Why Your Google Ads Smart Bidding Is Not Working

You switched to Target CPA six weeks ago. Cost per acquisition went up. You extended the learning period, then raised the target, then lowered it again, and the account still performs worse than it did on manual bidding.

The instinct at this point is to blame the bid strategy. It is almost never the bid strategy.

Smart bidding is a prediction engine trained entirely on the conversions you report back to it. When it underperforms, the overwhelmingly likely cause is that you are teaching it something untrue, a few thousand times a month, with total confidence.

Why smart bidding fails when conversion data is bad

Every automated strategy works the same way. Google observes an auction, predicts the probability that this particular user in this particular context will convert, and bids accordingly. That prediction is built by correlating past conversions with everything Google knows: query, device, time, location, audience signals, and hundreds of features you never see.

The model is genuinely sophisticated. It is also entirely dependent on the labels you provide. Every conversion you report is a training example that says "this was a success, find more like it."

Report a conversion that did not happen and you have created a false positive. Report the same conversion twice and you have doubled the weight of that example. Report a page view as a purchase and you have told the model to optimize for browsing.

None of this produces an error message. The model trains on what it receives and spends your budget on the resulting predictions.

How smart bidding actually works and what it needs

Three things determine whether automated bidding can succeed.

Volume. The model needs enough conversions to find patterns. Google's rough guidance is thirty conversions in thirty days for Target CPA and fifty for Target ROAS. Below that, the model is fitting noise. If you are short on volume, a conversion action further up the funnel is often a better training signal than a sparse one at the bottom.

Accuracy. Every conversion must correspond to a real business outcome. This is where most accounts fail.

Consistency. The definition cannot keep changing. Every time you add, remove, or redefine a conversion action, you invalidate part of the model's training and restart learning.

That third point explains a common pattern: an account that never stabilizes because someone adjusts the conversion setup every few weeks in an attempt to fix performance.

The most common conversion tracking problems that break smart bidding

Duplicate conversions. The classic cause is a conversion tag deployed in both the site template and Google Tag Manager. Everything is counted twice. Your reported CPA halves, which looks like success, so budget shifts toward whichever campaigns show the strongest numbers. Those campaigns are not better, they are inflated at the same rate as the rest, so you have reallocated budget on noise.

Conversions firing on page load. A tag set to fire when someone reaches a thank you page also fires when they refresh it, navigate back to it, or arrive from a bookmark. The model builds an audience of people who visit a URL, which is a different population from people who bought.

Everything marked primary. In Google Ads, conversion actions marked as primary are what bidding optimizes toward. When newsletter signups, PDF downloads, and purchases are all primary, the model treats them as equivalent, and it will efficiently find you a great many newsletter signups.

Missing or zero transaction values. Value based strategies need real revenue figures to distinguish a two hundred dollar order from a two thousand dollar one. When a data layer variable stops being populated, the tag still fires and sends nothing. Target ROAS then optimizes for volume while you believe it is optimizing for revenue.

Loose trigger conditions. A URL condition using "contains" rather than "equals" catches every path sharing that string, so a purchase conversion quietly starts including newsletter confirmations.

Broken deduplication. When you run both a browser tag and the server side or Conversions API path, both need a consistent event ID. Without it, every conversion reports twice.

How to audit your conversion actions in Google Ads

Open Tools, then Conversions, and work through the list deliberately.

Check the Include in "Conversions" column. Only actions that genuinely represent business value should be included. This is the single highest impact setting on the page, because it determines what bidding chases.

Check the counting setting. For lead generation, use "One" so a single prospect submitting three forms counts once. For ecommerce, use "Every" so a repeat purchase counts properly. Getting this backward is common and quietly distorts everything.

Check the attribution window. A thirty day window on a product with a same day purchase cycle credits conversions to clicks that had nothing to do with them.

Look at the status column. Actions showing "No recent conversions" are either broken or obsolete, and both are worth resolving.

Reconcile against your backend. Take a full month and compare Google Ads conversions against your CRM or ecommerce platform. Some variance is expected because of attribution differences. A gap approaching double is duplication.

Then verify in the browser. Run GTM preview mode, complete a conversion yourself, and confirm the tag fires exactly once with a populated value.

What to fix first

Work in this order, because each fix makes the next one measurable.

Remove duplicates. Highest impact and usually the simplest. Find the tag firing twice and eliminate one instance. Expect reported conversions to fall, which is the correct outcome.

Demote secondary actions. Set anything that is not a genuine business outcome to secondary. It still reports, it just stops steering bids.

Fix triggers. Move conversions off page load and onto real interactions. Tighten loose URL conditions.

Restore values. If revenue is arriving as zero or undefined, fix the data layer before touching any value based strategy.

Then wait. Do not adjust targets during the recalibration. You need clean data flowing for long enough that the model retrains on it.

How performance improves when conversion data is clean

The first thing that happens is that your numbers get worse. Reported conversions drop, sometimes by close to half, and cost per acquisition rises to match. Nothing about your actual business changed. You are simply no longer counting phantom conversions.

This is the moment most teams panic and revert. Do not.

Over the following two to four weeks, the model retrains on accurate examples. Bidding stops overpaying for auctions that only appeared to perform. Budget stops flowing toward campaigns that were winning on double counted events. Value based strategies become usable because revenue signals are intact.

The end state is not a dramatic improvement in the numbers you report. It is that the numbers you report finally correspond to what your business received, and every decision made from them is sound.

The uncomfortable conclusion

If smart bidding is underperforming, resist the urge to change the strategy. Audit the signal first. The campaign settings are almost never the problem, and no amount of bid strategy tuning compensates for an algorithm learning the wrong lesson at scale.

FAQ

Why is my Target CPA not working?

Usually because the conversion data feeding it is inaccurate rather than because the target is wrong. Duplicate conversions, conversions firing on page load, and non revenue actions marked as primary all teach the model to chase the wrong outcomes. Audit your conversion actions and reconcile Google Ads totals against your backend before adjusting the target again.

How many conversions does smart bidding need?

Google suggests roughly thirty conversions in thirty days for Target CPA and around fifty for Target ROAS. Below those levels the model is fitting noise rather than finding patterns. If your bottom of funnel volume is too low, optimizing toward a reliable mid funnel action often produces better results than a sparse purchase signal.

Should all conversion actions be marked primary in Google Ads?

No. Only actions representing genuine business value should be primary, because primary actions are what bidding optimizes toward. Newsletter signups, PDF downloads, and scroll events should be secondary. They still appear in reporting, they simply stop steering your bids. Most underperforming accounts have too many primary actions.

Why did my conversions drop after fixing tracking?

Because you removed duplicates that were inflating the count. Reported conversions commonly fall by a third to a half after deduplication, while actual sales stay flat. Annotate the change date so future trend analysis does not mistake it for a performance decline, and expect two to four weeks before bidding fully recalibrates.

How long does smart bidding take to learn?

The formal learning period is typically one to two weeks, but full recalibration after a conversion tracking change takes closer to two to four weeks depending on volume. Higher volume accounts stabilize faster. Avoid changing targets, budgets, or conversion definitions during this window, because each change restarts the process.