Algorithms treat conversion data as truth. If tracking counts twice, measures forms that were only opened or mixes low-quality enquiries with sales, the AI learns to look for exactly that. The result looks good in reports and bad in revenue.
The essentials in a definition
Weak conversion data are incomplete, duplicated, delayed records or ones that measure the wrong event. Since algorithms learn from these data, a confident AI can optimise very efficiently for a worthless result. Correct conversion measurement, complemented by enhanced conversions, is the foundation of any automation.
The garbage-in principle
Any learning system depends on the quality of its examples. In PPC, the examples are conversions. If they include events that are neither sales nor real leads, the algorithm will consider traffic that clicks but does not buy to be good.
The problem is discreet, because the numbers improve. More conversions at lower cost sound like success, even when real sales do not keep up.
Signs of bad data
Conversions that jump suddenly with no change in the business, conversion rates too high for the sector, identical values on every sale, and large gaps between platform, analytics and CRM. Any of these justifies an audit.
Another sign is a disconnect with the sales team: if they say the leads are poor but the reports say otherwise, measurement is counting what it should not.
Cleaning conversions in four steps
- List all active conversions
- Separate real value from activity
- Fix duplicates, values and firing
- Audit again every quarter
How to clean up tracking
List all active conversions and identify those that represent real value. Pause the rest or reclassify them as secondary. Remove duplicates, fix values and confirm the event fires only after the action is complete (for example after the form is submitted, not on the click).
Wherever possible, link a quality status (qualified lead, completed sale) so the system learns from outcomes and not merely from intentions.
Better numbers are not the same as better results.
Keeping data clean over time
Audit every quarter. Whenever the website changes, retest the events. Document the definition of each conversion and who may change it. Small site changes, such as a new form, break tracking without warning.
Set a simple rule: no new conversion becomes primary without testing and without two weeks of comparison against CRM data.
Decision table
| Problem | Symptom | Fix |
|---|---|---|
| Duplicated conversion | More conversions than real sales | Transaction ID and testing |
| Wrong event | Many empty forms or counted clicks | Measure successful submission or the sale |
| Inconsistent value | Impossible or fluctuating ROAS | Single value and currency rule |
| Delayed information | Offline sales arrive weeks later | Offline conversion import |
Common mistakes to avoid
- Counting everything as a conversion. Button clicks and page visits are not business results.
- Not validating after site changes. A checkout change can break measurement without warning.
- Relying only on platform numbers. Always compare with the sales system or CRM.
How to measure the result
Every month, compare the platform’s conversion count with sales in the internal system. A stable, explainable difference is acceptable; one that grows or fluctuates indicates a tracking problem to fix before optimising.
Limits and exceptions
Fixing conversion data is not always quick: it may require changes to the site, checkout or CRM, and the platform may need weeks to relearn. During that time, performance may worsen before improving. Plan the fix outside critical seasons, tell decision-makers there will be fluctuation and always compare with the sales system. If measurement cannot be fixed in the short term, simplify objectives and use campaigns with more manual control.
Illustrative example
Illustrative example: a booking service has a conversion that fires on every visit to the thank-you page, including page refreshes. The system learns to look for people who reload pages. By fixing the event to fire once per paid booking, cost per conversion rises in the report, but cost per real booking falls. The “worse” number was in fact the true one.
Frequently asked questions
How do I know my conversions are wrong?
Compare platform, analytics and CRM. Big gaps or conversions that do not match sales are warning signs.
Can I keep secondary conversions?
Yes, for analysis, but without using them as an optimisation goal.
How often should I audit?
Quarterly, and whenever the website or forms change.
How do I know my conversion data are good?
If they match, in order of magnitude, real sales and contain no duplicates or valueless events.
Does AI fix bad data on its own?
No. It learns from what it receives; bad data lead to bad optimisation, only faster.
If conversion data are bad, AI optimises efficiently for the wrong result; validate events, values and duplicates against the sales system before trusting automation.
Sources and recommended reading
- Google Ads Help: conversion measurement
- Google Ads Help: enhanced conversions
- Google Ads API: conversion management
Last editorial review: October 2026. Informational content; always confirm current conditions in the official sources before deciding.