AI firms urged to ditch tokens for outcome pricing

For an industry obsessed with innovation, AI has landed itself with a surprisingly old problem: how exactly should they charge for it? And right now, the dominant answer – tokens – is fundamentally flawed.

The bridge that failed

The AI industry has become obsessed with counting tokens, and that's a problem.

There are already stories of organisations burning through millions of tokens with little, if anything, to show for it. Some even created internal leaderboards celebrating token consumption, as if usage alone was proof of value. Unsurprisingly, the enthusiasm faded when the business case failed to materialise.

Tokens were meant to be an elegant pricing mechanism: simple, scalable and consumption-based. Pay for what you use. What's not to like?

The reality is far less compelling.

Tokens don't measure outcomes; they measure activity. They count words in and words out but remain completely blind to the value created. Whether AI generates a meaningless paragraph or eliminates hours of specialist human effort, the pricing metric is largely the same.

That's the fundamental flaw. Businesses buy outcomes, not token throughput.

Imagine hiring an auditor who charged based on the number of words in the documents they read rather than the risks they identified. It would be absurd. Yet that's effectively what token pricing does. Two organisations can achieve the same outcome, uncover the same risks and realise the same value, while incurring dramatically different costs simply because one task consumed more tokens.

Put simply, token pricing confuses consumption with value.

The predictability illusion

If token pricing is flawed as a measure of value, it's even worse as a mechanism for cost control.

Enterprises are struggling because token consumption is inherently unpredictable rather than due to a lack of planning. 

AI adoption doesn't grow in a straight line. Prompts get longer. Users become more sophisticated. New use cases emerge. What starts as a modest budget line can quickly become a runaway cost centre.

That's exactly what organisations are now discovering. In the rush to embrace AI and avoid being left behind, businesses focused on proving adoption rather than managing economics. The result is escalating spend, limited visibility and, in some cases, ROI that has all but disappeared.

The real irony is that token-based pricing punishes success.

The more valuable an AI use case becomes, the more people use it. The more people use it, the higher the bill. Scale, which should improve economics, does the opposite.

That's not how enterprise technology is supposed to work. Organisations invest in technology to create operating leverage, where costs grow more slowly than value. Token pricing breaks that relationship. It turns AI from a scaling engine into a tax on adoption.

For a technology built on the promise of efficiency, that's an uncomfortable contradiction.

Making it work

The companies extracting real value from AI are the ones applying the same commercial discipline they would to any other investment, all with three things in common.

Firstly, they're ruthless about where AI gets deployed. They don't scatter it across the organisation in the hope that value will somehow emerge. They target specific processes where AI can eliminate meaningful cost, accelerate delivery, or improve outcomes. Every deployment has a business case attached to it. If the value can't be measured, it doesn't get funded.

Secondly, they focus on economics that executives actually understand. Not tokens consumed, context windows used, or prompts executed. Instead, they measure cost per contract reviewed, cost per ticket resolved, cost per proposal generated or cost per forecast produced. These are metrics that connect technology spend directly to business outcomes. They create accountability. More importantly, they expose whether AI is genuinely delivering value or simply generating activity.

And lastly, they build for scalable returns. As adoption grows, value grows faster than cost. The economics improve through scale, standardisation, and process redesign. Success creates operating leverage, not larger invoices.

None of this is revolutionary. It's basic commercial common sense.

But it requires a level of discipline that the token model actively works against. When the primary metric is consumption, the conversation inevitably shifts towards usage rather than value.

Instead of asking "How many tokens did we consume?", start asking a far better question like "What business outcome did we buy?"

What needs to change

So, what's the alternative?

A return to rate cards and time-and-materials pricing feels like trying to buy a SpaceX ticket with gold sovereigns. It doesn't fit the technology.

But persisting with token pricing despite its obvious shortcomings is inertia.

AI should be priced around outcomes, not consumption.

Outcome-based pricing won't work for every use case, and defining measurable outcomes isn't always straightforward. But that's a far better challenge than relying on a metric with little relationship to business value.

Outcome-based pricing changes the conversation entirely. Instead of debating prompt volumes, token consumption and model utilisation, both parties focus on the same objective: delivering measurable business results.

The organisations that are pulling are measuring contracts reviewed, tickets resolved, code deployed, and hours eliminated, not prompts executed and tokens burned.

It's time to stop counting tokens and start counting outcomes.