Perfect marketing attribution doesn't exist. Ali Mains explains what each model does, why GA4 and Google Ads disagree, and how to build reports a CFO trusts.
If you've got three reports that each claim the same sale and a finance director asking which one is right, it helps to know that perfect marketing attribution doesn't exist. Not for you, not for the brands with the biggest data teams, and not for us. The goal that is reachable is decisions you can defend rather than a number with two decimal places on it, and in our experience reporting gets better the moment a marketing team says that to its finance director, because it stops being an exam and becomes a shared problem.
On a quarterly call earlier this year an education client who'd run three back-to-back in-person courses in three cities wanted to know one thing: did each event break even? HubSpot said one thing, Google Ads another, the email platform claimed almost everything. Nobody was lying. Nobody was right either.
This piece covers why that happens, what each model does and where it misleads, why GA4 and the ad platforms disagree, and why your numbers conflict with the product team's and what to do about it.
The short version
- Every ad platform and GA4 reports from inside its own walls, so their combined conversions can run to 1.5 to 2 times the deals actually sitting in the CRM, which is why perfect attribution does not exist and the realistic goal is decisions you can defend rather than a precise number.
- The fix is usually tracking rather than a better model, in order: server-side tagging, consent mode set up properly, and CRM source capture written into fields the moment a form fires, with CRM source capture doing most of the work.
- Clear Click builds the tracking, the source-of-truth table and the CFO-grade report for B2B teams whose numbers disagree; what each model does and where it misleads is covered below.
Why does every channel claim the same sale?
Because every platform reports from inside its own walls. Google Ads saw the click, so it takes the credit. Meta saw an impression a fortnight ago and does too. GA4 saw the final session, probably direct or brand search. Add up the conversions each one reports and you'll get something like 1.5 to 2 times the deals in your CRM.
So when someone tells me they're getting conflicting reports from different marketing teams and can't work out what's actually true, my answer is that all of them are true from where each platform sits, and none is the answer. Until you pick one table as the source of truth they'll keep arguing, and to be honest the gap between platform-claimed conversions and CRM deals is the first number worth writing down.
What do attribution models actually do, and where does each one mislead?
An attribution model is a rule for splitting one sale across the touchpoints before it, and anything that splits it across more than one touch gets called multi-touch attribution. In plain English, each is a formula that decides who gets the credit, and every formula is biased towards something.
- Last click. All the credit to the final touch. It flatters brand search and retargeting, starves whatever created the demand, and misleads whenever the sales cycle is longer than a week.
- First click. All the credit to the first touch. Good for what opens the door, useless for what closes it, and it misleads when the first touch was a blog post read eight months ago.
- Linear. Equal credit to every touch. About as informative as splitting a restaurant bill evenly when one person had the steak.
- Time decay. More credit the closer a touch is to the sale. Fine for short B2C cycles. In B2B it quietly turns into last click with extra steps.
- Position based. Usually 40% to the first touch, 40% to the last, 20% across the middle. Better than most, but 40/40/20 is a convention someone picked, not a finding about your customers.
- Data-driven. The platform estimates each touch's contribution by comparing journeys that converted with ones that didn't. It's the default in GA4 and Google Ads and the best of the six, but only sees what that platform can track and you can't audit the split.
Which should a B2B company use? In my head there's a rule of thumb. Sales cycle under 30 days and you sell online: data-driven in GA4 is fine. Sales cycle of 60 to 180 days: no click model will describe it, so measure CRM source at deal level plus pipeline created per channel, and treat GA4 as a directional read on the top of the funnel. Pick one, write down why, and keep it for two quarters.
Why does GA4 show fewer conversions than Google Ads?
It almost always does, and it's rarely a bug. Three things stack.
Attribution windows. Google Ads counts a conversion if the click happened up to 30 days before it by default, and reports it on the day of the click. GA4 reports it on the day it happened with its own lookback. Same conversion, two dates, two monthly totals.
