GOOGLE ADS SMART BIDDING MINIMUM CONVERSIONS REQUIRED

The question I get most often about automated bidding is not "which strategy" but "when." The honest answer is that the minimum conversions for smart bidding is the single gate I check before anything else, and I am stricter about it than Google is. Google will happily let you turn Smart Bidding on with almost no history; the interface does not stop you. In the accounts I manage, the number that actually predicts a good outcome is higher than the published minimum, and switching before you reach it is the most common self-inflicted wound I see on inherited accounts. This article is the threshold I wait for, what I do with an account that sits below it, and why I would rather leave a campaign on manual for another month than force the machine to learn from noise.

The smart bidding minimum conversions required, and the number I actually wait for

There are two numbers here and they get conflated constantly. The first is what Google publishes as the volume needed to evaluate performance. Google recommends measuring results over a period with at least 30 conversions, and 50 for Target ROAS, so the machine has enough data to calibrate and so you can read the outcome without fooling yourself. Google's own documentation on Smart Bidding is explicit that some strategies rely on a minimum volume of historical conversion data depending on campaign type, and that you can technically start without prior data.

The second number is the one that keeps me out of trouble: roughly 30 conversions in the trailing 30 days at the campaign level before I move to Target CPA, and closer to 50 before I move to Target ROAS. That is not a coincidence with Google's figure, it is the same figure treated as a floor rather than a starting line. Below 15 conversions a month I will not run Target CPA or Target ROAS at all; I keep Maximize Conversions or manual CPC and spend the month building the data instead of spending it on the algorithm's tuition. The gap between "you can switch" and "you should switch" is exactly where most accounts get burned, and understanding it is the difference between Smart Bidding working and Smart Bidding flailing. It is the same discipline I bring to reading a Smart Bidding learning period honestly rather than pretending a low-volume campaign will ever exit it.

Volume is not the only condition. The conversions also have to be recent, consistent, and tracked correctly. Thirty conversions spread evenly across the month is a usable signal; thirty conversions where twenty landed in one promotional week is not, because the model over-weights that spike. This is why I read the Smart Bidding pillar conditions as a package, not a single counter: enough conversions, arriving steadily, on a conversion action I trust.

Why forcing it early is worse than waiting

The intuition most people have is that Smart Bidding on thin data is merely slower, that it will find its footing eventually. That is not what I have watched happen. On too few conversions, the algorithm is estimating conversion probability from a sample so small that a handful of random wins or losses swings its bids hard. It over-corrects, chases the last conversion, and the campaign never leaves the learning phase. I have inherited accounts that had been "learning" for months because someone switched a 10-conversion-a-month campaign to Target CPA and then kept nudging the target every time it wobbled.

There is a compounding problem: every meaningful change restarts learning. Switch the strategy, move the target more than a little, change the budget by more than about 20 percent, and the model re-enters a learning window of one to two weeks. On a low-volume campaign that window may be longer than the time it takes to accumulate the next handful of conversions, so the campaign is permanently unstable. You are not paying for optimization at that point; you are paying for a model that never has enough recent data to settle. Forcing it early does not buy you a head start, it buys you a campaign that reads its own noise as signal.

The counterintuitive move, then, is patience. Leaving a campaign on Maximize Conversions or even manual for another few weeks while volume builds is not lost time; it is the cheapest way to reach the point where automated bidding actually has something to learn from. Instability from thin data behaves a lot like the instability I control with data exclusions in Smart Bidding after a tracking outage: in both cases the fix is to stop the model learning from data that misrepresents normal demand.

What I do when the account can't reach the threshold: micro-conversions

Plenty of accounts will never hit 30 primary conversions a month, and the answer is not to give up on Smart Bidding forever, it is to give the algorithm a denser signal. Micro-conversions are the lever. Instead of optimizing only toward the rare purchase or qualified lead, I define an earlier, more frequent action that correlates with the outcome I care about, and let the model learn from that volume.

Choosing a micro-conversion that actually predicts the outcome

The whole trick is correlation. A good micro-conversion is one that a genuine prospect does on the way to converting and a bad-fit visitor mostly does not: a pricing-page view with dwell time, a add-to-cart, a form-start, a configurator completion, a demo-video watch past a threshold. A bad micro-conversion is a raw pageview or a bounce-adjacent event, because it inflates the count without predicting revenue, and the model dutifully optimizes toward cheap, worthless traffic. I would rather have 40 form-starts a month that predict 8 sales than optimize directly on 8 sales the model can barely see.

Keeping the primary conversion as the truth

I set the micro-conversion as a secondary action for reporting and, where the campaign type allows, optimize toward it while watching the primary conversion as the real scoreboard. The point is to bootstrap: run Maximize Conversions on the micro-conversion until the account gathers enough primary conversions to migrate the campaign onto the real one. The micro-conversion is scaffolding, not the finished building, and I take it down once the primary volume can stand on its own.

Value rules: value-based bidding without enough conversions

The other tool for a low-volume account is conversion value rules, and it solves a different problem. When you want the system to bid toward value rather than raw conversion count but you do not have the 50-plus valued conversions Target ROAS wants, value rules let you tell Google that some conversions are worth more than others based on attributes you already know, before the sale even completes.

In practice I use them to encode business truth the pixel cannot see. A lead from a target region, a new customer versus a repeat one, a particular device or audience that I know from history closes at a higher rate or a higher basket value, all of these can carry a value multiplier. That turns a thin stream of same-looking conversions into a stream the algorithm can prioritize, so even Maximize Conversion Value has something meaningful to optimize toward on modest volume. It is not a way to fake data; it is a way to add the context you legitimately have. When I plan how those rules feed the auction, I treat them as part of the broader bidding strategy decision, not a bolt-on, because the value you declare is the target the system chases.

The caution is that value rules amplify whatever judgment you put into them. Overstate the value of a segment and the model over-invests in it; get the direction right and you steer spend toward your best customers on volume that would otherwise be too thin for value-based bidding at all.

How I migrate once the account finally clears the bar

When a campaign is genuinely holding above my threshold on steady, trusted conversions, the migration is deliberately boring. I confirm the conversion action is clean and the volume is recent, not a one-off spike. I start with Maximize Conversions to establish a real CPA baseline rather than importing a wishful target. I let it run a full learning window without touching it, because the fastest way to reset progress is to intervene on day three. Only then do I move to Target CPA using the observed CPA as the starting point, and I adjust in small steps.

For value-based bidding I do the same in sequence: Maximize Conversion Value first, then Target ROAS once I have enough valued conversions, with value rules carrying the context I trust. Across every one of these steps the same principle holds, the one this whole article is really about: Smart Bidding rewards accounts that feed it enough recent, honest signal and punishes accounts that force it before the data exists. The threshold is not bureaucracy. It is the line between a model that optimizes and a model that guesses, and waiting to cross it properly is the highest-return patience in the whole platform.