Google Ads Smart Bidding – when to trust it, when to override

Across generic search, Shopping, and Performance Max with sufficient conversion volume, smart bidding produces better bids than manual control. The platform is acting on more signal per decision than any operator can match. Manual CPC across the account, single-keyword ad groups, granular match-type segmentation, daily bid tweaks based on yesterday’s data: these were the levers good operators differentiated themselves on a decade ago, and they are mostly gone.

Brand search is where smart bidding falls down. Across our brand-search bid tests over the past 18 months, manual CPC has outperformed target ROAS in the substantial majority of cases, sometimes by margins severe enough to settle a test inside a fortnight. The largest single result was a fashion account where switching brand from target ROAS to manual CPC drove a 642% ROAS uplift, with revenue up 215% and CPCs down 71%. It sits at the extreme end of a pattern that has held consistent across 20 controlled bid-strategy tests we have run.

Why smart bidding usually wins

Smart bidding’s structural advantage is the volume and breadth of signal it can act on. A target ROAS or target CPA campaign is bidding against the predicted likelihood that any given query, audience, time-of-day, device, and historical-pattern combination will convert at the target you have set. The algorithm adjusts bids in real time across millions of micro-decisions, and every one of those decisions is informed by data the operator does not have line-of-sight to.

For most generic search activity in an ecommerce account, where the goal is finding incremental conversions across diverse query patterns, this is the right tool. Smart bidding’s edge is in volume and complexity, and generic search is volume and complexity. The compounding effect is real too. Smart bidding gets better with more conversion data, which means a clean conversion-tracking setup (single primary purchase action, accurate values) produces better bidding decisions, which produce more conversions, which feed a richer signal back into the algorithm. The flywheel runs in the right direction once you have set it up correctly.

Where the override pays for itself

Brand search

This is the most consistent pattern in our test data. Across our brand-search bid tests over the past 18 months, manual CPC has outperformed target ROAS in the substantial majority of cases, often by margins large enough that the result was unambiguous within a fortnight.

Three results are representative. One fashion-account test produced a 35% ROAS uplift with manual CPC against target ROAS, alongside a 38% lower CPC; the manual cell captured the same brand-search volume at materially lower cost. A second account, run over a longer test window, returned a 124% ROAS uplift with manual CPC, paired with a 53% reduction in CPC and a 45% increase in revenue, which directly proved the hypothesis that lower CPCs would let manual outscale target ROAS at scale. The largest single result we recorded came from a fashion brand whose manual-CPC cell drove a 642% ROAS uplift, with revenue up 215% and CPC down 71%.

Target ROAS bids dynamically against the predicted conversion value of each click, and on brand traffic the predicted conversion value is high because people searching for your brand are already part-way through their decision. So target ROAS bids aggressively. The conversion was already going to happen, though. The user knew the brand and was coming anyway, so the aggressive bid just buys the same conversions at a higher CPC. Manual CPC, by setting a fixed ceiling, captures the same volume at substantially lower cost. Target ROAS is doing exactly what target ROAS is designed to do; it is the wrong tool for brand traffic.

Not every brand-search test wins for manual. We have one test on a footwear account where target ROAS outperformed, and the result was statistically insignificant. The conditions where manual CPC reliably wins on brand are: a clean separation where brand traffic is not contaminating Performance Max or generic Shopping, brand search volume high enough for the test to be meaningful, and brand defence intent that does not depend on impression share specifically.

The impression-share caveat is worth its own note. Where the brand goal is defence, ensuring your brand always appears for brand searches against competitor bidding, target Impression Share can outperform both manual CPC and target ROAS. We have run footwear tests where target Impression Share reduced CPC by 31% while maintaining coverage. The decision tree on brand bidding becomes: manual CPC for cost-efficient revenue capture, target Impression Share for brand defence, target ROAS rarely the right answer.

Campaigns without enough conversion volume

Smart bidding is signal-dependent. Without enough conversion data flowing through a campaign, the algorithm cannot self-stand on target ROAS or target CPA. Bids become erratic and the campaign underdelivers against what a more human-led structure would achieve.

The threshold we work to is 50 conversions in the past 30 days as the minimum for a campaign to be self-standing on smart bidding. Below that, the operating principle becomes consolidate-then-segment: run a single broader campaign that builds conversion volume, prove the data is sufficient, and only split into more granular campaign structures once it is. This matters most for accounts with low total spend, accounts entering new markets where the conversion history is thin, and accounts that have been over-segmented.

More data per campaign produces better bids than a more granular structure with less data each. The instinct to segment for control is intuitive, but the granular structure that looks well-organised in the interface produces worse results in the auction.

