What’s the best way to pay for AI? Most people think AI free. It’s not. Like everything in life, it has to be paid for somehow. The free versions are designed to get you hooked.
So, what’s a fair way to charge for AI? Goldman Sachs says there’s an interesting shift from per-user licensing to outcomes-based pricing already underway.
The per-seat model made sense for traditional software. You buy a licence, someone uses it, you pay for the access. Clean. Simple. Familiar.
But AI isn’t traditional software. And charging for access to something that actually does things (rather than just enabling you to do things yourself) is a fundamentally different proposition.
The current token-based pricing model causes a lot of misunderstanding because it’s difficult to predict.
Tokens are the units of text AI models process. Roughly three quarters of a word each, if you want a rule of thumb. Every question you ask, every document you analyse, every response you receive burns tokens. And the bill adds up in ways that are genuinely difficult to predict or explain to a finance director.
We’ve seen it ourselves. Processing 170,000 rows of data across multiple Salesforce tables. Brilliant result, solved a problem nobody else could crack, but it cost roughly £50 of tokens and registered as two prompts in the system. Try explaining that billing model to a procurement team.
One of Kieron’s son’s AI agents burned through his budget in minutes because two of the agents started arguing with each other. Oscar’s solution was to stop them talking directly and route everything through a project manager.
Token pricing isn’t wrong. It’s just difficult to explain in a way that makes people (especially Finance people) deeply uncomfortable.
Which brings us to outcomes-based pricing. And why it makes so much more sense.
A college currently pays £20 every time an outsourced company checks a student’s application documents, verifying they’re in date, complete, and genuine. It’s manual. It’s slow. It’s expensive.
KnowledgeFlow does the same thing for about 30p of AI processing.
So what’s the right pricing model? A monthly seat licence disconnected from value? A token bill nobody understands? Or a per-outcome charge that directly connects what you pay to what you receive?
The answer feels obvious. And it’s where the industry is heading.
For buyers, this is good news. You pay for results. No more shelfware. No more token bills that arrive like a surprise at the end of the month.
For AI providers, it’s a healthy discipline. If your pricing is tied to outcomes, you’d better be confident in your outcomes.
We’re already having these conversations with some of our customers. Not because we have to, but because it’s the right way to think about the value AI can provide.
Watch this space because pricing models will change, and AI costs will inevitably rise. Inform yourself before you get any nasty surprises.