View-through conversions. Google Ads and Meta both credit an ad that was seen but never clicked. GA4 has no idea the impression existed. On an account running Performance Max or YouTube, that alone can be 20 to 40% of reported conversions.
Consent and modelling. When a visitor declines cookies, GA4 with Consent Mode sends a cookieless ping and models the gap. Google Ads models its own gap with different assumptions.
In practice, Google Ads answers "what did our ads influence?" and GA4 answers "what did we observe on the site?". Neither is the number you put in front of a CFO. The full list is in why GA4 and your ad platforms disagree.
What's the tracking problem underneath all of this?
Most attribution arguments I get pulled into aren't about models but about data that never made it into the table. Three fixes cover most of it.
Server-side tagging. Your own server collects events and forwards what's allowed, rather than the browser talking to Google and Meta directly. In plain English, ad blockers and Safari stop deleting your data before it exists. On UK B2B sites we typically see 10 to 25% more conversions recorded after the move.
10 to 25% more conversions: That is the typical increase in recorded conversions Clear Click sees when a UK B2B site moves its tracking server-side.
Source: Clear Click, from UK B2B implementations.
Consent Mode set up properly. A banner on its own does nothing. It has to pass consent signals to the tags, with the right defaults set before the page loads, or you're either on the wrong side of PECR or throwing away modelled conversions you were entitled to.
CRM source capture. Every lead entering HubSpot or Salesforce should carry first source, last source, campaign, landing page and click ID, written into fields the moment the form fires. Rely on the CRM's own guessing and you'll get "Direct" on half your deals. This is the most valuable fix on the list and it's a two-week job. Those three, in that order, are most of our tracking and measurement work.
Why do our attribution numbers conflict with the product team's, and how do we fix it?
This is the question I get asked most, usually phrased as "our attribution models keep conflicting with what the product team reports for user engagement". The answer is that the two teams are measuring different things, and both are calling it performance.
Marketing attribution assigns credit for an acquisition to channels, inside an attribution window, using a browser or click ID. Product analytics counts what a logged-in user does after sign-up, keyed on a user ID. One is about where a person came from, the other about what they did after arriving. They were never going to match. Three steps fix it.
First, agree the question before the model. "Which channel brings in users still active at 90 days?" is one marketing and product can both answer from the same table. "Which channel has the best ROAS?" is one product can't touch. Write it down and get both heads to sign it.
Second, build a single source-of-truth table. One row per lead or account: marketing source on the left, CRM stage and value in the middle, product engagement on the right, joined on a hashed email or account ID rather than a cookie. Every report, marketing's and product's, reads from this table.
Third, and this is Nic's line that I've stolen for every reporting project since: build the logic before the dashboard. If the definitions live inside a Looker Studio chart, every model change means a rebuild. If the raw touchpoints are stored once and the model is the final calculation on top, switching from last click to position based is a one-line change. That's the answer to "every time marketing changes their attribution model our entire reporting infrastructure needs rebuilding". Store the events, apply the verdict last.
Back to the education client. Their CRM created a new deal every time someone downloaded a free course, so one person could have six deals, and nearly 2,000 sat at "consideration". Marketing counted six conversions, product counted one user, both were technically correct. We moved the original deal through the stages instead, used lifecycle stages to separate returning customers from new, and set a rule for when a free lead is dead. The model never changed. The data did.
When should you stop refining attribution and run an incrementality test instead?
When the decision in front of you is worth more than about £20k of annual spend and attribution can't tell you whether the channel is causing sales or just standing near them.
Attribution, however good, is correlation. Brand search looks brilliant on last click because people already going to buy searched your name on the way in, which is why I keep saying over-reliance on Google Ads will eventually come back to haunt some companies. The only way to know if spend is incremental is to switch it off and watch. The practical version is a geo holdout: pause the channel in matched regions for four to six weeks and compare against the rest. The method is in our incrementality testing guide for CFOs. Beyond five or six channels the next step is a marketing mix model, covered in marketing mix modelling vs attribution.
