Smart Bidding: which strategy to choose and how to set its target
Automated bidding is neither magic nor a scam. It optimises towards whatever you labelled as success — so the result is decided by the quality of your conversion data and the target you set, not by which option you pick from a dropdown.
- The strategy is only a way of telling the algorithm your goal. Result quality comes from conversion data, not from the dropdown.
- Derive the target from margin, not from a historical average: break-even ROAS = 1 ÷ gross margin.
- One change at a time, then two to three weeks of calm. Two signals at once cannot be evaluated.
Automated bidding took over the job specialists used to do by hand: estimating, for every auction, how much is worth bidding. It does it better, because it sees signals that are not in the interface — device, time, user history, query context. What it does not do is the one thing people often expect: it will not decide for you what success is and what it may cost.
The strategies to choose from
| Strategy | What it does | When to use it | Risk |
|---|---|---|---|
| Maximise clicks | Gets as many clicks as the budget allows | A new account with no conversion data, short term | Clicks are not a goal. Do not leave it on beyond a few weeks |
| Maximise conversions | Gets as many conversions as the budget allows | Launch, when you do not yet know what CPA is realistic | It always spends the whole budget. Without a target, CPA can be anything |
| Target CPA | Holds the average cost per conversion at the value you set | Lead generation where conversions have similar value | Too low a target strangles the campaign: it stops spending |
| Maximise conversion value | Gets the highest total value the budget allows | Shops with widely differing order values | Can chase expensive, low-profit orders |
| Target ROAS | Holds the ratio of value to cost at the level you set | Shops with working conversion values and known margin | Depends entirely on the value you pass being correct |
A practical rule: if your conversions have differing values, you belong with the value strategies (maximise value, target ROAS). If they are similar in value, a CPA strategy is enough. A shop running on target CPA is telling the system that a €20 order and a €2,000 order are the same thing.
The target comes from margin, not from the average
The most common approach we see: take the historical CPA and set it as the target. That is freezing the current state, not steering.
Correctly, the target is derived backwards:
- Work out the gross margin on one conversion — for a shop from the average order, for an enquiry the margin on the job times the share that close.
- Break-even sits where CPA equals that. Break-even ROAS = 1 ÷ gross margin.
- Set the target above break-even, with headroom for fixed costs, management and fluctuation.
At a gross margin of 30 % the break-even ROAS is 3.33. A sensible target is therefore something like 4.2, not “4, because that is what came out last time”. The converter between metrics and a table of break-evens by margin is in ROAS, CPA, CAC and LTV.
When cosmetics run a 55 % margin and electronics 10 %, they cannot share a ROAS target. A single target means budget flows to where things sell most easily, not where they earn most. Split campaigns or asset groups by margin and give each its own target.
What you feed the algorithm: conversion data
This is the part that decides the outcome more than the choice of strategy. Smart Bidding optimises towards whatever you marked as a conversion. Include add-to-cart and you will get carts.
- One main conversion. An order, or a qualified enquiry. Measure the others but do not let them drive optimisation.
- The conversion value must be truthful. Ideally order value excluding VAT and shipping; better still margin, if you can pass it.
- Enough conversions. Roughly tens a month per campaign so the strategy has something to work from. When short, merge campaigns rather than leaving five of them starving.
- Send offline outcomes too. For enquiries, the difference between “a form arrived” and “a deal happened” is the most valuable thing you can give the system. See first-party data and enhanced conversions.
- Watch measurement completeness. After consent mode some conversions are missing or modelled, and the algorithm learns from an incomplete picture. See consent mode.
The learning phase: what happens and what not to do
After a change of strategy, target or a substantial change of budget, the system looks for a new equilibrium. It usually takes a few days to two weeks depending on conversion volume, and performance fluctuates during it.
What that means in practice:
- One change at a time. Raise the budget and loosen the target on the same day and you will not know which caused the result.
- Do not intervene mid-way. The worst move is reverting after three days because “it is not working”. That restarts learning.
- Move targets in steps. Going from ROAS 6 to 3 is a different campaign. Ten to twenty per cent at a time, with a gap between.
- Write the expectation down first. What you expect from the change and when you will check. Without it, a week later there is always a reason it actually went fine.
When the strategy will not spend the budget
A frequent situation: the campaign has a €2,000 budget, spends €700, and nobody knows why. Smart Bidding will not spend if it cannot meet the target. The order of checks:
- Is the target realistic? A target ROAS of 10 in a category that historically returns 4 means the strategy will decline most auctions. Loosen the target.
- Is there enough demand? Check impression share. High share plus unspent budget means an exhausted market — you need more queries or products, not more money.
- Are there enough conversions? At a handful a month the strategy has nothing to build on.
- Is targeting holding it back? Narrow audiences, a small feed, too many negative keywords.
This is exactly the situation where adding budget does nothing — and where the decision most often goes wrong. Covered in the article on budget.
Portfolio strategies: when yes
A shared strategy across campaigns lets the system move budget to wherever the opportunity currently is. It makes sense for campaigns with the same commercial logic and similar margin. It does not make sense when you put brand and non-brand into one portfolio, or categories at 10 % and 60 % margin — there you only hide where the money actually goes.
What to take away
Smart Bidding does not decide whether you make money. It decides how efficiently it will pursue the goal you give it. So the work that pays off is outside the strategy dropdown: properly measured conversions, a truthful conversion value, a target calculated from margin, and separate campaigns for things with different economics. After that, one change at a time and patience.
Common questions
What target CPA or ROAS should I set?
Derive it from margin, not from a historical average. Work out the gross margin on one conversion; break-even ROAS equals one divided by that margin, so at a thirty per cent margin it is 3.33. Then set the target above break-even with headroom for fixed costs, management and fluctuation. If you have categories with markedly different margins they need separate campaigns and separate targets — one target for a whole account sends money where things sell most easily, not where they earn most.
Why will Smart Bidding not spend my whole budget?
Because it cannot find enough auctions that would meet the target you set. The most common cause is too strict a ROAS or CPA target; the second is exhausted demand, which you spot through high impression share. It can also be too little conversion data, narrow targeting or an excess of negative keywords. Adding budget helps in none of those cases, because budget is not what is limiting the campaign.
How long does the learning phase last and what should I do during it?
Usually a few days to two weeks depending on how many conversions the campaign gathers. Performance fluctuates during it, and the worst thing you can do is revert the change after three days — that restarts learning. The rule is one change at a time, then calm, and evaluation after two to three weeks. Move targets in steps of around ten to twenty per cent rather than in a single jump.
- Explaining conversion lift, PPC LandBackground on why the quality of conversion data feeds through into the performance of automated bidding and how real contribution is measured.
- Google Ads September 2026: New Data Strength Tools, Common Thread CollectiveOverview of the tools Google uses to measure conversion data completeness. An agency source; the improvement percentages originate from Google.