Every paid search audit I run reaches the same meeting. Somebody pulls up Google Ads showing 60 conversions last month. Somebody else pulls up the CRM showing 41 leads. The room concludes that tracking is broken and asks me to fix it.
Usually nothing is broken. The two systems are counting different events, on different dates, under different rules. Until you can name which of those differences applies to you, you cannot tell a real tracking fault from normal accounting.
Here is how to take the gap apart.
The one that surprises everyone: click-date attribution
This is the single largest cause and the least known.
Google Ads reports a conversion on the date of the click that led to it, not the date the conversion happened.
Someone clicks your ad on 3 March. They convert on 24 March. Google Ads records that conversion against 3 March. Your CRM records the lead on 24 March.
Three consequences follow, and all of them look like bugs:
- Recent periods always look understated. Conversions from last week's clicks have not happened yet. Judging the last seven days is judging incomplete data.
- Historical numbers keep changing. March's total is still growing in April. You export a report, export it again a week later, and the numbers differ. Nothing went wrong.
- Month-boundary comparisons are unreliable. Late-month clicks converting after the month closes get attributed backwards, which quietly moves the line between two reporting periods.
If your sales cycle is longer than a few days, this alone can account for most of your gap. Compare a period that closed at least a full conversion window ago — for the default 30-day click window, that means not judging anything more recent than 30 days back.
Attribution windows
The default click-through conversion window in Google Ads is 30 days, configurable up to 90. Your CRM has no such concept — it stamps the lead when it arrives.
So a click on 1 March converting on 27 March is inside the window and counted. The same journey with a 45-day gap is outside it, and Google Ads never sees it at all. On long cycles this cuts the other way: your CRM shows leads that paid search genuinely produced and the platform never claims.
Check your actual window in Google Ads under conversion action settings. It is one of those settings people assume matches their sales cycle and almost never does.
View-through conversions
Display, YouTube, Demand Gen and Performance Max can count conversions from impressions — someone saw the ad, did not click, and converted later.
Your CRM has no way of knowing the impression happened. There is no click, no gclid, no referrer. The lead simply arrives as direct or organic.
This is defensible measurement, not fraud, but it inflates the platform number relative to anything click-based. Check whether the Conversions column includes view-through, and whether you are looking at Conversions or All conversions — the second includes secondary actions you probably did not intend to report on.
Modelled conversions
Under Consent Mode, when a user declines tracking, Google estimates conversions it could not observe and reports them alongside observed ones. In the EU and UK this can be a meaningful share.
Modelled conversions are, by definition, absent from your CRM. There is no record because no record was created.
The estimate is not arbitrary and it is not verifiable either. Accept it as a permanent component of the gap rather than something to reconcile away. The relevant question is whether it is stable, not whether it is correct.
Duplicate conversion actions
The mundane one, and the most common genuine fault.
A single form submission fires:
- A GTM tag on the thank-you page
- A hardcoded gtag snippet somebody left in the template
- An enhanced conversion event
- A separate "Lead" and "Form submission" action, both marked primary
Now one lead reports as two, three, or four. This is a real defect and it does need fixing.
To find it: open Tools → Conversions and look at every action marked as a primary conversion. Ask what event fires each one. If two actions can fire from a single user action, one of them belongs in the secondary column or should be deleted.
Then use Tag Assistant to load a thank-you page and watch what actually fires. Do not trust the tag list — watch the network.
Counting rules
Each conversion action is set to count every conversion or one per click.
Set to every, a customer submitting a form twice counts twice. Your CRM deduplicates by email and counts one.
For lead generation, one is almost always correct. For ecommerce, every is correct because two purchases are two purchases. Getting this backwards is common in inherited accounts and produces a persistent, confusing inflation.
Time zones
Google Ads reports in the account's time zone, set at account creation and permanently unchangeable. Your CRM uses its own.
A few hours of offset shifts conversions across day boundaries at the edges. On daily reporting this creates noise that looks like volatility. On monthly totals it mostly washes out, except at month boundaries where it does not.
Check the account time zone. In inherited accounts it is frequently a leftover from an agency in a different country.
The identity problem
Even with everything above corrected, you cannot match individual records unless the gclid reaches your CRM.
With auto-tagging on, Google appends ?gclid=... to your landing page URL. If nothing captures it into a hidden form field and stores it against the lead, you can compare totals but never a specific lead to a specific click.
Capturing it unlocks the thing that actually matters: which keywords produce leads that close, rather than which produce leads. That difference reorders most accounts' priorities, and it is the single highest-return piece of marketing infrastructure work available to a paid search programme.
It also enables offline conversion import, so closed revenue flows back to Google Ads and Smart Bidding optimises toward deals instead of form fills.
How to reconcile, in order
- Pick a closed period. At least one full conversion window in the past. Judging last week guarantees a mismatch.
- Confirm the counting rule. One per click for lead gen, every for ecommerce.
- Audit primary conversion actions. Every duplicate you find explains part of the gap.
- Isolate view-through. Compare click-through-only to the headline number.
- Estimate the modelled share. Consent Mode reporting will indicate it.
- Check both time zones.
- Write down the remaining gap and its causes.
If the residual is stable at twenty to forty percent with named causes, you do not have a tracking problem. You have two systems doing their jobs.
Which number goes in the board pack
Report the CRM number. It maps to pipeline and revenue, and it is the one finance will recognise.
Use the platform number for in-flight optimisation. Smart Bidding needs same-day signal, and CRM data arrives too late to steer bids.
Then write one sentence explaining the gap, and keep it in the report:
Google Ads reports 60; CRM records 41. The difference is click-date attribution on a 22-day average cycle, view-through conversions in Performance Max, and modelled conversions under Consent Mode. The gap has been stable within five points for six months.
That sentence ends the recurring argument permanently. It also tells you when something genuinely breaks — because a stable gap that suddenly moves is a real signal, and you cannot see it until you have established what stable looks like.
This is why every SEM engagement I take starts by verifying the data rather than by touching bids. Optimising against numbers nobody trusts is the most expensive habit in paid search, and it is almost universal.
If your two systems disagree and nobody can explain why, send me the account details and I will tell you which of the causes above is yours.