The funnel was never a map of how people buy. It was a way to draw a sales team’s pipeline on a whiteboard in 1898, and it has been quietly ruining marketing decisions ever since.
You know the shape. Awareness at the wide top, consideration in the middle, conversion at the narrow bottom. Prospects fall through it like coins through a sorting machine, and your job is to widen the mouth and reduce the friction on the way down. It is clean. It is measurable. It fits on a slide.
It is also fiction.
When Google and The Behavioural Architects ran large-scale shopping simulations across 31 categories to model how people actually decide, they did not find a funnel. They found what they named the “messy middle”: a space between the first trigger and the eventual purchase where people cycle between two modes, exploration (widening the set of options) and evaluation (narrowing it back down), over and over, sometimes for months.
There is no orderly descent. A buyer adds you to the shortlist, drops you, reads something that puts you back, forgets you, gets reminded by a colleague, compares you to two rivals, expands the set again. The journey is a scribble, not a slope.
This matters more in B2B, where the “buyer” is rarely one person. Gartner’s work on the buying journey describes six “buying jobs” (problem identification, solution exploration, requirements building, supplier selection, validation, consensus creation) that a group of roughly a dozen people revisit again and again, not in sequence but on a loop. Nobody is falling down anything. A committee is arguing in a Slack channel you will never see.
Here is the part that should change how you spend money.
Gartner found that across an entire B2B purchase, buyers spend only about 17% of their time meeting with any potential suppliers. Split that across the three or four vendors on the list and each one gets maybe 5 or 6% of the buyer’s attention. The other 83% happens without you in the room.
And by the time a buyer does turn up in your world, the decision is largely cooked. 6sense’s Buyer Experience Report found that B2B buyers are nearly 70% of the way through their process before they engage a seller, that 81% already have a preferred vendor at the moment of first contact, and that more than 90% have prior experience with at least one of the vendors they consider. The average cycle runs 11.3 months, and for most of it you are invisible to your own analytics.
Read that back. The form fill, the demo request, the “hand raise” your whole dashboard is built to celebrate, arrives after the buyer has quietly decided who they trust. You are not being moved down a funnel at that point. You are being shortlisted, or not, on memory built long before. Most of that preference forms while the buyer is still out of market entirely, nowhere near ready to act. The meeting is not persuasion. It is due diligence on a choice already made.
Which means the real contest happened while you were not measuring anything.
Now the uncomfortable bit. The channels doing the actual convincing (a podcast a founder half-listened to on a run, a recommendation dropped in a private Slack, a WhatsApp reply that said “just use these guys”, a talk someone saw at a meetup) leave no trace in your attribution model. This is what Chris Walker’s team at Refine Labs call dark social. It is most of the market, and it is unmeasurable by design.
So your attribution tool does the only thing it can. It credits the last click it can see. Someone who was sold months ago in a conversation you will never observe finally googles your name, clicks a branded search ad, and converts. The dashboard hands the ad the medal.
Refine Labs put a number on the gap. In a twelve-month study of 620 declared-intent conversions worth $21.5m in closed-won ARR, self-reported attribution disagreed with software-based attribution by around 90 percentage points. Podcasts were named by buyers as the source behind 53% of revenue. The software gave podcasts zero.
Your attribution model is not measuring what persuades people. It is measuring what happens to be standing nearest the till when they decide to pay.
Sit with the incentive that creates. Every quarter, the last-click harvesters look brilliant and the demand-creators look like a cost. So budget flows to the harvest and away from the sowing. You cut the podcast, kill the point of view, defund the thing people actually repeat to each other, and for two quarters the numbers hold because there is a backlog of demand still ripening. Then the pipeline thins and nobody can say why. You optimised the machine to death by measuring the wrong end of it.
None of this is an argument to stop measuring. It is an argument to stop pretending your model delivers verdicts. It offers directions. Hold it that loosely and it becomes useful again.
Three moves change the game more than a better dashboard will.
Ask people directly. Put a “how did you hear about us?” box on the form and read the free text, not the dropdown. It is unfashionably low-tech and it consistently surfaces the podcasts, posts and peer recommendations that the tracking never credits. When self-report and software disagree, the human is usually closer to the truth, because the human was there.
Test whether a channel actually causes anything. Turn it off in a region for a month. Hold back a group. If pipeline moves, the channel was doing work; if nothing moves, you were paying for coincidence. Incrementality is a blunt instrument and it is worth more than a year of tidy multi-touch reports.
Then spend to be present and memorable across the messy middle, not just to catch the hand-raise. Show up in the rooms where buyers actually loop: the communities, the shows, the inboxes of the people they ask. Say something a person would repeat without a link attached. That repetition is the mechanism, and it is precisely the part your model cannot see.
The founders who win the next few years will not be the ones with the cleanest funnel. There is no funnel to clean. They will be the ones who made peace with the mess, kept showing up in the 83% of the journey they cannot track, and stopped firing the channels that were quietly doing all the work.
Stop optimising a shape that does not exist. Start earning the memory that does.
The best work starts with a conversation, and a decision worth making.
Get in touch