A vendor I worked with in financial services spent the better part of last year defending a marketing budget. The number was real money — a mid-seven-figure demand engine, built over a decade, with a content team, a partner channel, an SDR floor, and an attribution model that could tell you which webinar touched which logo. The model was beautiful. It was also describing a building that buyers had quietly stopped walking into.
B2B buying behavior didn't change because someone decided it should. It changed because the tools sitting between a buyer and a decision changed, and most go-to-market leaders are still measuring the old building. If you run demand generation, content, or channel partnerships in pharma, manufacturing, or financial services, the uncomfortable part is this: the place where deals now get shaped is a place your dashboards cannot see.
The verdict, in one sentence: the belief that your funnel reflects how buyers actually decide is a leftover from a measurement era that has quietly ended, and the source of that belief is thinner than the confidence built on top of it.
Where the belief came from
It helps to trace how the field came to trust the funnel in the first place, because the trust has a specific origin and the origin was always partial.
In the 2000s, B2B marketing borrowed a story from a 2011 piece of research that became gospel: that buyers were already most of the way through their decision before they ever spoke to a salesperson. The number that traveled was "57 percent of the purchase decision is complete before contact." It got cited in keynote after keynote. It justified an enormous reallocation toward content marketing and digital demand generation, because if buyers were educating themselves, the job was to be present during that self-education.
Here is the thin part. That figure came from a particular study, a particular sample, a particular moment — and it was always a measurement of the channels we could observe. We saw search, we saw site visits, we saw gated content, so we declared those things the journey. The funnel was never a map of how people decided. It was a map of where our instruments could take readings.
For about fifteen years, that limitation didn't hurt much. The instruments and the behavior overlapped well enough. A buyer in industrial procurement really did Google a component, land on a spec sheet, download a whitepaper, and enter a nurture track. The reading and the reality lined up, so we forgot they were two different things.
What broke the overlap
The overlap broke when a layer of software inserted itself between the buyer's question and the open web. When a hospital's clinical staff want evidence on a therapy, a meaningful share of them now query a medical AI tool rather than a search engine — by some industry reporting, a large fraction of U.S. physicians use one such tool routinely, with query volumes in the tens of millions per month. When an HVAC contractor sizes a system, the question increasingly goes to a chat assistant, not a manufacturer's site. When a finance team scopes a vendor category, the shortlist often arrives pre-assembled by a model summarizing sources the buyer never clicked.
Industry analysts have started putting numbers on the shift. IDC has projected that a majority of B2B demand generation will be AI-mediated within a few years. Survey work on B2B technology purchasing has found buyers compressing journeys that used to take the better part of a year into roughly a quarter. The direction is consistent across every credible source: the self-education phase you built your content strategy to serve is moving inside systems that don't send you a referral, don't fire your pixel, and don't appear in your attribution model.
So your readings haven't gotten wrong, exactly. They've gotten narrow. You are measuring the part of the journey that still passes through your instruments, and concluding that's the whole journey, for the same reason your predecessors did in 2011. The belief survives because the dashboard still fills with data. It just isn't the data that decides anything.
How I'd actually decide what to trust
When a B2B leader asks me how worried to be, I don't answer with a maturity model. I ask three questions, because they separate the genuinely exposed from the merely anxious.
| Question | What a worrying answer looks like |
|---|---|
| Where does your category's buyer ask their first real question? | "We assume it's search and peer referral" — assume, not verified |
| Can you see your own content the way a model sees it? | Nobody on the team has checked what an AI tool returns for your top buying queries |
| What share of pipeline arrives already shortlisted? | Rising "we were one of three they called" with no traceable touch before that |
The first question is the whole game. If the buyer's opening question now lands inside a system that synthesizes an answer, then everything downstream — your content, your channel, your sales motion — is competing to be a citation in someone else's summary, not a destination. That is a different job than the one your team was hired for, and pretending otherwise is the expensive mistake.
Who should act on this, and who shouldn't
This matters most if you sell into pharma, manufacturing, or financial services, where the buyer is a professional resolving a high-stakes question and is exactly the kind of person reaching for a synthesis tool to do it faster. If your category is genuinely relationship-led — long procurement cycles driven by a handful of named accounts and human trust — the shift is real but slower, and a panic reallocation would cost you more than the threat does this year.
It also matters less if you're a small team without the budget to maintain a sophisticated funnel in the first place. You have less to unlearn. The hardest position is the one my financial-services client was in: a large, well-instrumented operation whose very sophistication makes the blind spot harder to admit, because admitting it means conceding that the beautiful model measures a shrinking room.
What this looks like lived out
I'll tell you what I changed in my own practice, not as a prescription but as evidence of where the logic leads.
Once a month I take the five questions a buyer in a client's category would actually type, and I put them into the AI tools that buyer is likely using. I read what comes back, and I note whether the client is mentioned, how, and beside whom. It takes about forty minutes. The first time I did it for a manufacturing client, they weren't in any of the five answers, and a distributor they'd been losing to was in four of them. No dashboard they owned had ever shown that, because no dashboard they owned was looking at the room where the question got answered.
The funnel still fills. The question is whether you've checked who is being read aloud in the room you can't see.
