Your conversion tracking is wrong when three numbers for the same week refuse to agree and nobody on the team can explain why. Pull conversions from the ad platform, conversions from analytics, and actual leads or orders from your CRM. Those three will never match exactly, but they should sit in a stable relationship you can describe out loud. When they do not, every bid, budget, and creative decision sitting on top of them is being made on fiction.
This matters more than it did five years ago. Google and Meta both bid automatically now, toward whatever you told them to count. Bad tracking no longer just misreports performance. It actively teaches the platform to buy the wrong thing, faster and at scale.
The three-number check
Pick one closed week. Not the current week, and not a week that contains a launch, an outage, or a holiday. Then write down three numbers side by side.
- Conversions reported by each ad platform, counted by the date of the click.
- Conversions or key events reported by analytics for the same channels.
- Real leads, bookings, or orders in the CRM or booking system, filtered to the same sources.
The platform number is usually the highest, analytics is usually in the middle, and the CRM is usually the lowest. That is normal. View-through windows, cross-device modeling, consent, and spam filtering all pull in different directions. What matters is whether the gaps are stable and explainable, not whether they are zero.
Repeat the exercise for the four weeks before it. If the relationship between the three numbers holds steady, your tracking is probably fine and your problem is elsewhere. If the gaps swing week to week, or one number is a multiple of another, something in the measurement layer is broken.
Six symptoms worth checking today
Conversions outnumber the humans. The account reports 180 conversions and sales worked 40 leads. That gap is almost always one event firing several times per visit, or several different events all mapped to the same conversion action.
The number is flat. A conversion count that lands within one or two of the same figure every week, regardless of spend, is not measuring behavior. It is measuring something automatic, like a page load or a bot.
Everything is credited to one page. When the thank-you page or the homepage owns nearly every conversion, the event is probably firing on page view rather than on the action you care about.
Two platforms both claim the same sale. Add the conversions from Google, Meta, and TikTok. If the total is meaningfully larger than the orders in the CRM, each platform is claiming the same outcomes and no cross-channel view exists to catch it.
Mobile and desktop rates look nothing alike. A large unexplained split usually means a consent banner, a tag that fires late, or a form that behaves differently on a phone. It rarely means mobile users are five times less interested.
Nobody has touched the setup since the last site launch. Rebuilds break tags quietly. So do form plugin updates, new booking tools, and CRM migrations. If the tracking has not been checked since something structural changed, treat it as suspect.
How to confirm it in about an hour
Start at the end of the funnel and work backward. Submit a real lead through the live site on a phone, using a name you will recognize, and then follow that single record everywhere.
Watch it arrive in analytics through the debug view while you submit. Confirm it fires once, on the action, with the parameters you expect. Then check whether it reached the ad platform, and how long that took. Then find it in the CRM and confirm the source recorded there matches the campaign you actually clicked.
While you are in the accounts, look at each platform's own tag diagnostics for warnings about inactive tags, missing order identifiers, or duplicate events. Those warnings are frequently sitting there unread and describe the problem precisely.
One live test rarely proves the whole setup is healthy. It reliably proves when it is not.
What to fix, in order
Decide what counts. Most accounts drift because everything got marked as a conversion: form fills, phone clicks, PDF downloads, newsletter signups. Pick the one action that means real revenue is likely, count that as the primary conversion, and demote the rest to secondary events you still measure but never bid toward.
Deduplicate. Pass an order or lead identifier with every conversion so the same outcome cannot be counted twice. This is the single highest-value fix in most retail and ecommerce accounts.
Fix the plumbing before the reporting. A dashboard built on broken events is a faster way to be confidently wrong. Our guide to the website tracking setup every marketing site needs covers the events, consent handling, and naming that keep this stable.
Then close the loop. Once the counting is trustworthy, send qualified outcomes back to the platforms so bidding optimizes toward revenue rather than raw form fills. That process is laid out in optimizing for qualified leads.
Keep the sources clean. Inconsistent campaign tagging recreates the same confusion a quarter later. UTM tracking, done properly is the short version, and fixing attribution chaos covers what to do when the platforms and the CRM still disagree.
If the footprint is large, several platforms, several accounts, several locations, this reconciliation stops being an afternoon of work. That is the case for an outside paid media audit, where the tracking layer gets reviewed alongside structure, budget, and creative, and every finding comes back with a dollar estimate attached.
Related questions
Should the ad platform and analytics ever match exactly? No. They use different attribution models, different windows, and different rules about what counts as a conversion. Aim for a stable, explainable gap rather than an identical number.
How often should conversion tracking be checked? Quarterly for a stable setup, and immediately after any website rebuild, form or CRM change, platform migration, or consent update. Those four events cause most of the breakage we see.
Can broken tracking actually make performance worse? Yes. Automated bidding buys more of whatever you tell it worked. Feed it the wrong conversions and it spends efficiently in the wrong direction, which is more expensive than not using automation at all.
Start with a Diagnostic Call
If your numbers do not reconcile and you want a second read on why, we can look at it together. See how paid media works here, or Start with a Diagnostic Call.