Why GA4, Meta and Google Never Agree With Your CRM

Below: the seven reasons, a table of which number to trust for which decision, a reconciliation a team can actually run, and the governance point that stops this meeting repeating.

Follow one person. On Tuesday she sees your Meta ad on a phone, does not click, and that evening searches your brand on a laptop and clicks the Google ad. She reads two pages, leaves, returns nine days later from a newsletter and fills the form. Sales qualifies her three weeks after that, and the deal closes in November.

Meta may count a view-through conversion, dated Tuesday. Google Ads counts one also dated Tuesday, because that is when the click happened. GA4 counts one nine days later and credits the newsletter. The CRM records a lead nine days later, a qualified lead three weeks after that, and revenue in November. Every one is correct inside its own rules.

That single journey contains six of the seven reasons. The seventh, what the word lead means, causes the loudest arguments and gets fixed in a meeting rather than in a tool.

Reason 1: different attribution models and lookback windows

Each system decides for itself which touch earns credit and how far back it will look. None of it is standardised, and every default is tuned to the vendor's own product.

• Meta defaults to a 7-day click plus 1-day view setting, configurable per ad set, and only sees touches on Meta's own surfaces.
• Google Ads uses data-driven attribution for most conversion actions by default, with a conversion window set per action, commonly 30 days and adjustable.
• GA4 runs its own model over its own data, with its own lookback windows, and its acquisition reports use last non-direct click, which is a third rule again.
• The CRM has an implicit model nobody chose: whatever the lead source field held when the record was created, or what a salesperson typed.

Two consequences. A journey longer than a platform's window is invisible to it, which is why B2B accounts with a three-month cycle see the largest gaps. And because each platform sees only its own touches, a journey crossing Meta and Google is counted in full by both: not double counting inside either dashboard, but double counting the moment you add them up. See attribution window.

Reason 2: view-through and engaged-view conversions

Meta counts a conversion when somebody saw the ad and did not click, inside its view window. Google Ads has its own versions for Display, Video and Demand Gen in separate columns, including engaged-view conversions on video. GA4 counts none of it, because no click reached your site, and the CRM has no idea an impression happened.

On an account running upper funnel social this one reason can be most of the gap, and it is not fraud. It is a different definition of contribution. The rule: know which columns include view-through before you quote them, and never compare one that does against one that cannot.

Reason 3: the click date against the conversion date

The reason most teams have never heard of, and it explains gaps that look like data loss. Ad platforms report a conversion back on the day of the click or impression that earned it. GA4 and your CRM report it on the day it happened. The same conversion lands in two different weeks.

For e-commerce with a same-day cycle that is noise. For lead generation with a two-month sales cycle it smears the curves apart, and a recent week in a platform dashboard is always incomplete: conversions are still being backdated into it. This week's Meta number against this week's CRM number compares a figure that will keep growing to one that will not.

Reason 4: deduplication and identity

Each system identifies a different unit. GA4 works with a browser identifier and a session, the ad platforms with a person inside their own logged-in graph, the CRM with an email address. A lead who opens the ad on a phone, researches on a work laptop and signs on a third device is one person to the CRM and three to the browser.

Deduplication differs again. Browser and server events must share an event id or they count twice, a duplicate submission may be one conversion and two CRM records, and a returning customer is a new conversion to the pixel and an existing contact to sales. See cross-device journeys for how much of this a stored click ID solves.

Reason 5: the data the browser never sends

A browser tag must load, run and reach a third-party endpoint before it counts anything, and four things stop it. Ad blockers block the request. Safari and iOS restrict cookie lifetimes, so a return visit weeks later has forgotten the first one. A declined consent banner means the tag never fires. And ordinary breakage, a tag missing from the thank you page or a redirect that drops parameters, is commoner than anyone expects.

This hits browser-measured numbers hardest, which is why GA4 usually sits below the platforms and often below the CRM. Server-side tracking on your own domain removes most of it, because the event leaves your server rather than the visitor's browser. The comparison is in server-side versus browser tracking.

Reason 6: modelling and thresholds

What a platform cannot observe, it estimates. Google Ads models conversions it believes happened but could not measure, including consent-declined and cross-device paths, and reports the estimate in the same column as the observed ones. GA4 applies its own modelling and withholds detail where the numbers are small enough to identify individuals, so a segmented report can total less than the unsegmented one.

None of that is dishonest, and a modelled figure is often closer to the truth than a raw count. But a model estimates a population, and a CRM record is a named person you can phone. Do not expect the two to reconcile row by row.

0:45

The most human reason, and the one that produces the loudest meetings. The pixel was asked to count form submissions. Sales counts people worth calling. Finance counts signed contracts. All three call their number leads, and nobody has written down which one the reporting means.

