Why did my smart bidding go flat?
Smart bidding goes flat when the signals it was trained on stop arriving. Target CPA, target ROAS, and Maximize Conversions all learn from conversion data. When a conversion action stops firing, a value field goes empty, or an offline import quietly breaks, the model keeps bidding but on weaker information. Performance rarely falls off a cliff. It drifts flat, then decays. The fix is usually restoring a signal, not changing the bid strategy.
Google Ads smart bidding went flat: the 5 signals you stopped sending
The account did not break. The data feeding it did. Most flat smart-bidding accounts are not a bidding problem, they are a measurement problem that looks like one.
1. Smart bidding does not stall on its own
Target CPA, target ROAS, and Maximize Conversions are not static rules. They are machine-learning systems that set a bid at every auction using the signals available at that moment. Google's Smart Bidding documentation is explicit about the dependency: the system learns from conversion data, and missing or delayed conversion feedback weakens the model's input.
The consequence is easy to misread. When a signal stops flowing, the auction itself does not change. Clicks and impressions can hold steady while the model quietly loses the ability to tell a good click from a bad one. The account looks healthy in the top-line numbers and slowly stops improving where it matters.
That is why the first question on a flat account is not "which bid strategy should I switch to?" It is "which signal did we stop sending?" The five below account for most of the flat accounts we diagnose.
2. Signal one: the primary conversion action
Smart Bidding optimizes toward whatever conversions you mark as primary. If the qualified-lead event gets demoted to secondary, or a soft event like a page view is marked primary, the model optimizes the wrong thing with full confidence. It is not broken, it is just aiming at the wrong target.
A common version of this: a tracking change re-creates the conversion action, the new one defaults to secondary, and the model keeps bidding toward the old, now-dead action. Clicks continue, conversions "drop," and the account looks flat. Search Engine Journal's primary-versus-secondary framework walks through exactly this failure mode.
Check this first because it is the cheapest fix in the list. Open the conversion actions, confirm the event you actually want is primary, confirm it is the only one the strategy is using, and confirm the counting method still matches how the business records a lead.
3. Signal two: conversion value, for target ROAS
Target ROAS and Maximize Conversion Value cannot distinguish a $50 lead from a $5,000 sale unless the value is passed. If values are missing, static, or wrong, the bidder has no revenue signal to optimize toward.
Watch for the value field silently emptying. A tag or data-layer change can stop passing the transaction amount while the conversion event itself keeps firing. The conversion count stays healthy, ROAS goes flat, and nothing in the account appears broken. This is the hardest of the five to catch because it leaves almost no visible trace.
Verify that the value is arriving on the conversion action and that it reflects real revenue, not a placeholder. A flat ROAS on healthy volume is, more often than not, a value field that stopped telling the truth.
4. Signal three: enhanced conversions, the first-party fallback
Browser and device privacy changes have removed a large share of the third-party identifiers smart bidding used to match. Enhanced conversions replaces some of that with first-party data, hashed email and phone, that you already collect. When it is off or misconfigured, a portion of your conversions simply stop being attributable.
Since April 2026 Google has consolidated enhanced conversions for web and leads under a single on/off setting, accepting user-provided data from website tags, Data Manager, and API connections. If that setting is off, or the hashing is broken, the model is learning from a smaller, noisier sample than it should.
Confirm enhanced conversions is on, that the user-provided data fields are populated, and that the hashing is happening before the data leaves your system. A surprising number of accounts have the toggle on while the data underneath it is empty.
5. Signal four: offline conversion imports
For businesses where the sale happens after the click, in a call, a quote, or a booked job, the click is not the outcome. Offline conversion imports close that loop by feeding the real result back to the bidder. Google's offline conversion documentation reports a median 10% increase in conversions for advertisers using offline imports with first-party data.
There is also a 2026 deadline hiding in this signal. Starting June 15, 2026, legacy offline conversion imports and enhanced conversions for leads uploads moved to the Data Manager API and were blocked in the Google Ads API for uploads. An import that has run quietly for a year can stop working on a migration date, and the account goes flat while nobody notices the upload died.
