise conversion value) | Values and volume | E-commerce with margin-derived targets; lead generation with value rules |
| Target impression share | Share of eligible impressions at a chosen position | Impression share and position, with a maximum CPC | None (not conversion-based) | Brand terms, visibility goals, defensive campaigns |
| Maximise clicks | Most clicks within budget | Optional maximum CPC | None | Traffic goals, early data gathering |
| Manual CPC | Whatever the advertiser decides, click by click | Bids per keyword or ad group | None | Small controlled tests, campaigns without usable conversion data |
Google groups the first four as Smart Bidding. In the current interface target CPA and target ROAS are set as optional targets on maximise conversions and maximise conversion value respectively, rather than as separate strategies; the behaviour is the same, and older accounts may still show the standalone names. Enhanced CPC, which once sat between manual and automated bidding, has been retired for Search and Display campaigns.
Maximise conversions and conversion value, with and without targets
Without a target, maximise conversions spends the budget to get as many conversions as it can, and maximise conversion value spends it to get as much value as it can. The budget is the only constraint, and the documented behaviour is that the strategy will try to spend it. These are the right starting strategies for a campaign with no history: they tell you what the campaign can achieve at the budget, and that observed figure is the basis for a target later.
With a target, the same strategies are told to hold an average cost per conversion or an average return. The target changes the question from "how many at this budget" to "how many at this price". Google documents that the target is an average the strategy aims for, not a cap on any single conversion: some conversions will cost more and some less. A target also means the budget may go unspent when the model cannot find enough auctions it believes will meet it.
The choice between the conversion-count and conversion-value pair rests on whether conversions differ in worth. For a store, orders differ by value and the value-based strategies are correct. For a lead form, the leads may be equal, in which case counting them is right, or they may differ by product interest or company size, in which case assigning values by rule — and then using the value-based strategy — produces better leads at the same cost. Value rules and per-action values are set in the conversion configuration, which the measurement cluster covers.
Target CPA and target ROAS: setting the number
The target is where most Smart Bidding problems begin. A target set from economics, as the bidding hub describes, is a ceiling on cost or a floor on return that the business can afford. A target set from the campaign's recent average is a statement of what it has been doing. The two should be reconciled: the campaign should run at or better than the economic limit, and it should get there from where it currently is in steps.
For target ROAS, start from the Break-even ROASMinimum ROAS to avoid a loss. that gross margin implies and add the profit the business requires; the ROAS glossary entry and the ROAS calculator cover the arithmetic. Then compare that figure with the campaign's trailing return. If the economic target is far above what the campaign achieves, setting it at once will collapse spend; set it part of the way there and move it in steps as the campaign adjusts. For target CPA, the CPA calculator gives the maximum from customer value and conversion rate (CVRShare of visits or clicks that convert.), and the same stepwise approach applies.
Two documented details affect the number. The attribution model and the conversion window decide how many conversions and how much value the campaign is credited with, so a target set under one model needs revisiting if the model changes. And conversion valuesThe monetary value an advertiser assigns to a conversion, such as the order's revenue or a lead's expected worth. that include returns, cancellations or margin adjustments give a different ROAS from gross order values, so the target has to be expressed on the same basis as the values the campaign receives. The glossary's definition of ROAS in Google Ads sets out exactly what the platform's conversion value / cost includes.
Moving a campaign from no target to a target ROAS
A home-fitness retailer launches a Search campaign on maximise conversion value with a £300 daily budget and no target. After four weeks it spends £8,400 and returns £29,000 in tracked revenue, a ROAS of about 345%. Gross margin on the range is 40%, so break-even ROAS is 250%, and the business wants a 20% margin after ad cost, which implies a target of roughly 400%. Rather than setting 400% at once, the team sets 370%; spend dips to £240 a day for ten days while the model adjusts, then recovers to £290 a day at a delivered ROAS of 378%. Three weeks later the target is raised to 400%, spend settles at £270 a day and delivered ROAS at 405%. Monthly revenue is slightly lower than at launch, from £29,000 to £27,500, but profit after ad cost rises from £3,200 to £3,900, which was the point.
Target impression share and manual CPC: when they still fit
Not every campaign has a conversion goal. Target impression share bids to show the ad for a chosen share of eligible searches at a chosen location on the page (absolute top, top, or anywhere), with a maximum CPC the advertiser sets. It fits brand campaigns where the goal is to occupy the top of the page whenever the brand is searched, visibility campaigns in a launch, and defensive campaigns where a competitor is bidding on the advertiser's name. The maximum CPC matters: without it the strategy will pay whatever is needed, and an aggressive share target on competitive terms can be expensive.
Manual CPC remains the right choice in a few cases. A campaign with no conversion tracking and no intention to add it. A small test where the advertiser wants to know exactly what was bid. A campaign whose goal the system cannot observe, such as phone calls that are not tracked. And the early days of an account where the team wants to see raw auction prices before delegating. Maximise clicks, with a maximum CPC, is the automated cousin for traffic goals and gathers conversion data faster than manual bidding on a new campaign, which makes it a common bridge to Smart Bidding.
