Your AI has a meter now

THE VALUE GAP · No 2

Your AI has a meter now.

The all-you-can-eat licence is being retired. You cannot claim value you have not netted against the meter.

———

“A cynic is a man who knows the price of everything and the value of nothing.”

— Oscar Wilde, Lady Windermere’s Fan, 1892

For most of the last two decades, enterprise software was sold the way a gym sells membership. You paid a fixed fee once a year, you got a licence, and whether you used the thing twice a day or ten thousand times, the number on the invoice did not move. The economics were simple, and they were forgiving. Waste was invisible because waste was free. You could deploy badly, adopt half-heartedly, leave whole modules dark, and the finance conversation never changed. The meter, such as it was, did not run.

That era is closing, quietly and fast. AI has a meter now, and it is running while you read this.

Look at what the platforms themselves have just done. ServiceNow — the backbone of service management for much of the enterprise world — has set its legacy licensing to end-of-sale on 1 July 2026, replacing it with three AI-native tiers in which the AI “assists” are consumption-metered rather than bundled into a flat seat.¹ This is a vendor decision, and it is not an isolated one. GitHub has moved Copilot onto token-based billing, so the cost of a coding assistant now rises and falls with what it actually consumes.² Salesforce prices its Agentforce agents at roughly ten cents per action — a figure the company itself has put forward, and one worth sitting with, because an agent that acts a million times a month is no longer a rounding error.³ The pattern across all three is the same. The all-you-can-eat licence is being retired. In its place is a taxi meter.

A meter changes behaviour. It makes every action visible and, eventually, accountable. On the face of it this is good news for anyone who cares about value, because the thing that was once free to waste is now priced. But there is a trap folded inside the meter, and it is the whole subject of this piece.

———

Roughly 72% of the total cost of running enterprise AI sits outside the model invoice altogether — in integration, data pipelines, orchestration, storage, security and the human oversight around it.⁴

— EY

———

The meter on your invoice measures one thing: the model calls. It does not measure the rest. EY’s work on the true cost of enterprise AI puts nearly three-quarters of the total spend outside that visible line item, in the plumbing and the people around the model.⁴ The FinOps Foundation, whose entire discipline is the accounting of cloud and AI cost, has put it more sharply still: there are nine distinct cost layers in a production AI system, and only one of them is metered on the bill you actually receive.⁵ You can watch the token counter tick over in real time and still have almost no idea what the system costs you, because the counter is watching the cheapest floor of a nine-storey building.

This is where consumption pricing meets the argument of the book this series accompanies. The value gap has never been a technology problem — it is an approach problem, and the meter has just made the approach problem impossible to ignore. For twenty years you could run an AI or a platform program without ever netting its value against its cost, because the cost was a fixed line you had already agreed to and mentally written off. Now the cost moves with usage, most of it is invisible, and the old habit — count the activity, admire the adoption curve, declare a win — produces a number that means nothing. Prompts served is not value. Actions taken is not value. Tickets deflected is not value. Each of those is a reading on the meter, and a meter reading is a cost, not a return.

You cannot claim value you have not netted against the meter.

That line is the discipline in one sentence. Value is what is left after you subtract what it cost to produce — all of what it cost, including the roughly three-quarters that never appears on the invoice. A program that reports a benefit without netting it against the full consumption cost is not reporting value. It is reporting activity with a dollar sign in front of it, which is a more dangerous thing, because it looks like proof.

Here the book’s oldest observation does the heavy lifting. Technology amplifies whatever you already have. Give a metered, consumption-priced AI to an organisation that already measures outcomes net of cost, and the meter becomes a gift — a live, honest signal of where value is being made and where it is merely being spent. Give the same meter to an organisation that has never named the outcome or netted the cost, and it becomes a bill that grows every quarter for a benefit no one can prove. The meter does not create the discipline. It reveals whether the discipline was ever there. And the discipline is not new. Netting realised value against true cost is the same test that separated the platform investments that paid back from the ones that merely went live — the ones that lifted IT maturity, drove real automation and improved the experience, from the ones that bought a licence and hoped. AI has only made the bill impossible to ignore.

Which returns, as this work long does, to a single question: who in your organisation owns the outcome of this system, net of what it consumes? Not who approved the licence. Not who watches the usage console the vendor so helpfully provides. Who is accountable for the number that remains once the full meter — all nine layers of it — has been subtracted. If you can name that person, the shift to consumption pricing is the most useful thing to happen to your business case in years, because it finally gives them a true cost to net against. If you cannot name them, the meter will keep running regardless, and the gap between what you spend and what you can prove will widen with every action your agents take.

Mind the gap. It was there long before the meter. Consumption pricing has simply installed a meter beside it, and switched it on.

The organisations that pull ahead from here will not be the ones that spend the most on AI, nor the ones that spend the least. They will be the ones who can stand in front of the meter and say, with a number they would defend, exactly what was left over once it stopped.

———

THE VALUE GAP — The full argument, and the BRIDGE framework for closing the gap between what you spend and what you realise, is in Bridge The Value Gap, out now at rodneyhobbs.com.

———

References

¹ ServiceNow, licensing transition to AI-native tiers with consumption-metered assists; legacy SKUs end-of-sale 1 July 2026 (vendor claim).

² GitHub, move of Copilot to token-based billing, 2025.

³ Salesforce, Agentforce pricing of approximately US$0.10 per action, 2025 (vendor claim).

⁴ EY, analysis on the total cost of enterprise AI — approximately 72% of cost sitting outside the model invoice, 2025.

⁵ FinOps Foundation, FinOps X — “nine cost layers, only one metered,” 2025.