The feature is not the position.

Open eight banking AI vendor sites and you will read the same page. Real-time scoring. Explainable outputs. Model monitoring. Low-code integration into the core. Every one of those claims is probably true, and not one of them tells a risk committee why your name belongs on the shortlist.

A feature list is an inventory of what the engineering team shipped. A position is a claim about where you sit inside a decision the buyer is already making, and it has to hold up when nobody from your company is in the room.

Which is most of the time. Gartner surveyed 646 B2B buyers between August and September 2025 and found that 67% would prefer to buy with no rep involved at all. Forty-five percent said they used AI somewhere in a recent purchase, which means your feature page is now being summarised by a model before a human ever reads it. Fluent copy that says nothing distinctive gets compressed into nothing.

Start with the change, not the capability

Gartner’s buying research also carries a number that should reorganise how you write: 99% of B2B purchases are driven by organisational change. Nobody wakes up wanting fraud scoring. They want it because something moved.

In banking those movements are unusually legible. A new chief risk officer inherits a model inventory nobody trusts. A fraud loss line finally reaches the board pack. A core migration opens an integration window that will close in eighteen months. A DORA register of information gets assembled for the first time and reveals how many providers sit under one critical function.

Position against the movement. “Scores transactions in under 40ms” is a specification. “Gets a fraud model through second-line validation in one cycle instead of three” is a position, because it names the thing that is actually stuck.

The alternative you are really up against

April Dunford’s framework breaks positioning into five parts: competitive alternatives, unique attributes, the value those attributes create, the characteristics of buyers who care most, and the market category you choose to sit in. Vendors in this sector reliably get four of them roughly right and the first one badly wrong.

Ask a fintech AI team who they compete with and you get a list of other fintech AI vendors. Ask the bank and you get a different answer: build it internally with the data science team who are already half-finished; wait for the incumbent core provider to ship something like it in a release next year; buy a smaller module from a supplier already inside the building for a different purpose; do nothing for another budget cycle and manage the loss.

Those alternatives have advantages your competitor set does not, and the biggest one is friction. An internal build needs no procurement, no security review and no new line in the register of information, and it comes with political cover for whoever proposed it. The incumbent is already contracted. Doing nothing costs zero this year, which matters enormously in a year when the cost base is under pressure. If your positioning only beats other vendors, you win the bake-off and lose the decision.

“AI” was never a position and it is now less than one

In 2019 MMC Ventures reviewed 2,830 European companies that public databases had classified as AI companies. In roughly 40% of cases the researchers found no evidence that AI was material to the value proposition. Worth being fair about the finding: most of those labels were applied by third-party trackers rather than claimed by the companies themselves, and the firms simply did not correct them. The label was cheap, so it inflated.

Seven years on, the label has stopped carrying information in the other direction too. The Bank of England and FCA found that 75% of UK financial services firms were already using AI, with a further 10% planning to start within three years. When three quarters of your buyers already run AI in production, telling them you use AI describes them as accurately as it describes you.

The category still matters, though. Choosing to be “an AI fraud platform” puts you in a comparison set where explainability and latency are table stakes and price gets squeezed. Choosing to be “model validation acceleration for financial crime teams” puts you in a set of two or three, where the buyer’s frame of reference becomes the cost of a validation cycle rather than the cost per transaction scored. The product underneath is identical in both cases; the second framing just gives the buyer a comparison you can win.

A test you can run this week

Take your homepage hero and your first discovery slide, and check four things.

• Could a competitor put their logo on it without changing a word? If yes, you have written a category description, not a position.
• Does it name a decision or a moment, or does it name a capability? “Before your next model validation” beats “AI-powered risk intelligence”.
• Would the person who has to defend this purchase internally repeat the sentence verbatim in a committee, in their own voice, without embarrassment? That is the only distribution channel that matters, because they will be in rooms you never enter.
• Is there a number attached that a bank could check? Cycle time, false positive rate, hours of analyst review removed, weeks saved on onboarding. Unverifiable adjectives get discounted to zero by second line.

What changing the position actually costs

Repositioning is cheap in production terms and expensive in political terms. The deck is a day. The website is a fortnight. The hard part is that a real position excludes buyers, and somebody in your commercial organisation has built a pipeline number on the buyers you are about to exclude.

That argument is worth having in the open rather than settling by drift. A position nobody will defend in a forecast review collapses back into a feature list within a quarter, and you end up paying for the rewrite twice.