One limit worth knowing about: attribution can't see seasonality or the economy either. When a prospect on a discovery call asked what the analytics would tell us about a dip in enquiries, I said there are a few things analytics won't tell us, and those are two of them. So we run a seasonality analysis paired with the macro factors we can get data on, and it sits next to attribution in the monthly report, which means a quiet month can be read for what it is rather than pinned on a channel.
What should a CFO-grade monthly report show, and who should build it?
One page. Spend by channel, leads by channel from the CRM (not the platforms), pipeline created, revenue closed, cost per lead and cost per opportunity, against last month and the same month last year. If it needs a legend explaining data-driven attribution, it isn't a CFO report.
Once that report is built, the same source-of-truth table is what makes CRM data trustworthy enough for sales and finance to run off, not just marketing. That handoff is what our CRM and revenue operations team builds.
The roadshow client got exactly this. Three events, one campaign. We segmented by location, set tiered goals for each (break even, target, stretch) and pulled it into one dashboard fed by the CRM. The MD could see two events had covered their costs by week three and the third needed a different venue. That's a decision, not a precision exercise, which is the point of our reporting and dashboards work.
Do you need marketing attribution software? Under about £15k a month across two or three channels, no. The £1,000 to £3,000 a month a tool costs is better spent on tracking and a CRM that captures source. Above that, especially with phone or offline sales, tools that stitch CRM revenue back to the click earn their keep. Buy the tool after the logic is built, never instead of it.
And do you need a marketing analytics agency or an in-house analyst? An analyst is the right answer if you have one with time. In reality most B2B marketing teams we meet have someone who is 20% analyst and 80% something else, and doing well at both. A marketing analytics agency should deliver three things: the tracking fixed and documented, the source-of-truth table built and owned by you, and a monthly report your finance director reads without questions. Our data and analytics work is built around those three deliverables, with PPC attribution intelligence on top for the paid budget decisions.
Questions we get asked
Why do our attribution reports conflict with the product team's numbers?
Marketing attribution credits channels for acquisitions inside a lookback window using click or cookie identity; product analytics counts logged-in user behaviour after sign-up. Agree one question both teams can answer, then report it from one shared table joined on account or email ID.
Why does GA4 show fewer conversions than Google Ads?
Google Ads counts view-through conversions, uses a 30-day click window by default, reports on click date and models consent gaps its own way. GA4 sees none of the impressions, so a 20 to 40% gap is normal on accounts running Performance Max or YouTube.
Which attribution model should a B2B company use?
Data-driven in GA4 for a directional read, and CRM first and last source at deal level for anything that goes to the board. Over a 60-day sales cycle, pipeline created per channel is the number that matters.
Do we need attribution software?
Not under roughly £15k a month of spend across a handful of channels. Above that, or with phone and offline sales in the mix, a tool that ties CRM revenue back to the click is worth the £1,000 to £3,000 a month.
What should a marketing analytics agency actually deliver?
Working tracking with documentation, a source-of-truth table you own and can query, and a monthly report your CFO can read in five minutes. Dashboards are the last 10% of that work, so if the proposal leads with them, ask what logic sits underneath.
How do we stop rebuilding our reporting every time the model changes?
Store raw touchpoints and CRM events once, in a warehouse or even a well-kept sheet, and apply the attribution model as the final calculation layer rather than baking it into each report. A model change is then one rule updated, and every dashboard picks it up.
Where we fit
We're for B2B marketing teams with three reports that disagree and a finance director who has stopped believing any of them. We fix the tracking, build the table and put a one-page report in front of the board. For Hubject the budget modelling alone saved £40k, and for AM Phillip Trucktech the same discipline delivered 32% lower cost per lead.
For a second pair of eyes, get in touch and bring the two reports that disagree most. If attribution has started to feel like the wrong question, read our piece on market share as a marketing metric.
