Smart Bidding is doing exactly what you configured it to do, and that is the problem
Most Google Ads accounts feed their smart bidding algorithms the wrong conversion data, and the algorithm confidently optimises toward the wrong goal. Conversions are high, ROAS looks acceptable, but the conversion mix includes form views, scroll depth, and page visits alongside actual revenue events. The bidder learns toward the average signal. Auditing primary vs secondary conversion actions takes 20 minutes. Most accounts have never done it.
Most Google Ads accounts are feeding their smart bidding algorithms the wrong conversion data, and the algorithm is confidently optimising toward the wrong goal.
The problem starts with what counts as a conversion. A lot of accounts have multiple conversion actions tracked at the account level: phone calls, form fills, page visits, scroll depth, even session duration. When smart bidding reads all of these as signals, it learns toward the average signal, which includes a lot of noise.
The result is a bidding strategy that looks like it is working. Conversions are high. The algorithm is triggering frequently. ROAS looks acceptable. But if you pull apart the conversion mix and look only at the actions that actually correlate with revenue, the story changes. The algorithm optimised for whatever was numerically dominant, not whatever was commercially meaningful.
Primary vs secondary conversion actions
The fix is not complicated. Google has a "primary" and "secondary" conversion action setting that most accounts do not use correctly. Primary conversions feed the bidding algorithm. Secondary conversions are for reporting only.
If you have soft conversions sitting in your primary column, they are actively degrading your bidding signal quality.
Common offenders we see in audits:
- A "Newsletter signup" conversion fired by a footer-form widget, set to primary, weight 1. The bidder learns that anyone willing to give an email is a worthwhile target. It bids harder on the audience pattern that gives emails. That audience converts to revenue at a tenth of the rate of your actual buyers.
- A "Scroll depth 75 percent" event set as a conversion, primary, weight 1. The bidder optimises toward people who scroll. Scrolling is not buying.
- A "Time on site 60 seconds" engagement event treated as a conversion. The model is now optimising toward bounce-resistant traffic, not converting traffic.
- Multiple form conversions counted independently: "Quote request", "Contact us", "Demo booking" all primary, all weight 1, all triggered by the same form on the same thank-you page through different paths. The bidder counts each conversion two or three times.
- Phone-call conversions set to primary on accounts where the call volume is mostly support and customer service calls, not new-business calls.
Each of these individually looks defensible. In aggregate, they tell the bidder that "engagement" matters more than "revenue". The bidder is right to follow that instruction. The instruction was wrong.
The 20-minute audit
For each tracked conversion action, ask one question: if the algorithm maximises this, does it maximise revenue?
If the answer is no, the action should not be primary. Either move it to secondary (still tracked, not used for bidding) or remove it entirely.
The audit is mechanical:
1. Open Conversions in Google Ads 2. Sort by Last 30 days, all enabled actions 3. For each action: is this firing on an event that corresponds to actual revenue? If yes, primary. If it correlates with revenue but is not revenue itself (a lead, a qualified call), primary with a sensible value. If it is a soft engagement event (page view, scroll, time), secondary or delete. 4. Re-check after 30 days. If the bidder has stabilised on the new signal mix, performance usually improves measurably.
The accounts that get this right run the audit quarterly, not as a one-off. Conversion actions accumulate. Someone adds a "View pricing page" event for reporting and forgets to set it to secondary. Six weeks later, the bidder is partly optimising toward that.
Why this matters more in 2026
Smart Bidding has always rewarded clean signal. With AI Max for Search now mandatory from September, the auto-bidding layer has more authority over more of your campaigns than ever before. The same algorithm is making more decisions with the same conversion configuration you set up two years ago.
If your conversion configuration was good enough for manual bidding, it may not be good enough for AI Max. The cost of bad configuration is now amplified by the volume of decisions the algorithm makes on your behalf.
What to actually do
- Audit your conversion action configuration this month. List every action, mark each as primary or secondary, justify in one sentence.
- Tie at least one primary conversion to actual closed revenue via offline conversion import. The closer the bidder's training signal is to your real revenue, the better the bidding becomes.
- Set up alerts for changes to conversion actions in the account. Anyone (including the platform itself via auto-tagging) modifying conversions should generate a notification.
- Re-audit quarterly. Drift is inevitable. The point is to catch it before it changes campaign behaviour.
- If you cannot answer "what is the bidder actually optimising for?" in one sentence, the configuration needs work before any campaign-level change matters.
When did you last audit which conversion actions are feeding your bidding strategies?
If you want a free audit of which conversion actions in your account are degrading the bidding signal, book a free audit and we will rank them by impact before any other recommendation.
Get a free PPC audit from the team that wrote this.
We'll review your Google Ads or Microsoft Ads account and show you three specific things we'd change in the first 30 days.
