No reporting meeting is complete without the question: "so how many conversions did we actually get?". Google Ads says 412, GA4 counts 368, the CRM records 344. All three numbers are correct. They measure different things.
The data-driven model, and why it is no longer a choice
Data-driven attribution distributes conversion credit using the historical data of that specific conversion action: instead of applying a fixed rule, it estimates the actual contribution of each interaction along the path. It is now the default model for most conversion actions.
The other rule-based models — first click, linear, time decay and position-based — are no longer supported: actions that used them were upgraded to data-driven. That leaves two real options:
| Model | What it does | When it makes sense |
|---|---|---|
| Data-driven | Assigns credit by each interaction's estimated contribution | The default: multi-touch paths, any non-trivial funnel |
| Last click | All credit to the final click | Only where historical continuity matters, or in very simple single-channel accounts |
The model does not create conversions. Switching models does not sell more: it moves credit between campaigns. But because Smart Bidding optimises against that credit, a model change changes where the budget ends up. Treat it as a structural decision, not a reporting preference.
Why Google Ads and GA4 will never agree
The differences are not errors — they are different design choices. Four of them explain almost every gap:
- The date. Google Ads credits the conversion to the day of the interaction that led to it. GA4 records it on the day it happened. With a two-week purchase cycle, at month end the two systems are looking at two different sets of events.
- The scope. Google Ads measures Google Ads. GA4 measures every channel and hands credit to organic, direct, email and referral too: the same conversion can read as paid search in Ads and organic in GA4.
- Which interactions count. Google Ads includes interaction types GA4 does not see the same way, starting with view-through conversions and offline imports.
- Consent and modelling. Under Consent Mode part of the data is modelled, and the two systems model at different points in the chain. We covered this in Consent Mode v2 and tracking.
The working rule
Use Google Ads to optimise campaigns: it is the system Smart Bidding learns from and the only one consistent with its decisions. Use GA4 to decide the channel mix and to understand the full path. A report that puts the two columns side by side and subtracts one from the other produces arguments; a report that states which question it is answering produces decisions.
Conversion windows: the lever almost nobody touches
The conversion window sets how long after an interaction a conversion can still be credited to it. It is the setting with the most direct effect on budget decisions, and it is nearly always left at its default.
- Long purchase cycle. If your average customer compares for three weeks and the window is 30 days, you are at the edge: some conversions fall outside it and the campaigns doing early work look unproductive.
- Impulse purchase. A long window on a fast-converting ecommerce inflates credit for interactions that contributed little.
- Consistency across actions. Different windows on different actions make campaign comparisons unreliable, because campaigns are not collecting credit over the same span of time.
How to choose properly: look up the median time from first touch to close in your CRM, and set a window covering at least 75% of it.
The lead generation case
In lead generation attribution has an extra problem: the moment that matters — the sale — happens off-site, weeks later. Any model applied to the submitted form is crediting an event that is not yet a result.
The answer is not a different model: it is sending the outcome back. Upload offline conversions with their real value and data-driven attribution starts distributing credit based on what produced customers rather than forms. That is the loop we described in enhanced conversions for leads and the Data Manager API: without it, any debate about models stays academic.
What happens to upper-funnel campaigns
Data-driven attribution recognises contributions last click erased. That is why video campaigns, Demand Gen and in-market audiences look like entirely different campaigns under one model versus the other — and why judging them on last click leads to switching them off.
The reverse is also true, and it is the uncomfortable part: brand campaigns tend to lose credit, because data-driven recognises that the final click often closes a journey that began elsewhere. If your brand campaign only holds up the account's numbers under last click, the model is not making things worse — it has stopped hiding an acquisition problem.
The verification checklist
- Check the model on every conversion action, not just the primary one: automatic migrations left mixed configurations in a lot of accounts.
- Check which actions are included in "Conversions": that set is what Smart Bidding actually optimises toward.
- Align conversion windows with the real purchase cycle, measured in the CRM.
- Pick one source of truth per question and say so in the report, instead of comparing incomparable columns.
- Close the loop with offline conversions if you sell leads: it is the prerequisite for any sensible attribution.
If your reports tell three different stories and nobody knows which to follow, the free audit starts exactly there: models, windows and consistency across Google Ads, GA4 and your CRM. You can also see how we work and which services cover this ground.