Check the import is still running, still uploading fresh data on schedule, and that the upload endpoint has been migrated if you were on the legacy path.
6. Signal five: consent mode, the invisible gap
Consent mode tells the bidder how much data is missing because a user declined tracking. Without it, the model sees a conversion drop and cannot tell whether demand fell or consent hid it. The result is a bidder reacting to a phantom.
The signal is invisible by design. Nothing in the account flags "consent signals missing." The only trace is a conversion count that understates reality and a model that bids as if the business got worse. For advertisers in privacy-regulated markets, this is frequently the gap between a flat account and a recovering one.
Confirm consent mode is configured, that the consent state is being passed on every relevant event, and that the model can distinguish granted from denied. The fix is configuration work, not spend, which is why it belongs on this list before any talk of restructuring campaigns.
7. What the model sees: healthy versus broken
| Signal | Healthy: what the model learns | Broken: what the model gets instead |
|---|---|---|
| Primary conversion action | Every qualified lead and sale, counted correctly | Bids toward a dead or demoted event, clicks look fine while conversions vanish |
| Conversion value | Real revenue per transaction, so target ROAS aims at profit | A static or empty value field, ROAS flat on healthy volume |
| Enhanced conversions | Hashed first-party data recovers consented users lost to privacy changes | A share of conversions go unattributed, the model learns from a smaller sample |
| Offline imports | Post-click outcomes feed back, closing the loop after the sale | The bidder never learns what happened after the click |
| Consent mode | The model models the gap and adjusts for hidden demand | The model treats consent loss as a real demand drop |
8. The diagnostic order: cheap to expensive
- Confirm the primary conversion action is the right event, primary, and counted correctly. Free to check, free to fix.
- Confirm the conversion value is arriving and reflects real revenue. Free to check, cheap to fix.
- Confirm enhanced conversions is on and the first-party data is actually populated and hashed. Free to check.
- Confirm offline imports still run on schedule and have been migrated off the legacy endpoint if needed. Free to check.
- Confirm consent mode is configured and passing consent state. Configuration work, no added spend.
Only after all five are clean should the conversation turn to bid strategy, budget, or creative. A widely repeated practitioner threshold, not an official Google rule, is about 30 conversions a month for target CPA and 50 for target ROAS. Below that, the model has less to learn from, and results are noisier no matter how clean the signals are.
9. When it actually is the strategy
Sometimes the signals are clean and the account is still flat. That is the moment to consider the strategy, not before. A target CPA set below what the market can deliver, or a target ROAS with no historical support, will pin the bidder to a point it can rarely win. But changing the strategy while a signal is broken just moves the flat line somewhere else.
Restore the signals first, give the model time to re-learn, then judge the strategy. A conversion or value change typically needs days to weeks, not hours, before the flat line moves. Watch the trend, not a single day.
Why Google Ads smart bidding stopped improving: the five missing signals
Because the data feeding the bidder changed, not the bidder. Smart bidding is only as good as the conversions, values, first-party data, offline outcomes, and consent signals it receives. When one of those stops flowing, the account drifts flat while looking healthy. The fix is restoring the signal, then giving the model time to re-learn, before touching the bid strategy.
Fair questions
Does smart bidding need a minimum number of conversions to work?
A widely repeated practitioner threshold is about 30 conversions a month for target CPA and 50 for target ROAS. It is a rule of thumb, not an official Google limit. Below that, the model has less data to learn from and results are noisier.
Should I switch to manual bidding if smart bidding goes flat?
Usually not first. Check whether a signal stopped flowing before abandoning the strategy. Manual bidding removes the data problem but also removes the auction-time optimization. Restore the signal, then decide.
How long should I wait after fixing a signal before judging the results?
Give the model time to re-learn. A conversion or value change typically needs days to weeks, not hours, before the flat line moves. Watch the trend across several days, not a single day.