Why did multi-touch attribution stop working?

Because the touches stopped producing data. Gartner found 67% of B2B buyers prefer a rep-free experience, Pew found AI summaries cut clicks from 15% to 8%, and consent refusal removes the rest. The research happens without a trackable session.

Is self-reported attribution reliable?

Directionally, yes. It suffers from recall bias and skews toward the last thing remembered, but after a couple of hundred free-text responses the pattern is clear, and it captures channels no analytics platform can see.

What is a holdout test?

Pausing a channel in matched geographies or audiences for four to six weeks and measuring the pipeline difference against the control group. It is the only widely available method that proves causation rather than correlation.

Is AI-referred traffic worth tracking if the volume is small?

Yes. Semrush's analysis of over 500 high-value topics found AI search visitors converted at roughly 4.4 times the value of traditional organic visitors, because they arrive having already compared their options.

How do you measure marketing when the journey is invisible? - Tack Media | AI Digital Marketing Agency in Los Angeles

Q: Does GA4 fill in the missing data?

Partly, and only as an estimate. Google's behavioral modeling requires thresholds including 1,000 events per day with analytics storage denied over seven days. Below those thresholds you see undercounted raw data with no warning.

Short answerStop trying to trace every touch and measure three things instead: total pipeline against total spend, self-reported attribution collected at the point of conversion, and controlled holdout tests. Multi-touch attribution broke because most research now happens in AI answers and private channels that leave no referral data.

The short answer

The short answer Stop trying to trace every touch and measure three things instead: total pipeline against total spend, self-reported attribution collected at the point of conversion, and controlled holdout tests. Multi-touch attribution broke because most research now happens in AI answers and private channels that leave no referral data. Why the journey went dark This is not.

The short answer

Stop trying to trace every touch and measure three things instead: total pipeline against total spend, self-reported attribution collected at the point of conversion, and controlled holdout tests. Multi-touch attribution broke because most research now happens in AI answers and private channels that leave no referral data.

Why the journey went dark

This is not a tracking configuration problem you can solve with a better tag manager. Four separate forces removed the data, and three of them are permanent.

Buyers deliberately avoid identifying themselves

Gartner surveyed 646 B2B buyers between August and September 2025 and found that 67% prefer a rep-free buying experience and 70% prefer a completely digital, self-service journey. A buyer who does not want to talk to you will not fill in a form to get information they can obtain anonymously.

Research moved into answers, not clicks

Pew Research Center tracked 900 US adults in March 2025. When Google showed an AI summary, users clicked a traditional search result 8% of the time versus 15% without one, and only 1% of visits produced a click inside the summary itself. That research still happened. It simply produced no session, no referrer and no campaign parameter.

AI is now a research layer of its own

In the same Gartner work, 45% of buyers reported using generative AI during a recent purchase, and buyers used an average of seven information sources. Notably, 69% of buyers turn to sales reps to validate AI-generated insights. So the assistant shapes the shortlist, then a human conversation confirms it. Your analytics sees only the conversation.

Consent and modelling removed the rest

Where cookies are declined, GA4 substitutes estimates. Google states that behavioral modeling “uses machine learning to model the behavior of users who decline analytics cookies,” and it only turns on once a property clears thresholds including 1,000 events per day with analytics storage denied across seven days. Under those thresholds you get raw undercounted numbers with no indication that anything is missing.

The four measurements that still work

Method	What it answers	Cost and effort	Where it fails
Blended CAC and pipeline coverage	Is total marketing spend producing total pipeline	Low. Two numbers a month	Cannot allocate between channels
Self-reported attribution (“how did you hear about us”)	What the buyer believes influenced them	Low. One required form field	Recall bias, last-thing-remembered skew
Geo or audience holdout tests	True incremental lift of a channel	High. Needs volume and patience	Needs enough conversions to reach significance
Branded search and direct traffic trend	Whether demand creation is working	Low. Search Console plus analytics	Lags spend by weeks or months

Used together these cover the ground that multi-touch attribution used to claim. Blended numbers keep you honest at the top. Self-reported attribution tells you what the buyer noticed. Holdouts prove causation. Branded search shows whether awareness is compounding.

How to tell which applies to you

Match the method to your deal shape.

Short cycle, high volume, low ticket. Platform-reported conversions plus a blended CAC check are usually enough. You have the volume to run holdout tests, so run them quarterly and trust them over platform attribution when the two disagree.

Long cycle, low volume, high ticket. Attribution modelling is statistically meaningless at 30 deals a year. Use self-reported attribution on every inbound, tag opportunities in the CRM with the source the buyer names, and measure pipeline created per quarter against spend from the previous quarter.

Anything with a research-heavy purchase. Track share of AI answers alongside rankings. Semrush’s analysis of over 500 high-value marketing topics found AI search visitors were worth 4.4 times more than traditional organic visitors on conversion rate, because they arrive having already compared options. A small, invisible channel can be your best one.

What we’d do

TACK has built measurement systems for 300+ brands since 2009, and we build them in this order.

Add one required field to every form. “How did you hear about us,” free text, no dropdown. Dropdowns bias the answer toward whatever you already believe. Free text tells you what buyers actually say, and after 200 responses the pattern is unambiguous.
Make the CRM the source of truth, not the ad platform. Every platform will claim the same conversion. Only the CRM knows what closed. Push offline conversions back into the ad platforms so bidding optimises to revenue rather than form fills, which is core to how we run paid media and CRO.
Set one holdout per quarter. Pause a channel in matched geographies for four to six weeks and measure the pipeline difference. It is uncomfortable and it is the only method that proves causation.
Track citation share as a leading indicator. Run a fixed prompt set across the major assistants monthly and record who gets named, which is how we report on SEO, AEO and GEO work. It moves before branded search moves, and branded search moves before revenue does.
Report on a single page. Spend, pipeline created, pipeline closed, blended CAC, self-reported source mix, branded search volume. Six numbers. Anything longer stops being read by month three.

Common mistakes

Believing the platform. Add up the conversions claimed by Google, Meta and your analytics and you will exceed the number of real customers, often by a wide margin. Every platform counts a touch it saw. None of them counts the touches it did not.

Killing channels on last-click data. Demand creation activity rarely wins the last click by design. Cutting it because it shows a poor last-click ROAS is how companies quietly remove the thing that was filling the top of the pipeline, then discover the hole two quarters later.

Waiting for perfect attribution before deciding. It is not coming back. Buyers have chosen anonymity and answer engines have removed the click. Directionally correct measurement acted on quickly beats precise measurement that arrives too late to matter.

The bottom line

The buyer journey is invisible because buyers want it that way and the interfaces changed to accommodate them. Measure incrementality, ask buyers directly, and hold yourself to blended numbers rather than platform-reported ones.

If you want a measurement model that survives the next platform change, book twenty minutes at calendly.com/tack-media-agency/talk-to-an-expert or call TACK at 310-620-1141. Engagements start at $5,000 per month.