How to Manage Google Ads Smart Bidding Strategies
Learn when to switch to Smart Bidding and how to scale it without resetting learning.
Open Toolkit
Learn when to switch to Smart Bidding and how to scale it without resetting learning.
Open ToolkitSmart Bidding is Google’s machine learning system for setting bids automatically, based on the likelihood a click will turn into a conversion. It’s powerful, but it isn’t a set-and-forget switch. Turn it on too early, feed it the wrong signals, or push targets too hard too fast, and performance drops instead of improving.
This guide covers when to switch to Smart Bidding, how much conversion data you actually need, how to scale Target CPA and Target ROAS campaigns without breaking them, and the campaign structure that gives Smart Bidding the best chance to work.
Smart Bidding needs conversion data to learn from. Before you switch, check three things.
Conversion tracking is accurate. If your conversion tracking is double-counting, missing conversions, or tracking the wrong actions (like every form view instead of every form submission), Smart Bidding will optimise toward the wrong outcome. Fix tracking first. Everything else comes second.
The campaign has enough history. A brand new campaign with two days of data has nothing to learn from. Enhanced CPC is no longer an option (Google retired it in March 2025), so the choice is manual CPC for a short baseline, or Maximise Conversions from day one with no target, which is what we do: it learns from the first conversions it sees and a target can be added once there are enough of them. Either way, give it two to four weeks before judging it, unless the account already has strong conversion history from a similar campaign to draw on.
The account can tolerate a learning period. Every time you switch bid strategies or make a significant change to one, the campaign enters a learning phase of roughly one to two weeks. Performance is often less stable during this window. Don’t switch strategies right before a big sale period or a launch. Switch when you have room to absorb a rough week or two.
A good rule of thumb: manual bidding is for the short period of building data, if you use it at all. Once the data exists, Smart Bidding will almost always outperform manual bidding, because it can adjust bids for each individual auction based on device, time, location and audience signals that a human simply can’t factor in bid by bid.
This is the most common reason Smart Bidding underperforms: switching before there’s enough data for the algorithm to learn from.
Maximise Conversions. No strict minimum. But the telltale signs a campaign is ready to switch to Maximise Conversions are that it’s limited by budget and it’s converting consistently within the budget constraints.
Target CPA. Aim for at least 30 conversions in the last 30 days at the campaign level before switching, and ideally 50 or more. Below that, the algorithm doesn’t have enough examples of what a “good” conversion looks like, and it will make bidding decisions on too little evidence. If a campaign is under this threshold, run Maximise Conversions first and switch to Target CPA once volume builds up, or consolidate similar campaigns so they share a bigger data pool.
Target ROAS. Same threshold as Target CPA, at least 30 conversions in 30 days, but this time each conversion needs an accurate value attached to it. Target ROAS is only as good as your conversion value data. If every purchase is tracked with the correct order value (not a flat number), and refunds are excluded or subtracted, the algorithm can genuinely tell a $30 order from a $300 order and bid accordingly. Without that, you’re giving it a return target to hit with no reliable way to measure return.
If an account doesn’t have enough conversions for Target CPA or Target ROAS, that’s not a reason to force it. Run Maximise Conversions until volume catches up, then add a target CPA (that is what Target CPA now is under the hood), or run Maximise Conversion Value with an optional target ROAS when real values are attached to each conversion.
Once a campaign is stable on Target CPA or Target ROAS, the temptation is to push it harder straight away. Do it gradually instead.
Change targets in small steps. Move a Target CPA or Target ROAS target by no more than 10 to 20 percent at a time. A bigger jump resets the learning phase and can cause a sharp drop in volume while the algorithm recalibrates.
Space changes out. Give each change one to two weeks to settle before making another one. Changing targets weekly, or worse, daily, keeps the campaign permanently in a learning state and it never gets the chance to optimise properly.
Scale budget, not just targets, when you want more volume. If a campaign is hitting its target comfortably and you want more conversions, raising the budget is often more effective than loosening the CPA or ROAS target. A higher budget lets the algorithm bid on more auctions at the same efficiency, rather than bidding less efficiently on the auctions it already had.
Use seasonality adjustments for short-term spikes. For a sale, promotion or known demand spike lasting one to seven days, use a seasonality adjustment rather than changing the target itself. This tells Google to expect a temporary shift in conversion rate without permanently altering what the algorithm has learned. Google built the tool for short events and says it may not work well beyond about fourteen days, so a longer season is a job for the target and the budget, not an adjustment.
Consider portfolio bidding once you have multiple similar campaigns. Grouping campaigns with a shared goal into a portfolio bid strategy pools their data, which can help smaller campaigns benefit from the learning of larger ones. This works best when the campaigns genuinely target the same kind of customer and outcome, not when they’re being grouped just for convenience.
