Multi touch attribution tools and the four models Google stopped supporting

Before you buy a multi touch attribution tool, check whether the models you want still exist in the platforms the data comes from. Google Ads has withdrawn four of them. Google Analytics offers three models in total, two of which are last click. Any product demonstration that shows you a linear or a time decay curve has either built that curve itself from path data or is showing you something the platform no longer supports, and those are very different purchases.

What the two documentation pages actually say

Google Ads, on its attribution models page, read on 22 October 2026: "Note: The first click, linear, time decay, and position-based attribution models are no longer supported by Google. Conversion actions that used the deprecated attribution models have been upgraded to use data-driven attribution. You can also switch to the last click model, which is still supported."

Google Analytics, on its attribution and attribution modelling page, read on 22 October 2026: "There are 3 attribution models available in the Attribution reports in Google Analytics properties: Data-driven attribution, Paid and organic last click, Google paid channels last click." The same page adds: "All attribution models exclude direct visits from receiving attribution credit, unless the path to key event consists entirely of direct visits."

Those two notes together redraw the market. The four withdrawn models were the ones that made multi touch legible to people who are not analysts, because a linear split is easy to explain in a meeting. What remains in the platforms is one algorithmic model whose weights you cannot inspect, and last click.

So what is left for a tool to do

The third row is where an attribution tool earns its place in a business with a long cycle or an offline close. Joining a signed contract back to the touches that preceded it is work the ad platforms cannot do for you, and it does not depend on which model is in fashion.

The last row is the one to press hardest. Attribution divides credit for conversions that happened. It cannot tell you what would have happened without the spend. Only a holdout can answer that, and a vendor who will not discuss holdouts is selling a reporting layer, which is fine as long as everybody knows it.

The two numbers that will not reconcile, and why that is normal

Expect your tool and Google Analytics to disagree. One documented reason is the direct visit rule quoted above: Analytics excludes direct from credit unless the whole path is direct, and most independent tools do not. A second is the conversion window and the attribution setting on each conversion action, which change every number downstream. A third is simple timing, because platforms credit a conversion to the date of the click and analytics tools often credit the date of the conversion.

The useful discipline is to pick one source of truth per decision, name it on the report, and print the count the rate was taken over beside it. The denominators behind each common metric are set out in our metric dictionary, and the paid search reports and settings that change the figures are in paid search analytics.

A limit on how far back you can look

Multi touch work tempts people into long lookbacks, and retention gets in the way. Google's data retention documentation, read on 22 October 2026, describes controls that set "the amount of time before user-level and event-level data stored by Google Analytics is automatically deleted from Analytics servers". Data already deleted does not return when the setting is changed. Check the current setting before promising anybody a two year path analysis.

One figure worth stating as absent: there is no published, independent accuracy benchmark comparing the major multi touch attribution products on the same dataset. We looked for one to cite and there is none. In its absence, the only honest comparison is the one you run yourself on your own data, with the method written down.

A disclosure, because we sell a measurement of the thing this page discusses. AI Knows Us reads five AI surfaces and reports two separate findings, whether a business is named and whether its own pages are cited, and they are never added together. Our own reading on 22 October 2026 was named 0 and cited 0 across 635 readings, published in full as our own null result after 507 new pages. Where a rival is stronger, plainly: Profound publishes SSO and SAML, SOC 2 and up to nine engines; our rendered pricing page carries none of those words and names five surfaces, not nine. The cheapest plan of ours that reads all five is Growth at 299 dollars a month in the US and the UK, and the 149 dollar SEO Start plan reads 0 questions, 0 credits and 0 engines, so it is not an AI visibility plan.

Questions buyers ask about multi touch attribution

Is data driven attribution better than last click

It uses more of the path, and you cannot inspect its weights. Last click is crude and completely transparent. Pick by whether the decision needs explaining to somebody who will not take an algorithm on trust.

Can I still use a linear model

Not in Google Ads, whose documentation says first click, linear, time decay and position based are no longer supported, and not in Google Analytics, which offers three models. A third party tool can compute one from path data, but it will then disagree with the platforms and you should expect that.

Why does my tool report more conversions than Google Ads

Different windows, different models, different treatment of direct visits, and different dates credited. The gap is structural rather than a bug. Name one source per decision.

Do I need attribution or incrementality

Attribution divides credit for what happened. Incrementality asks what would have happened anyway, and needs a held out group. Budget questions need the second one.

When can we judge whether attribution work has paid off

Judge it on decisions rather than on elapsed time: how many budget moves were made because of it, whether the two sources now disagree by a known and explained amount, and whether every rate on the report carries its denominator.

The touch that leaves no path at all

When an assistant answers a buying question, there is often no click to attribute and sometimes no link at all. That touch cannot enter any attribution model, so it has to be measured directly, on a recorded prompt set, with citation and naming counted separately against the number of readings.