Skip to article
Growth & Acquisition

Why Your Marketing Numbers Don't Match—and How to Decide Where to Spend

Google Ads, GA4, Meta and your CRM rarely agree. Learn why, and use the Attribution Disagreement Ledger to decide budget before the numbers converge.

· Zenko Protocol · 9 min read
Zenko fox compares advertising, analytics and CRM reports beside a shared ledger.
Illustrative artwork. Reporting settings and examples are explained in the article.

Your Google Ads account says one thing, GA4 says another, Meta claims a third, and the CRM shows a fourth figure for pipeline. Finance has a fifth. A mismatch can indicate a tracking fault, but it can also arise because each system counts different things, under different rules, at different times. Check both possibilities.

Exact agreement is not always a meaningful goal. You can explain each gap, decide in advance which number governs which decision, and set a dated review rule before you move money. That is what the Attribution Disagreement Ledger on this page is for.

This guide is for marketing and growth leads, and for the finance partners who have to sign off budget. It does not claim to settle which platform is right, and it sets no universal tolerance for how big a gap is acceptable.

Why each system counts differently

Most marketing attribution discrepancies come from a short list of causes. Each is covered below with the platform documentation that supports it. The documentation was read on 18 October 2026.

  1. Credit rules. Each system decides how to share credit between touches. Google Ads currently supports last click and data-driven models, and describes data-driven as the default for most conversion actions (About attribution models, Google Ads Help).
  2. Windows. Each system looks back a set number of days from a conversion. Different windows give different counts. The attribution windows guide explains this in detail.
  3. Date basis. Google Ads standard conversion reporting uses the date of the attributed ad interaction rather than necessarily the day the conversion happened (Understand your conversion tracking data, Google Ads Help). A tool that reports by conversion date will place the same sale on a different day.
  4. Modelled conversions. Google explains that some reported conversions are modelled, including cross-device estimates where direct observation is unavailable (same page). Consent-mode modelling can also add modelled conversions to the Conversions column (About consent mode modeling, Google Ads Help).
  5. Counting settings. A conversion action can count every conversion or one per interaction (Google Ads Help, 6270625).
  6. What the system can see. A CRM only credits interactions it can observe. HubSpot, for example, counts ad clicks only from connected Facebook, Instagram, LinkedIn or Google ad accounts, and its revenue-attribution documentation excludes outside interactions unless they use UTM-tagged links to tracked pages (Understand attribution report definitions, HubSpot Knowledge Base). See why your CRM and ad platforms disagree.
  7. Reporting lag. Google notes that reporting lags can take up to 24 to 48 hours (Google Ads Help, 6270625). In a business with a long sales cycle, CRM stage dates can also be restated later. For comparing channels at the same cohort age, see compare channels at the same cohort age, which Zenko published on 4 October 2026.

For the Google Ads and GA4 pair specifically, see why Google Ads and GA4 report different conversions.

The rules change underneath you

Historical comparisons are less safe than they look, because platforms change their rules without asking you.

  • Google Ads. Google states that the first click, linear, time decay and position-based models have been retired and affected conversion actions moved to data-driven attribution (Google Ads Help, 6259715). The page shows no effective date, so check your own account's change history. Whether your history is affected depends on your settings.
  • GA4. Google's page says changing the reporting attribution model affects both earlier and subsequent data in applicable event-scoped reports; user- and session-scoped traffic dimensions are unaffected, whereas lookback-window changes affect subsequent reporting rather than rewriting earlier data (Select attribution settings, GA4 Help). Within one product, one setting rewrites history and another does not.
  • Meta. On 3 March 2026 Meta announced that click-through attribution for website and in-store conversions would be limited to link clicks, with shares, saves and other non-link clicks moving to a category renamed "engage-through", and that the engaged video view threshold would fall from 10 seconds to 5 seconds. Rollout was staggered from later in March 2026, and Meta said billing was unchanged (Simplifying Ad Measurement for a Social-First World, Meta for Business). Meta gave reducing misalignment with tools such as Google Analytics as its reason.

The practical lesson is to record the setting and the date of each change next to every number you log.

Attributed is not the same as incremental

An attribution report tells you which touches a system credited. It does not tell you what would have happened without the spend. In a 2019 study, Gordon et al. found that observational estimates often differed from randomised experimental results even after using extensive observed data. This summary relies on the Kellogg research abstract, not a claimed full-paper review.