Efficiency gains come from improving the inputs Smart Bidding uses, not from fighting the algorithm.
Improve conversion tracking quality. Enhanced conversions and offline conversion imports (like matching a lead to a closed sale in your CRM) give the algorithm a clearer picture of which clicks actually turned into revenue, not just which ones filled in a form. This alone often improves efficiency more than any target adjustment.
Feed it real value, not flat values. For lead generation accounts, this means passing through lead quality data where possible (a qualified enquiry weighted higher than a general one). For eCommerce, it means making sure the actual order value is what gets sent back, including shipping and tax treatment being consistent. A flat value per lead is the trap. It turns Target ROAS into Target CPA with a built-in bias towards whichever action carries the highest figure and against the cheapest, and the algorithm will pay almost anything for the former. We watched that one setting halve a campaign’s leads and double its click cost inside a month before it went back to Maximise Conversions.
Tighten targets slowly as performance stabilises, rather than leaving an easy target in place indefinitely. A campaign that’s been comfortably beating its Target CPA for a month has room to move that target closer to your true breakeven point.
Reduce account fragmentation. Several small campaigns competing for the same searches split your conversion data across multiple pools, and each one learns more slowly than one larger campaign would. Consolidating overlapping campaigns, where it makes sense for the business, gives Smart Bidding more data to work with per campaign.
Watch for target changes made too soon. If you’ve changed a target in the last week, resist the urge to change it again because performance looks off. That instability is often the learning phase, not the target itself failing.
The account structure around a bid strategy matters as much as the strategy itself. Here’s the approach that gives Smart Bidding the best foundation to work from.
Start with phrase match, tightly controlled. Phrase match campaigns targeting your core, bottom-funnel search terms give you a controlled base of high-intent traffic and clean data. This is where you build the conversion history the rest of the account will lean on.
Layer in a tightly controlled AI Max campaign. AI Max for Search extends your reach using Google’s AI to match more search terms and generate assets, but “tightly controlled” is the key phrase. Keep a solid negative keyword list in place, use brand exclusions where needed, and review the search terms report regularly so you can see exactly what it’s matching to and adjust as it scales.
Add broad match with extensive negative keywords once the account has data to protect. Broad match gives Smart Bidding the widest pool of search terms to find efficient conversions in, but it needs guardrails. Build out negative keyword lists covering irrelevant intent (informational searches, job seekers, competitor terms, wrong locations or services) so broad match spend stays on searches that are genuinely relevant, not just technically related.
Give every ad group a full, varied set of assets. Google has said Ad Strength itself is a diagnostic and plays no part in the auction or in bidding, so chase the substance rather than the label: fifteen distinct headlines and four descriptions give the system more combinations to test and more signal about what converts. Weak or repetitive copy limits what the algorithm has to work with, no matter how well the bid strategy itself is set up.
Together, this structure works in stages: phrase match builds a clean, controlled data foundation, AI Max extends reach without losing control, and broad match (once the data and negative lists are solid) gives Smart Bidding room to find additional efficient volume. Strong, varied ad copy ties it all together by making sure every auction Smart Bidding wins is backed by an ad worth clicking.
Smart Bidding rewards patience and clean inputs. Switch once conversion tracking is solid and there’s enough history to learn from, change targets in small, spaced-out steps, and put your effort into improving conversion data quality rather than chasing the target itself. Build the account structure so it earns Smart Bidding’s trust with data before asking it to stretch further. If you’re not sure your account is there yet, a Google Ads audit will show you where the gaps are.
Maximise Conversions has no strict minimum and is the right starting point for a new campaign. For Target CPA, aim for at least 30 conversions in the last 30 days at campaign level, ideally 50 or more; Target ROAS needs the same volume with an accurate value on every conversion. Below those levels the algorithm is guessing from too few examples, so either run Maximise Conversions until volume builds or consolidate overlapping campaigns into one data pool.
When the campaign has been converting consistently for a month, you know the cost per conversion it achieves, and you want to hold it there. Set the first target at or slightly above that achieved figure. A target set well below it does not lower costs, it lowers volume, because the system bids less and wins fewer auctions. Tighten in 10 to 20 percent steps with one to two weeks between moves.
A change of more than about 20 percent to a target, a change of bid strategy, or a large budget change can put the campaign back into learning for roughly a week, and performance is usually less stable while it relearns. Small, spaced changes avoid most of it. Resist changing a target again inside the first week because results look off; the instability is often the learning phase, not the target failing.
Only if every conversion carries a real value. Lead forms usually do not. When every lead is given the same flat figure, Target ROAS behaves like Target CPA with a built-in bias towards whichever action carries the highest number and against the cheapest, and it will overpay for the former. Run Maximise Conversions, with a target CPA once volume allows, until you can pass qualified-lead or closed-sale values back from your CRM.