If your campaign is below threshold, the right sequence is to consolidate first, transition to Maximise Clicks while you build conversion data (it pushes traffic without requiring a conversion target the algorithm cannot reliably hit), then move to Maximise Conversions once the data is sufficient, and finally to target ROAS once you have a defensible target. We have seen this work on home textiles accounts, where Maximise Clicks produced the volume needed to switch to Maximise Conversions, which then drove higher CVR and revenue once the data was in place. Starting with a segmented structure and a smart bidding strategy the data cannot support produces a campaign that looks good in the interface and underdelivers in the auction.

Picking the wrong smart bidding strategy

The third common failure happens within smart bidding itself. Target ROAS, target CPA, Maximise Conversions, Maximise Conversion Value, and Maximise Clicks are not interchangeable, and picking the wrong one for the campaign’s objective produces results that look like smart bidding failing when actually the wrong tool was used.

Target ROAS beats Maximise Conversion Value for scaling, on the test data. A home improvement account ran a head-to-head between the two on the hypothesis that Maximise Conversion Value would scale better for purchases. The result was the opposite. Target ROAS drove 238% higher last-click ROAS and 154% more revenue, with no scaling benefit from the alternative. Maximise Conversion Value sounds like a stronger version of target ROAS in the interface, but in our tests, target ROAS proved more disciplined for accounts with a defensible ROAS target.

Maximise Clicks is the right strategy when an account is still building conversion data. It pushes traffic and gets the conversion log populated. Once that volume is sufficient, switching to Maximise Conversions produces materially higher CVR and revenue. Staying on Maximise Clicks past the switching point is a common error; the interface does not prompt for the switch, and the operator has to make the call.

Target CPA versus target ROAS. For ecommerce, target ROAS is almost always the right choice between the two. Target CPA can be cleaner for lead generation, but mixing the two within the same ecommerce account creates confused signal across campaigns that should be working towards a coherent business outcome.

What still goes wrong even when the strategy is right

Operators often choose the right bid strategy and still degrade smart bidding’s performance through habits that have not adjusted to how the algorithms actually work.

The most common is daily target adjustments. Google’s smart bidding strategies operate on a roughly monthly cycle, and day-to-day demand fluctuates within that cycle. Operators changing targets daily based on a noisy week of data are giving the algorithm contradictory signals, and the bidding becomes worse as a result. Where target adjustments are needed (because spend is consistently overrunning, for example), they should be deliberate and gradual.

Pausing campaigns too quickly is the second. New campaigns have a learning phase, and short tests under that window produce noisy results that look like the campaign is not working when actually it has not yet had time to optimise. The minimum sensible test window for a new bid strategy is 30 days, ideally Google’s 30.4-day measurement window for clean comparison.

Over-grooming search terms is the third. Excluding non-converting search terms on a 30-day lookback removes signal from the algorithm. Some non-converting search terms are part of how customers research before buying and contribute to conversions later. Some are relevant queries where the landing page or ad copy is the actual problem. Where a search term is genuinely irrelevant, exclude it. Where it is relevant but not directly converting, leaving it in the system gives the algorithm more matching to work with.

The fourth runs upstream of bidding entirely. Smart bidding is only as good as the conversion data feeding it, and if your primary conversion action mixes purchases with funnel events like add-to-cart or page view, smart bidding is optimising against confused signal whatever bid strategy you have chosen.

The decision framework

Start with conversion tracking. Single primary purchase action, accurate value. If either is missing, fix it before changing bid strategy. Smart bidding is fed by conversion data, and broken tracking starves it.

Then check campaign volume. If the campaign has at least 50 conversions in the past 30 days, smart bidding can self-stand. If it does not, consolidate first and build volume on Maximise Clicks, then transition to Maximise Conversions when the data supports it.

For brand search, default to manual CPC for cost-efficient revenue capture, or target Impression Share for brand defence. Target ROAS is rarely the right call.

For generic search, Shopping, or Performance Max with sufficient volume, use target ROAS. Use Maximise Conversion Value only if you have a specific reason; on the test data we have, target ROAS outperformed.

If you are actively learning, whether that is testing new creative, new audiences, or a new market, Maximise Conversions gives the algorithm room to find conversions before you constrain it with a ROAS target. Move to target ROAS once you have a defensible target.

And test AI Max where existing broad-match coverage has hit its ceiling. Where existing coverage is already efficient, the lift is unreliable.

These decisions are not made once. The right bid strategy changes as the campaign matures and the platform itself evolves.

FAQ