Conversion tracking that counts real inquiries

An account that counts the wrong events will optimize toward them with total commitment. This is the failure I find most often and the one that costs the most, because it is invisible from the dashboard. The graphs go up. The reports look healthy. The phone does not ring any more than it did before, and nobody can explain why.

The events that are not conversions

Here is what I regularly find sitting in the primary conversion column, all of them counting toward the number that Smart Bidding chases.

• Thank you page views that fire on page load rather than on submission, so bots and refreshes count.
• Form starts or field focus events tagged as conversions during a tracking setup that was never finished.
• Phone number clicks counted as calls. On mobile a click is not a call. Some fraction never dial, and some dial and hang up.
• Time on page or scroll depth imported from analytics because someone wanted more data points.
• Newsletter signups and PDF downloads sitting alongside sales inquiries with equal weight.
• The same conversion counted twice through both a direct tag and an analytics import.

Any one of these will pull bidding toward the cheapest audience that produces that event. Cheap events attract cheap traffic. The account learns to buy people who load pages and never buy anything.

What a real conversion looks like

A conversion should be an event that a human being at your company would recognize as a lead. A submitted form with contact details you can act on. A phone call that lasted long enough to be a conversation, not a wrong number. A booked appointment. A completed purchase. If your sales team would not put it in the pipeline, it should not be a primary conversion.

The primary and secondary distinction is the tool for this. Primary actions drive bidding. Everything else goes to secondary, where it is still visible in reporting but does not steer the machine. Most accounts should have one or two primary actions. If you have seven, you have no target at all.

How to fix tracking, in order

1. Audit what exists. Open the conversions table. For each action, check the source, the counting setting, and whether it is primary. Note the last recorded conversion date. Actions that have not fired in months are usually broken rather than unpopular.
2. Set counting correctly. Lead generation actions should count one per click. Ecommerce should count every purchase. Getting this backwards inflates lead accounts substantially.
3. Move everything that is not a real inquiry to secondary. Do it in one change, then leave the account alone for two weeks. Bidding will re-learn, and performance will look worse before it looks honest.
4. Fix the fire trigger. Conversions should fire on successful submission, confirmed by the form handler, not on arrival at a URL that anyone can reach directly.
5. Set call tracking to a meaningful duration. Sixty seconds is a reasonable floor for most service businesses. Below that you are counting hang-ups.
6. Turn on enhanced conversions for leads. This sends hashed first-party data so that conversions can be matched when cookies fail, which is now routine rather than rare.
7. Implement consent mode properly if you serve the EEA or the UK. Consent mode v2 with the ad_user_data and ad_personalization parameters is a requirement there, and Google has tightened enforcement steadily through 2025 and 2026. Without it you lose both audience building and modeled conversions, and enhanced conversions will not work as intended.
8. Import the outcome, not just the inquiry. If you use a CRM, feed qualified and closed statuses back as offline conversions. This is the step almost nobody does, and it is the difference between optimizing for leads and optimizing for customers.

Where it gets genuinely hard

Some businesses cannot cleanly track the thing that matters. A restaurant, a walk-in clinic, a shop. If the sale happens offline and you have no booking system, you are measuring a proxy no matter what you do. In those cases I prefer a small number of honest proxies, calls over sixty seconds and direction requests, and I set expectations that the account will be steered partly by what the owner observes in the business rather than by the platform alone.

Low-volume accounts have a different problem. If you get a handful of real inquiries a month, that is not enough signal for Smart Bidding to work well on the sale itself. Here it can be correct to keep a higher-volume secondary event as a bidding signal, provided it correlates with real inquiries and you check that correlation rather than assuming it.

What to do next

Open the conversions table today and look at one column: which actions are primary. Then ask, for each of them, whether you would pay cash for that event happening. Everything that fails the test goes to secondary this afternoon. Expect the reported numbers to fall. That fall is not a loss. It is the first accurate reading the account has produced.