Somewhere in the setup of almost every underperforming account we audit, there is a moment where somebody switched on automated bidding.
It is an easy switch to justify. Google recommends it. The interface nudges toward it. It promises to find people likely to convert, which is exactly what you want. So it gets turned on, often in week one, and then the account spends six months wondering why the algorithm is not delivering what the algorithm promised.
The answer is almost never that Smart Bidding is bad. It is that it was switched on before there was anything to be smart with.
Smart Bidding Isn't Smart. It's Trained.
The word "smart" does a lot of damage here, because it suggests the system arrives knowing things. It doesn't. It arrives knowing nothing about your business, and it learns entirely from the conversions you feed it.
When you run Target CPA, the algorithm looks at everyone who converted, builds a profile of what those people had in common, and then bids more aggressively on searches that resemble them. Device, location, time of day, query wording, browsing history, hundreds of signals you never see.
That process needs examples. Not a handful. Enough that patterns separate from noise.
Give it four conversions and it cannot tell the difference between a real pattern and a coincidence. Two of your four converters happened to be on iPhones at 9pm, so now it thinks iPhone evenings are gold. That is not intelligence. That is superstition built from a sample too small to mean anything.
An RV internet company we audited had spent roughly $50,000 running automated campaigns and produced about four conversions worth the name. The algorithm had been optimizing that entire time. It had simply never been given enough real outcomes to optimize toward, so it optimized toward noise and kept billing for the privilege.
Thirty a Month. And Google Agrees.
Our rule is 30 quality conversions per month in a campaign before we hand bidding to the algorithm. That number is not ours.
Google's own documentation sets 15 conversions in the past 30 days as the hard minimum to use Target CPA, and recommends 30 as the level where it actually performs. We use the recommended number rather than the minimum, because the minimum is where the feature becomes available, not where it becomes good.
Two details that catch people:
It is per campaign, not per account. Fifty conversions spread across five campaigns is ten each, and each campaign learns on its own data. An account can look comfortably above the line in aggregate while every individual campaign sits well below it.
It is a rolling 30 days, not a lifetime total. Four hundred conversions from two years ago do nothing. The algorithm cares about recent behavior, because your market, your prices, and your competitors have all moved since.
If you are running a business where 30 conversions a month is genuinely out of reach, that is not a failure. Plenty of good businesses have a smaller number of much larger deals. It just means automated bidding is the wrong tool, and the rest of this article matters more to you, not less.
What You Do Instead, and Why It Works Better Than Waiting
The alternative is not "wait around on manual CPC until you qualify." It is a deliberate build phase with its own goal.
We launch with manual CPC and exact match, targeting the tightest possible set of keywords. Not a broad net. The specific searches we already believe are the money searches for that business.
The objective in this phase is not volume. It is page position on the searches that matter. We want to be in the top positions for a small number of high intent terms, consistently, and we watch impression share on those specific terms to confirm it. For a steel building manufacturer we work with, the entire early phase was about beating 15 to 20 competitors in one state on a handful of terms, rather than being visible everywhere for everything.
This does three things at once:
- It produces conversions faster than a broad build would, because you are only paying for searches with real intent behind them.
- It produces conversions you can trust, which matters enormously for what comes next.
- It teaches you which keywords actually convert, which no algorithm can hand you and which stays valuable long after you have automated.
You are not idling. You are manufacturing the training data the algorithm will need, deliberately, instead of hoping it accumulates by accident. It is the least glamorous phase of running Google Ads and the one that decides whether everything after it works.
Thirty Bad Conversions Are Worse Than Five Good Ones
Here is where the threshold gets misused. People chase the number.
If you need 30 conversions before automated bidding works, the temptation is to make 30 conversions appear. Mark newsletter signups as conversions. Count add-to-cart. Count page views. Suddenly the campaign clears the bar and Target CPA goes on.
You have now trained the algorithm to hunt for newsletter signups. It will get very good at that. Your cost per newsletter signup will fall beautifully while your actual sales sit flat or decline, because every bid decision is now pointed at the wrong outcome.