What none of these strategies do is consider the probability of conversion. An account that runs them on campaigns with good conversion data is leaving the model's main advantage unused; the usual reason is habit rather than a decision, and the usual fix is a controlled move to a conversion-based strategy with the before-and-after measured at the account level.
Changing targets and budgets without resetting learning
Google documents a learning period after a bid strategy is created or significantly changed, during which performance can fluctuate while the model adjusts. The status column shows "Learning" during that period. What counts as a significant change is not published exactly, but the help content lists changing the strategy type, changing conversion actions, and large changes to targets or budgets as triggers.
The practical rules that follow are industry practice rather than documented thresholds. Change targets in small steps rather than in one move, and wait for the campaign to settle before the next step. Change one thing at a time, so that the effect of a target change is not confused with the effect of a budget change made the same day. Avoid changing targets during a period that is already unusual, such as a sale or a holiday, because the model will learn the unusual period as normal. And resist reacting to a few bad days: the target is an average over a window, and daily figures vary.
Budget changes interact with targets, which is why the budgets guide covers the budget side separately. A target campaign that is not spending its budget will not spend more if the budget is raised; the constraint is the target. A target campaign that is limited by budget will spend more if the budget is raised, and may find the additional conversions cost more than the average, which is a marginal cost question rather than a learning problem.
Learning, limited and status messages
The bid strategy status column is the main feedback the system gives, and each state calls for something different.
| Status | What it means | What to do |
|---|---|---|
| Learning | The strategy was created or significantly changed and is adjusting | Wait; avoid further changes; judge performance after the period |
| Limited (by budget) | The strategy could spend more at the target but the budget caps it | Raise the budget, or accept the constraint; see the budgets guide |
| Limited (bid limits) | Minimum or maximum bid limits in a portfolio strategy are constraining bids | Loosen or remove the limits, which Google generally advises against setting |
| Limited (low conversion data or similar wording) | Too few conversions for the strategy to learn reliably | Broaden the conversion action, pool campaigns, or run without a target longer |
| Misconfigured | A setting conflicts with the strategy, such as a missing conversion action | Fix the named setting |
| Inactive | No campaigns or ad groups are using the strategy | Assign it or remove it |
| Eligible | Normal operation | Review on the usual cadence |
The exact wording of the limited sub-statuses varies by campaign type and changes over time, so read the hover text in the account rather than relying on a list. The general rule is that "Learning" asks for patience, "Limited" names a specific constraint and the fix is to loosen it or accept it, and "Misconfigured" names a specific setting to repair. The guide to status messages in Google Ads reads these bid strategy statuses together with campaign, ad, keyword and conversion action statuses in a weekly sweep.
Portfolio strategies and shared targets
A portfolio bid strategy is a single strategy applied across several campaigns, with one target shared between them. Google documents it as a way to pool data and manage a target in one place, and, for target CPA and target ROAS portfolios, as the place where minimum and maximum bid limits can be set. The pooled data is the main benefit: three campaigns with modest conversion volume each learn better as one portfolio than as three separate strategies.
The cost is that the target is shared. The portfolio will allocate spend across its campaigns to hit the target overall, which means one campaign can run above the target while another runs below it. That is correct behaviour if the campaigns are economically equivalent and wrong if they are not — a high-margin and a low-margin product line in one target ROAS portfolio will be optimised to a blended target that is right for neither. Group campaigns into a portfolio when they share a goal and a margin; keep them separate when they do not, and accept thinner data as the price of a correct target.
Bid limits in a portfolio are documented as available and generally discouraged, because they stop the model bidding high where it predicts a valuable conversion and low where it does not. They have a place as a temporary guard rail while a new portfolio learns, after which removing them is usually the better choice. Portfolio strategies can also be paired with a shared budget, which is a budget decision covered in the budgets guide; the two are separate objects and one can be used without the other.
Performance Max uses the same strategies — maximise conversions or conversion value, with optional targets — and its targets are set the same way; the structural questions it adds, such as how many campaigns to split by margin, are in the Performance Max structure guide. The author's Google Ads expertise page describes how bidding decisions sit within account management as a whole.
Teams weighing a second search platform should read the Google Ads versus Microsoft Ads comparison, which covers how automated bidding and conversion volume behave differently on each network.
If performance fell after a switch and it is unclear whether the strategy is learning, constrained or being judged on incomplete data, the results dropped after a bid strategy change diagnosis separates those cases.
Whether to run maximise conversions on its own or add a target, and when to make the switch, is set out in the comparison of maximise conversions and target CPA.
Common mistakes
- Setting a target before there is a baseline. A new campaign with an aggressive target spends little and teaches nothing.
- Targets from averages, not margin. The campaign holds a number that has nothing to do with profit.
- Counting conversions that differ in worth. Cheap leads win when the goal was good leads.
- Changing the target and the budget on the same day. Two learning triggers, one unreadable result.
- Reacting to daily variance. Targets are averages over a window; a bad Tuesday is not a trend.
- Bid limits on a portfolio left on permanently. The model's main advantage switched off as a precaution.
- Pooling unlike campaigns in one portfolio. A blended target that is wrong for every member.
- Judging a strategy change at the campaign level. Conversions moved between campaigns; the account total is the honest figure.