The drift compounds. A submission that failed email verification is a conversion to Meta and nothing to the CRM. A prospect who filled two forms is two conversions and one contact. A lead typed in after a trade show is a CRM record with no click behind it. Before blaming a tool, take one week and count the CRM records with no ad click, and the platform conversions that produced no CRM record. That is usually the whole argument.

The fix is a written definition of qualified lead that sales agrees to, and one CRM stage that represents it. Everything downstream depends on that sentence existing.

There is no single true number, and hunting for one is the mistake. There is a right instrument per question. Settle it once with this table.

Note what the table does not contain: a column reconciling the four totals. Nobody needs that number. Every real decision maps to one of these rows.

Once, properly, on one week and one campaign. The point is not to make the numbers equal but to name and size every reason they differ, so the next meeting starts from a shared list rather than a shared suspicion.

1. Pick one campaign and one week that finished at least 30 days ago, so no window is still filling.
2. Export the platform conversions for that week with both the click date and the conversion date, the click ID and the campaign.
3. Export every CRM record created that week, plus every record whose first touch names that campaign, whatever week it was created in.
4. Join on the click ID first, then hashed email, then phone number. Record which key matched: that alone tells you how good your identity chain is.
5. Classify every unmatched row into exactly one of the seven reasons, one row per record, in a spreadsheet with a reason column.
6. Count the classes and sort them. The largest is your next project, and the rest become footnotes in the monthly report.

The classification step is the whole value. A gap of 45 conversions is an argument. A gap made of 18 view-through, 12 outside the window, 9 with no click ID and 6 duplicates is a project plan, and only the click ID rows need fixing.

A gap is not a symptom. A gap that behaves unpredictably is. Three things separate a setup working as designed from one that needs work.

• Healthy: stable and explainableThe platforms sit above the CRM by roughly the same proportion each month, the direction never flips, and most unmatched rows classify as window and timestamp effects.
• Broken: a step changeThe gap jumps on a particular date rather than drifting. That is a deployment, a consent banner change or a tag that stopped firing, and it has a cause you can find in a changelog.
• Broken: no click ID on the recordA large share of CRM records with no click ID means the identity chain never survived the journey. No reporting fixes that, because the join key was never stored.

The last is the common diagnosis in lead generation, and it is worth stating plainly: if the click ID is not captured at the first visit and carried onto the CRM record, no model or dashboard recovers it later. See pixel and CRM mismatch.

The technical half is a week of work. The governance half is an hour, and it is what stops the meeting repeating. Three numbers, defined in a document everyone can open, reviewed quarterly and never renegotiated in the meeting itself.

• The optimisation number

  What each ad platform counts, in its own dashboard, on its own model. For daily decisions inside that platform. Never compared across platforms and never added up.

• The efficiency number

  Cost per qualified lead: platform spend over CRM qualified leads, on one written definition of qualified that sales signed off. What the weekly meeting runs on.

• The revenue number

  Closed revenue from the CRM by close date, first and last touch shown together. Slow, always late, and the only one anyone outside marketing believes.

Write next to each one who owns it, which system it comes from and how often it refreshes. Then when somebody asks why Meta says 84 and the CRM says 39, the answer is that those are the optimisation number and the efficiency number, both correct, and the meeting moves on.

LeadJourney does not try to make the four numbers equal, and no honest tool would. It builds the record underneath them. Tracking runs server-side on your own domain at 95%+ accuracy, so most of reason five disappears, and the click IDs are captured at the first visit: gclid with gbraid and wbraid, fbclid, li_fat_id, msclkid, plus the UTMs and the landing page. That anonymous first click is joined to the person at the form fill, the call or the booking, and the journey follows your CRM stages to the closed deal.

Because the join key is stored rather than inferred, the reconciliation above stops being an annual exercise. Cost per qualified lead and cost per closed deal are standard reports, models switch between first click, last click, linear, time decay and position based without re-tracking, and no attribution window cuts a long cycle short. The CRM stages you choose go back to Meta, Google Ads, LinkedIn and Microsoft as conversions with the deal amount and currency attached, server to server, matched on the click ID and deduplicated on the Meta side, so the platforms optimise on qualified pipeline instead of form fills.

What it will not do is make Meta agree with Google Ads. Both keep their own models, windows and view-through rules. What changes is that each total traces back to the same named records, so a gap is explainable in an afternoon instead of debated for a quarter. Native CRM integrations cover HubSpot, Salesforce, Pipedrive, Close, Attio, GoHighLevel, ActiveCampaign and Odoo, anything else connects by webhook, and setup takes about 21 minutes.

Read verified reviews on Trustpilot, G2 and leadjourney.io/testimonials.

Related reading: the true cost per lead for the efficiency number, multi-touch attribution for B2B lead generation for the models, server-side versus browser tracking for reason five, pixel and CRM mismatch and GA4 revenue attribution for the two commonest versions of this problem, and LeadJourney vs Google Analytics 4.