Google retired four of its six attribution models in mid-October 2023. Plenty of the guides explaining all six have not caught up. If you are picking between marketing attribution models for B2B this quarter, Google Ads and Google Analytics 4 will run two of them, and the more useful question is what neither one can see.
This is a working guide to the models, what each one hides, and why last-click keeps winning arguments it should lose.
What a Marketing Attribution Model Actually Decides
An attribution model is a rule for splitting credit between the touches you already recorded. That is the whole job. It does not find touches, and it cannot add the ones you missed.
So the argument teams have about models usually sits downstream of a coverage problem. Picture a deal where six people read about you across four months, and your tags recorded three of those moments. Every model in the world is now splitting credit across three moments, confidently, to two decimal places.
| Model | How it splits credit | What it hides |
|---|---|---|
| Last-click | All of it to the final touch. | Everything that created the demand. Brand search and direct look like origins rather than the destinations they are. |
| First-click | All of it to the first recorded touch. | Everything that closed the deal, and the fact that “first” only means first inside your window. |
| Linear | Evenly across every touch. | That the touches were not equal. A newsletter open and a pricing page visit are not the same event. |
| Time-decay | More credit the closer a touch sits to the conversion. | Long consideration. An early touch earns less for being early, whatever actually happened in it. |
| Position-based | The model’s set split: 40% to the first touch, 40% to the last, 20% across the middle. | Nothing in particular, and that is the problem. The split is a convention, not a measurement. |
| Data-driven | Weights learned from converting and non-converting paths in your own account. | Its own reasoning. Google keeps the weights to itself, so whoever owns the budget takes the split on trust. |
Google Retired Four of the Six Models in 2023
In April 2023 Google announced it was retiring first-click, linear, time-decay and position-based from Google Ads and Google Analytics 4. The sunset landed in mid-October 2023. Google gave adoption as the reason: fewer than 3% of conversions in Google Ads used those four. Accounts sitting on a retired model moved to data-driven automatically.
So inside Google’s own tools the menu is two items long. Last-click, or data-driven.
Google also says that switching to data-driven typically produces a 6% increase in conversions for advertisers. Read that slowly, because it is Google describing its own product. Switching model does not win you a deal you were going to lose; it changes how the deals you already closed get credited. So treat a jump in reported conversions as a question to open rather than a result to bank. If your board watches the number move 6% in a quarter when nothing changed in the market, someone will ask why.
Why Last-Click Keeps Winning Even Though It Is Wrong
Everyone in B2B knows last-click undercounts the top of the funnel. It keeps winning anyway, and not because marketers are careless.
Last-click is reproducible. Two people pull the same report on the same day and get the same number. Nobody defends a methodology in a budget meeting, because there is no methodology to defend, only a rule a child could restate. When the CFO asks how a channel earned its credit, “it was the last thing they clicked” ends the question.
Data-driven cannot do that. Its weights come out of a model Google does not open up, tuned on paths inside your own account. Data-driven does use more of the recorded path than last-click does, and that is the honest case for it. The person whose budget it reallocates still cannot check the arithmetic, and in our experience that is a governance problem before it is a statistics problem. Governance problems decide which report gets used on a Monday.
There is a second reason, less flattering. Last-click reliably flatters paid search and brand terms, and those are usually the channels run by whoever pulls the report.
The Ninety Day Ceiling Every Model Shares
Now the constraint that outranks the model choice.
Google Analytics 4 only looks back so far. For most key events the default window is 90 days, and you can shorten it to 60 or 30. Acquisition events default to 30 days, with 7 as the alternative. Engaged view events sit fixed at 3 days. Ninety days is the longest look the tool takes, under any model.
Data table
| Key event type | Days |
|---|---|
| Engaged view, fixed | 3 |
| Acquisition, shortest | 7 |
| Acquisition, default | 30 |
| Other events, shorter | 60 |
| Other events, maximum | 90 |
A touch outside that window is not underweighted. It is absent. Last-click, data-driven, all four of the retired models: every one of them divides a pie baked inside those 90 days.
Set that against how B2B software gets bought. Gartner reports that a typical buying committee for a complex solution runs to six to ten decision makers, each working from four or five pieces of information they gathered themselves. Gartner also reports that buyers spend roughly 17% of their total purchase time meeting suppliers at all, and 5% to 6% with any single vendor once several are competing. Its March 2026 survey found 67% of B2B buyers would rather buy without talking to a rep, up from 61% in its June 2025 survey.