Automated bidding only works if the signals are accurate. A campaign with 8 real bookings a month and honest tracking is in better shape than one showing 40 conversions made of button clicks, and it will get to genuine readiness sooner, because it is measuring something real while it grows.
If you are not certain your conversion actions are clean, that is the prerequisite, and we wrote about what goes wrong there in the first piece of this series.
Every Change You Make Restarts the Clock
Once automated bidding is on, it is not a switch. It is a system in a state, and that state is fragile in ways the interface never mentions.
Material changes push a campaign back into learning: changing bid strategy, moving your target substantially, large budget swings, significant structural edits. During learning, performance is genuinely unreliable, and judging results in that window tells you nothing.
We manage a coffee equipment account where the true cost per acquisition had settled around $250 while the target was still set at $200. That gap is a problem: when your target sits meaningfully below what the account actually achieves, the algorithm gets throttled and can effectively stall. Raising the target to $250 and lifting the daily budget to match fixed it, but the change reset learning and the following week dipped before recovering.
That dip was not the change failing. It was the system recalibrating.
Which leads to the rule that saves the most accounts: judge automated bidding over a month, never a week. There is natural ebb and flow. Google will overspend one day and pull back the next to hold the target. A bad week inside a good month is noise, and reacting to it with another change starts the learning period over, which produces another bad week, which invites another change.
Two practical notes that come up constantly:
- Pausing is survivable, briefly. Up to about 10 days off and a campaign generally picks back up where it was. Past that, expect it to re-enter learning and underperform for a stretch.
- Scale down gradually rather than switching off. Reducing budget in steps when you are at capacity is fine. Turning everything off and back on is what does the damage.
The Signal That Beats Anything On Your Website
There is a tier above website conversions, and almost nobody uses it.
Your site can tell Google that a form was submitted. It cannot tell Google whether that lead was any good. Two form fills look identical to the algorithm: the one that became a $40,000 order and the one that was a student doing homework.
If you have a CRM, you can send back what actually happened. The lead that became a customer gets reported as a conversion worth more than the one that never answered the phone. Now the algorithm is not optimizing toward form fills. It is optimizing toward form fills that turn into revenue.
This changes the math on the 30-conversion threshold too. Fewer, better-qualified signals with real outcomes attached can outperform a larger pile of undifferentiated ones. It takes setup work, and it is worth it for any business where lead quality varies a lot, which is most businesses selling anything considered.
How to Know You're Actually Ready
Four checks. All of them, not three.
1. Thirty conversions in the last 30 days, in the campaign you want to automate. Open the campaign, set the date range to the last 30 days, and read the number. Not the account. Not last quarter's best month.
2. Those conversions are real. Segment your conversions column by conversion action and look at what is actually in there. If a meaningful share is add-to-cart, page views, or clicks, you do not have 30. You have some smaller number plus padding.
3. The volume is steady, not a spike. Thirty conversions because of one promotion is not a pattern. Look at three consecutive months.
4. You know what you can afford to pay. Target CPA means naming a number. If you have not worked out what a customer is worth and what you can spend to acquire one, you will pick a target out of the air, and a target set too low will strangle the campaign as surely as no data at all.
When those four are true, move. Start with Target CPA set near your current actual cost per acquisition rather than your hoped-for one, give it a month without touching anything, and judge it then.
One naming note, because it causes confusion: Google renamed "Maximize conversions with a Target CPA" to simply "Target CPA" in mid 2026. Same bidding behavior, different label. If you are following older instructions and cannot find the option, that is why.
The Uncomfortable Version
Most accounts we audit are not being held back by their bidding strategy. They are running a strategy that requires evidence they never collected, and the fix is not a better setting. It is a few months of deliberately building the thing the setting needs.
That is a harder sell than flipping a switch, which is exactly why the switch keeps getting flipped.
We do this all day long, and if you would rather hand the build phase to someone who has done it a few hundred times, we can. If you would rather run it yourself, run the four checks above before you automate anything. That alone will put you ahead of most of the accounts that land on our desk.
