After the Funnel.

What the funnel actually assumed

Strip the picture back and three assumptions are carrying it.

The first is that the buyer gathers information personally. A person reads the page, downloads the paper, sits through the webinar, takes the call.

The second is that gathering is expensive, and therefore gradual. Attention is scarce, evaluation runs over weeks, and the vendor gets repeated opportunities to be present during that stretch. Nurture sequences exist entirely because of the gap between one act of human attention and the next.

The third is that persuasion accumulates. Each impression leaves a residue, and the residue eventually becomes preference.

A funnel is not a model of a buyer. It is a model of an information asymmetry, and of how long a human takes to close it.

Remove the time and the diagram loses its subject.

The gathering moves

What is changing is not the buyer's psychology. It is who performs the search.

A buyer states a problem in ordinary language. An assistant retrieves, compares, discards, and returns a short list with reasons attached. By the time a human looks at anything, the field has already been narrowed by a process that read faster than any person could, did not get tired, and did not encounter your retargeting.

The first thing the buyer sees is not your page. It is a synthesis in which you either appear or do not.

This is not yet how most enterprise buying works, and I want to be careful about that. It is how a growing share of the first pass works, and the first pass is the part the funnel was built to own.

What happens to each band

The top of the funnel was awareness, measured in impressions. Under an assistant, awareness becomes retrieval eligibility. Either the system can find and parse your position on the question, or you were absent from the consideration set before anyone knew a consideration set was being formed.

The middle was where nurture lived. It compresses hardest, because the machine has no residue. It does not warm to you over six months. It holds a corpus and a set of citations, and the seventh impression means nothing to it because there was never a first one in the sense the funnel meant.

The bottom is largely intact. The decision, the risk, the negotiation and the signature remain human, and they remain slow. What has changed is that the human arrives at the bottom holding a shortlist they did not personally assemble.

So the honest description is not that the funnel is dead. It is that its front half has been intermediated, and the vocabulary built for that half now describes a process nobody is running.

The question that replaces it

Under the old model, the qualifying question at each stage was whether a human had been moved: did they notice, did they engage, did they come back.

The replacement question is colder and considerably easier to fail.

Can a machine retrieve, verify and cite what you published.

Three verbs, three separate failure modes.

Retrieve. Is the claim in a form that can be found and parsed at all, or is it inside a video, a gated PDF, a carousel, or a paragraph that never states its own conclusion. A position that requires a human to infer it does not survive retrieval.

Verify. Is the claim traceable to something that stands up. A date, a jurisdiction, a named framework, a version, a source that can be checked independently. A claim with nothing behind it is not treated as wrong. It is treated as unsupported, which in practice is worse, because wrong claims get corrected and unsupported ones get dropped.

Cite. Is the material attributable to a named entity, at an address that persists, in a form that can be pointed back to. Attribution is not vanity here. It is the mechanism by which a machine can hand accountability back to a human at the end of the chain.

Most marketing copy fails all three. Not because it is dishonest, but because it was built for a different job: to produce a feeling in a reader who was already paying attention. It carries no checkable claim, because a checkable claim was never what it was for.

What this asks of published expertise

The practical consequences are less glamorous than the shift sounds.

Structure stops being a formatting preference. If the argument is buried, the argument does not exist. A page that states its claim, defines its terms once, and answers the question it is named after is not writing for a search engine. It is writing so that the claim can be lifted out intact.

Provenance stops being an academic habit. Regulation has dates and version numbers. Frameworks have authors and revisions. An assertion tied to those things can be checked; an assertion floating free cannot, and the difference now determines whether the assertion travels.

Volume stops being self-evidently rational. Under the funnel, publishing more was defensible because impressions accumulated. Under retrieval, thirty near-identical pages on the same topic are a liability rather than an asset. The system selects one passage, not thirty, and a diluted corpus makes that selection worse.

Publication is no longer the act of putting something where people will see it. It is the act of putting something where a machine can check it.

Where this connects to callable expertise

The argument I keep returning to on this site is that expertise is becoming callable: structured so that a system can retrieve it, apply it, and hand back an answer carrying its own evidence and its own limits.

The buying question and the callable question turn out to be the same question approached from opposite ends. An assistant shortlisting vendors is doing a crude version of what an agent consulting a body of expertise does. It queries, it evaluates what it gets back, and it discards anything it cannot stand behind. The buyer's assistant is the first caller most published expertise will ever have, and it is a demanding one, because it has no relationship with you and no reason to be generous about ambiguity.

That is why I treat structure and provenance as architecture rather than as polish. They are the conditions under which knowledge remains usable once a person is no longer the one reading it.

What I am not claiming

Several things here are unsettled, and pretending otherwise would be the same mistake the funnel's defenders make in the other direction.

I do not know what share of buying journeys currently involves an assistant doing the gathering, and neither does anyone quoting a precise figure. The measurement infrastructure for this barely exists.

Relationships, referrals, conferences and reputation still route an enormous amount of enterprise buying, probably more than machine retrieval does today. The people who decide are still people.

The retrieval layer itself is unstable. What gets surfaced and cited changes with the model, the product surface and the index behind it, and any strategy built on the specifics of today's behaviour will be recalibrating within a year.

And the funnel is not dead. Stages remain a useful internal fiction for organising work, in the way that any simplification is useful as long as everyone remembers it is one.

What survives

The stages were always a proxy for something more durable. Underneath awareness, consideration and decision sat a single question: at the moment this problem gets named, does this organisation appear, and does what it says hold up under examination.

That question is not going anywhere. What is changing is who asks it first, how quickly they ask, and how little patience they have for material that cannot answer.

The funnel measured whether a person had been persuaded. The thing replacing it measures whether the work can be checked. Those are not the same discipline, and the second one is considerably harder to fake.

© 2026 Nabeel Khan. After the Funnel is published under CC BY-NC-ND 4.0. Quote it, cite it, do not repackage it.