Do not take 90 days as a verdict on your business. Look up the median days from first recorded touch to closed won in your own CRM, and hold it against the window you have set. If the median comes in shorter, the window spans your typical deal. If it comes in longer, then for at least half your deals the window shuts before the deal does, and you know that from your own data rather than from an article.
The rest happens where no tag reaches. A colleague forwards a PDF. Someone asks a question in a private Slack group and gets three vendor names back. A director reads a review thread at 11pm and quietly drops you from the shortlist. Six weeks later that lands in your CRM as one direct visit to a pricing page, credited to nothing.
What to Decide With When the Model Cannot Decide
Attribution is good at direction and diagnosis. Which pages appear on converting paths. Which campaigns stopped appearing at all. Where the drop sits between the demo request and the accepted lead. Use it for that daily.
It answers the causal question badly, and the causal question is what the budget meeting is about: if we stop spending this, what do we lose? Every model works from paths that already happened, in a world where the spend was running. Three things answer it better, and none needs a new platform.
Pause something and watch. Turn a channel off in one region or one segment for a defined period, hold everything else steady, and compare. Blunt, and it costs real pipeline. Its answer also holds still when someone changes an attribution setting, which is the whole point.
Split by geography. Run the campaign in half your markets and hold the other half back. The held back markets show what happens without it, which is the thing attribution never observes.
Agree on cost per accepted lead, with sales in the room. Not cost per MQL. Accepted means a salesperson looked at it and said yes, this is a real opportunity. Swap your attribution model tomorrow and that number still means the same thing, because a human wrote it into the CRM.
One piece of housekeeping. If your measurement plan still opens by preparing for the end of third-party cookies, rewrite it. Google confirmed on 22 April 2025 that it will not deprecate third-party cookies in Chrome, after moving the date from 2022 to 2024 to 2025 and then to a user choice prompt it never shipped. It has since retired most of the Privacy Sandbox APIs. The cookie was never the main constraint on B2B attribution. Long cycles, large committees and private rooms were, and all three are still here.
Pick last-click if you need a number everyone can reproduce. Pick data-driven if you can live with weights you cannot check. Then stop asking either one to settle a question it was never able to settle, and go run a test.
FAQ
What are the marketing attribution models?
Six are commonly described: last-click, first-click, linear, time-decay, position-based and data-driven. Since mid-October 2023, Google Ads and Google Analytics 4 run only two of them, last-click and data-driven. The other four are worth understanding anyway, because the reporting you inherit was often built around one of them.
Which attribution model is best for B2B?
None of them wins outright, and your sales cycle length matters more than the choice. If your median deal closes inside 90 days, data-driven uses more of the recorded path than last-click does. If your median runs longer, at least half your deals outlive the window, so pair whichever model you pick with a holdout test.
Why did Google remove four attribution models?
Google announced the retirement in April 2023 and completed it in mid-October 2023, citing adoption: fewer than 3% of conversions in Google Ads used first-click, linear, time-decay or position-based. Accounts on a retired model switched to data-driven automatically.
What is the difference between last-click and data-driven attribution?
Last-click gives all credit to the final touch, using a fixed rule anyone can restate. Data-driven learns weights from converting and non-converting paths inside your own account. Google does not publish those weights, so you cannot check how it reached its answer.
How far back does Google Analytics 4 look?
For most key events the default lookback window is 90 days, adjustable down to 60 or 30. Acquisition events default to 30 days with a 7 day option, and engaged view events stay fixed at 3 days. Ninety days is the maximum, so an older touch receives no credit under any model.
Is multi-touch attribution worth it for B2B?
Yes, when your cycle fits inside the lookback window and most touches happen on channels you can tag. Much less so when buying committees do their reading on review sites, peer forums and forwarded documents, because no multi-touch model can weight a touch it never recorded.
Do we still need to plan for third-party cookies disappearing?
No. Google confirmed on 22 April 2025 that it will not deprecate third-party cookies in Chrome, and it has since retired most of the Privacy Sandbox APIs. Plans written around the phase out need updating. Long cycles, large committees and untagged research are the real limits on B2B attribution.
What should we report to the board instead of attributed revenue?
Cost per accepted lead, agreed with sales, plus the result of whatever test you last ran. Accepted means a salesperson confirmed a real opportunity. Change your attribution model and that number keeps its meaning, which attributed revenue does not.
COSEOM Team
COSEOM is a multilingual B2B growth agency. Since 2008 the COSEOM team has run international SEO, generative engine optimization, paid media and demand generation programs for B2B software and technology companies, in English, German, Spanish and Portuguese. This account publishes our research, playbooks and product guides. More about COSEOM