That study covered 15 US experiments on one platform and is not recent. Use it for the principle only. It does not size the gap in your account. For lift testing and marginal CAC, see the paid-growth saturation test rather than treating a ledger as a substitute for an experiment.

The Attribution Disagreement Ledger

Download the Attribution Disagreement Ledger (CSV templates). Includes both linked tables and a decision record; no sign-up required.

The ledger puts every source's number for one channel side by side, names the cause of each gap, and records what you agreed to do about it before you saw the result. Copy the two tables below into a spreadsheet. Both use the same Row ID.

Zenko supplies no thresholds. Trust tiers, gap bands and actions are defined by your team and written down before you read the numbers. Nothing in this template is a benchmark.

Table A: what each system says

Row ID Channel or campaign Period and cohort age at read Source system Metric and its definition in that system Reported value Attribution model Click, engaged-view and view windows Date basis (interaction or conversion time) Includes modelled conversions? (Yes, No, Unknown) Counting (one or every) Date settings last changed
1
2
3

Table B: why they differ and what you will do

Row ID Reference source for this decision Gap against reference source Touches the system cannot see Named cause(s) Evidence for the cause Trust for this decision (team-defined tier) Pre-agreed action (hold, cap, test or cut) Owner Next review date Restatement notes
1
2
3

Illustrative example (hypothetical, invented round numbers)

This example is invented to show how the columns work. It is not Zenko or client data and no figure in it is a benchmark.

Row ID Source system Reported value (illustrative) Named cause Trust for this decision (team-defined) Pre-agreed action
1 Ad platform 100 conversions Wider window and view-through credit Low for budget cuts, medium for trend Test
2 GA4 70 conversions Different model and channel scope Medium Hold
3 CRM 40 opportunities Cannot see untagged touches; cohorts still maturing High for pipeline, immature for this period Hold until review date

The three figures are not supposed to match, because 100 conversions, 70 conversions and 40 opportunities are different things. The ledger's job is to stop anyone comparing them as if they were the same unit.

How to use the ledger

  1. Pick one channel. Do not start with the whole mix.
  2. Fill Table A from the systems themselves. Copy the model, windows, date basis and counting setting from each system's settings, not from memory.
  3. Name the cause of each gap. Use the seven causes above. If you cannot name a cause, write "unexplained" rather than guessing. An unexplained gap is a reason for caution.
  4. Assign trust by decision, not by system. A CRM may be the better reference for pipeline, and a platform may be the better reference for in-flight creative comparisons. Decide this per decision.
  5. Pre-commit an action for each gap band. Your team defines the bands and chooses among hold, cap, test or cut before reading the result. See budget rules while sources disagree.
  6. Review at a matured cohort age. Set the review date for when the cohort has had time to reach the stage you are judging.
  7. Log restatements. When a number changes later, record why.

For related measurement planning, see the weekly acquisition-economics scorecard. For the metric definitions behind CPL, CPQL and CAC, see the acquisition economics playbook.

What the ledger cannot do

  • It does not prove causation. Only a well-designed experiment can estimate lift.
  • It does not tell you which gap size is acceptable. Google describes discrepancies "often up to 20%" as expected because of different attribution models (Google Ads Help, 6270625). That is Google's description of one situation, not a decision threshold, and you should not adopt it as one.
  • It does not fix tracking faults. If a tag is broken, repair it first.
  • It does not replace finance review. Ask your finance partner to agree the trust tiers and actions before the budget meeting.

Frequently asked questions

Which number should I trust?

The one you decided in advance governs that decision. The ledger asks you to assign trust per decision, because a system that suits pipeline reporting may not suit creative comparison.

Is a gap between Google Ads and GA4 normal?

Google documents differences associated with attribution rules, lookback settings, processing and date basis. Whether a particular gap is acceptable depends on what you are comparing; the documentation does not establish your business’s tolerance.

Can I make the numbers match?

Not necessarily: they may count different things. You can align settings where possible, such as comparing on a conversion-time basis, and then explain what remains.

Does attribution show me what my ads caused?

No. Attribution assigns credit under a rule. Randomised experiments are the stronger method for estimating lift.

How long should I wait before judging a channel?

That depends on your sales cycle. Compare channels at the same cohort age and set the review date accordingly.

Next step

Copy the ledger and fill in one channel before your next budget meeting.

Sources checked on 18 October 2026

Account-specific settings: Check Meta’s current settings in your own account; this guide does not assume a universal default.