Someone in the meeting has three numbers for the same month: the ad platform reports one conversion count, analytics reports a lower one, and the CRM shows a different figure again. The instinct is to ask which one is right.
None of them is, because that is the wrong question. The three systems are not counting the same thing, so agreement was never available — the useful skill is reading the gap, not closing it. A stable, explainable discrepancy is a healthy measurement setup. A discrepancy that suddenly changes size, or one you cannot account for, is the signal worth investigating. This explainer covers what each system actually counts, the structural reasons they diverge, and how to tell a definition difference from a genuine tracking failure.
What each system is actually counting
Before comparing numbers, name what each one measures. Most reconciliation arguments dissolve here.
- The ad platform counts conversions it can credit to its own clicks or impressions, within its own lookback window, using its own credit rules. It is answering: "of the conversions that happened, how many can I connect to an ad interaction I served?"
- Your analytics platform counts events it observed on your site, grouped into sessions, and assigns each to a traffic source using its model. It is answering: "what happened on the site, and where did the visitor appear to come from?"
- Your CRM or order system counts records that a human or a payment created. It is answering: "what business actually landed?" — usually with no view of the marketing path at all.
Three different questions, three different denominators — and therefore three different totals.
The structural reasons the numbers diverge
Six mechanics account for most of the gap. None of them is a bug.
1. Attribution windows differ
Every platform has a lookback window — how long after an interaction a conversion still gets credited. If one system looks back further than another, it will naturally claim more conversions, including ones that landed weeks after the click. Windows are configurable in most platforms and differ by default between them, so check the current setting in each platform's own documentation rather than assuming they match.
2. The credit model differs
An ad platform generally credits conversions to itself when it saw a qualifying interaction; an analytics platform distributes credit across the path it observed using whichever model is selected. Last-click, first-click, position-based, and data-driven models each brighten different channels. Compare a self-crediting platform total to a multi-touch analytics total and the paid numbers will diverge by design. Which lens you are looking through is the first thing to state out loud — the deeper treatment is in our marketing measurement guide.
3. The date a conversion is stamped with
Ad platforms commonly report a conversion against the date of the click that earned it; analytics and CRMs report it on the date it happened. For a same-day sale that's identical. For a considered purchase, the same event lands in two different rows — possibly two different months. This alone can make a month-end comparison look broken, then quietly resolve as later conversions backfill.
4. Definitions and deduplication
"Conversion" is a configuration, not a fact. Systems differ on whether they count every conversion or one per person, whether multiple conversion actions are counted together, and how they deduplicate repeat submissions. A form that fires on both submit and thank-you page load double-counts in one system and not another. Before comparing totals, confirm both sides are counting the same action the same number of times.
5. Consent, blocking, and modelled data
Analytics generally requires a tag to fire in the visitor's browser. Consent choices, tracking protection, ad blockers, and network failures all suppress some share of those events — so analytics tends to under-count relative to systems that record conversions server-side, like a CRM or payment processor. Meanwhile, platforms increasingly estimate conversions they could not directly observe. That's a comparison between a partly observed number and a partly modelled one: a difference in kind, not a rounding error.
6. Cross-device and view-through paths
A visitor who clicks on a phone and converts on a laptop breaks a session-based count but may still be joined by a platform with a logged-in identity graph. Some platforms also credit conversions after an ad was seen but not clicked, where analytics sees only a later direct or organic visit. Both produce conversions in one system with no matching row in the other.
How to read the gap in practice
Reconciliation is a reading exercise, and it goes faster in this order.
- Pick a reference. Decide which system is your source of truth for reported results — usually the one closest to money (CRM, orders, payments). Everything else is a lens onto it, not a competitor.
- Compare like for like. Same date range, time zone, conversion action, currency, and — where you can — the same attribution window. Most "impossible" gaps shrink once the comparison is fair.
- Measure the ratio, not the difference. Track the gap as a proportion over several periods. A steady ratio is a definition difference you can annotate and stop re-litigating; a moving ratio is the actual finding.
- Spot-check one conversion end to end. Take a known order and follow it through all three systems. One traced record teaches you more than a month of aggregate comparison.
- Write down the explanation. Record why the systems differ and roughly by how much. Reconciliation is only expensive the first time; undocumented, next quarter's team pays it again.
When the gap is a bug, not a definition
Definition gaps are stable and directional. Suspect a real problem when you see:
- A sudden change in the ratio with no change in campaigns, tags, or consent setup — especially a step change on one date.
- Zero conversions where there were some, which usually points at a broken tag, a changed page URL, a consent-banner change, or an expired container.
- A near-exact doubling, the classic fingerprint of a duplicate tag or a conversion firing twice in one flow.
- A gap in one channel only. Definition differences affect everything a platform measures; a single-channel anomaly points at tagging or missing campaign parameters on one source.
- Conversions the CRM cannot corroborate at all, worth checking for form spam and bot traffic before it reshapes a budget.
Date the change, then line it up against what you altered and what the platform altered. A shift that starts on the same day as a tracking edit, a consent change, or a platform release note has explained itself — the same classify-before-you-react posture that governs responding to ad-platform changes.
What to report, and to whom
Once the gap is understood, use each number for the job it is good at.
- Report business results from the system of record. Revenue, orders, and qualified leads come from the CRM or payment system, because those numbers are not subject to platform definitions.
- Optimize inside a platform using that platform's numbers. Bidding and creative decisions are comparisons within one lens, where its own conversions are the correct input — even if the total differs from your CRM.
- State the lens on every dashboard. Which model, which window, which source. Two people reading one chart through two assumed models will disagree indefinitely and never find out why.
FAQ
Why does my ad platform report more conversions than analytics?
Usually a combination of a longer lookback window, self-crediting, credit for conversions seen but not clicked, cross-device joins analytics cannot make, and modelled conversions filling in what could not be observed. Analytics also loses events to consent choices and tracking protection. Platform higher than analytics is common enough to be unremarkable.
What size of discrepancy is normal?
There is no universal number, and anyone quoting one for your setup is guessing. What matters is that the gap is explainable and stable in proportion. A consistent ratio is normal; a ratio that changes without a change in campaigns, tagging, or consent deserves investigation.
Should I try to make the numbers match exactly?
No. Exact agreement between systems with different windows, models, and collection methods isn't achievable, and pursuing it burns time you could spend on decisions. Explain the gap, document it, and monitor its stability.
How do I know whether tracking is broken or the definitions just differ?
Definition gaps are steady and affect everything a platform measures; breakage tends to be sudden, dated, and localized to one channel, form, or page. Trace a single known conversion end to end, then check whether the change lines up with a tagging edit, a site change, or a documented platform change.
Read the gap, then watch what moves it
Three numbers that disagree are not a crisis — they are three lenses reporting honestly on different questions. The teams that stay calm name a system of record, compare like for like, track the ratio rather than the difference, and write the explanation down once. What they watch for afterwards is whatever moves that ratio: an attribution default that changes, a consent rule that tightens, a platform that starts modelling more of what it can no longer see. Those shifts land in your reports before anyone announces them internally. Track every analytics, attribution, and ad-platform change as it happens on Moz News, clustered from trusted sources with every source shown.
