How Do You Measure Demand You Cannot Attribute?

How Do You Measure Demand You Cannot Attribute?

You stop trying to attribute it and start asking buyers directly. The demand that shows up as "direct" traffic in your analytics is real, it is often your best-converting demand, and no amount of tooling will tell you where it came from. The fix is a question on your form, not a better dashboard.

This is the argument we have most often with marketing teams. Somebody wants to cut the channel that cannot be attributed, and the channel that cannot be attributed is usually the one doing the work.

Here is why the data is genuinely missing, what that means for how you report, and what to do instead.

Why Does So Much Traffic Show Up as Direct?

Because browsers deliberately stopped telling you. The default referrer policy in modern browsers is strict-origin-when-cross-origin, which MDN documents as "the default policy if no policy is specified, or if the provided value is invalid," noting that "previously the default was no-referrer-when-downgrade."

Under that policy, a cross-origin request sends only the origin, not the path or query string. So you can learn that someone came from a domain, and not which page. And when a site sets no-referrer, MDN states the header "will be omitted: sent requests do not include any referrer information," so you learn nothing at all.

Add the places a link gets pasted where there is no referrer to send. A Slack message. A WhatsApp thread. An email client. A PDF. A conference slide someone photographed. Google Analytics 4's own definition of Direct is "the channel by which users arrive at your site/app via a saved link or by entering your URL," matched when "Source exactly matches (direct) AND Medium is one of ((not set), (none))." That bucket is where all of the above lands.

Is This Getting Better or Worse?

Worse, and search is the biggest reason. SparkToro published a study on 9 June 2026, authored by Rand Fishkin, using Similarweb's desktop and mobile web panel for January to April 2026 in the US. It found 68.01% of Google searches ended without a click, up from 60.45% in 2024.

The 2024 version of that study framed it more starkly: per 1,000 US Google searches, "only 374 clicks go to the open web."

So the trend runs in one direction. More of the buyer journey happens in places that send you no data: inside search results, inside AI answers, inside private messages, inside someone's inbox. Treating unattributed demand as an error to be fixed is going to get less workable every year, not more.

What Is Actually in Your Direct Bucket?

Four quite different things, and lumping them together is why the number feels useless. There is genuine direct traffic, someone typing your URL because they already know you. There is private sharing, a link passed in a message. There is stripped referrer traffic, a real referral whose source got removed. And there is everything GA4 could not classify, which its documentation calls Unassigned, "the value Analytics uses when there are no other channel rules that match the event data."

Only the first of those is truly direct in any meaningful sense, and it is the one that most clearly indicates brand strength. The other three are referrals wearing a disguise.

You cannot separate them in the data. You can separate them in conversation, which is the whole point of what follows.

What Question Should You Put on Your Form?

One open field: "How did you hear about us?" Free text, not a dropdown, and not required. That is it.

Make it free text because a dropdown constrains answers to the channels you already believe in, which defeats the purpose. The value is in the answers you did not anticipate: a specific podcast, a named person, a Slack community, a conference talk, a comparison article you did not write.

Make it optional because a required field on a demo form costs you conversions, and the people who skip it are not the people whose answer you need. In our experience a meaningful share of people answer it anyway, and their answers are more useful than any attribution model, because they name things no model contains.

Is Self-Reported Attribution Reliable?

Not precisely, and that is fine, because you are not using it for precision. People misremember. They name the last touch they consciously noticed rather than the first one that mattered. Someone who read three of your articles and then heard you mentioned on a podcast will say "podcast."

What self-reported data is good at is telling you which channels exist in your buyers' world at all. If 30 people write in the name of a community you have never invested in, that is a finding no dashboard would ever produce. If nobody ever mentions a channel you spend heavily on, that is also a finding.

So use it directionally, in aggregate, read monthly rather than per-deal. Compare it against your platform data and treat disagreement as information rather than as an error in one of them. That posture is the same one we argued for in B2B SaaS attribution.

What Else Can You Measure That Is Not Attribution?

Three things, all of them leading indicators of demand you cannot trace. Branded search volume, which tells you whether more people are actively looking for you by name. Direct traffic volume as a trend rather than a source. And the conversion rate of unattributed traffic compared with everything else.

That third one is the argument-ender in most internal debates. Unattributed traffic usually converts better, because someone arriving with no referrer often arrives with intent: they were told about you by someone they trust. If your direct traffic converts at three times your paid traffic, cutting the activity that generates it is a strange plan.

Branded search is the cleanest of the three to track over time, and we went into how in measuring branded search demand.

How Should You Report This to a CFO?

Honestly, with the unattributable portion named as a line rather than hidden. Our recommended shape is three buckets: demand we can trace, demand we cannot trace, and total. Then, next to the untraceable bucket, the self-reported answers as qualitative evidence.

The instinct to force everything into an attributed model is understandable and it backfires. A model that assigns 100% of pipeline to trackable channels is not more rigorous than one that admits 40% is unclear. It is less rigorous, because it has invented a precision it does not have, and a finance team that later discovers this will trust none of your numbers.

Saying "we cannot trace 38% of pipeline, here is what those buyers told us" is a stronger position than a tidy pie chart. It is also more defensible when someone asks how the number was produced, which is why we prefer efficiency measures over attribution fiction, as in CAC payback versus CAC ratio.

Does This Mean You Should Stop Tracking?

No. Track everything you can track properly, and stop pretending the rest is knowable. Tag your own links consistently, set a referrer policy on your own site deliberately, keep your channel definitions stable so year-on-year comparisons mean something.

Where tracking genuinely helps is in comparing like with like over time. Your paid channels are measurable and should be measured hard. Your email is measurable. Your organic search is partly measurable. Those numbers are real and worth optimising.

The mistake is extending that standard of proof to everything. Demanding attribution before funding an activity systematically defunds word of mouth, community, podcasts, conference talks and reputation, which is to say it defunds most of what actually makes a B2B company known. That bias is the real cost, and it is the bias we described in brand against performance marketing.

What Would We Do First?

Three things this month, none of which requires new tooling. Add the open text question to every form you own and start reading the answers weekly. Pull your last twelve months of direct and unassigned traffic and compare its conversion rate to every other channel, so you know what you are dealing with. And take a baseline of branded search volume so you have a trend line to look at in six months.

Then hold one rule internally: no channel gets cut on the basis of missing attribution alone. Cut it because buyers never mention it, because the numbers you can see are bad, or because you have something better to do with the money. Not because the data is absent, since absent data is now the default condition rather than a signal.

The deeper point is that measurement has got worse while marketing has not. That is uncomfortable, and pretending otherwise produces confident decisions built on invented certainty. The honest version, with a named gap and real buyer quotes filling it, is less tidy and much more useful.

If you want help building the parts you can measure and reporting the parts you cannot without fudging it, we are happy to talk. You can find us at phoenix.studio.