Tokenizing the Planet

Something quietly remarkable happened this month. Two industries that have almost nothing in common — a hotel front desk and a hospital ward — began worrying about precisely the same thing, in almost the same words. Not about whether AI works. About what it costs to run, one human at a time.

I have spent the past year arguing that hospitality needs a number it does not yet keep: Token Cost Per Guest, or TCPG — the total cost of the AI tokens a property consumes, divided by the guests it serves. The argument was simple. We meter the electricity, the labour minute, even the bar of soap, yet the fastest-growing variable cost in the building — the cost of the intelligence itself — hides inside a software invoice where no one can see it, govern it, or defend it when the bill arrives.

Then I read Laura Dyrda’s reporting in Becker’s Hospital Review, under the headline “Health systems race to rein in AI costs.” It could have been written about hotels. Health-system leaders described consumption-based token billing that, mishandled, in the words of Care New England’s Tomas Gregorio, “that’ll cripple your organization” and quietly destroy the return they were promised. At Houston Methodist, Michelle Stansbury described a regime where no AI agent is built without a clear return, where teams document the economics, and where an agent that only adds cost is simply switched off.

That is my TCPG argument, arriving in a different uniform. So let me make the parallel explicit — because once you see it in two industries, you see where it is going in all of them.

TCPG — the guest as the unit

In hospitality, the denominator is the guest. Take everything the property spends on AI tokens in a period — the pricing engine that nudges the rate overnight, the messaging bot that answers at midnight, the agent that drafts the upsell — and divide it by the guests served. That is TCPG.

Its first job is humble: to make an invisible cost visible, and visible properly — fully loaded, including the orchestration, the retries, the storage, and the human who checks the output. Its second job is to protect you in either market. If the AI boom runs on, TCPG is your hurdle rate; you spend against value, not hype. If the bubble deflates and token prices collapse, the operator who knows their TCPG harvests that collapse straight to margin, while the one who bought the bundle captures nothing, because they never knew the number to begin with.

And it carries one inviolable caveat: TCPG is half of a ratio, never a figure to be minimised on its own. Chase it downward in isolation and you reach for the cheapest, weakest model in the one business where the experience is the product. The discipline lives in the ratio — cost against what the intelligence returned — never in the cost line alone.

TCPP — the patient as the unit

Port that intact to healthcare and you get Token Cost Per Patient — TCPP. Total AI token cost divided by patients served, or per encounter, or per member per month to match how payers already count. The mechanics are identical; only the denominator changes.

If anything, healthcare proves the case harder than hospitality does. The killer use case writes itself: ambient AI scribes now listen to clinician-patient conversations and draft the notes, which means tokens are spent on every encounter — the textbook condition for a per-patient meter. The operators in Dyrda’s reporting are already living the discipline without the name. They describe the consumption model as something that can cripple an under-prepared organisation; they switch off agents that only add cost; they insist, as the sector’s own analysts do, that AI must prove value rather than merely promise it.

The reporting also surfaces a wrinkle hospitality has not yet named, and should: the Goldilocks problem of provisioning. Buy too many tokens and you have paid for capacity you never used; buy too few and you hit overage and throttling; and tokens sit idle overnight when the staff have gone home. That introduces a second axis beneath cost-per-unit — utilisation — the gap between the tokens you provisioned and the tokens you actually converted into a served patient, or a served guest. It is the same discipline, one layer deeper.

The one difference worth stating plainly: in a hotel, minimising token cost toward the cheapest model degrades an experience. In a hospital, it can degrade a diagnosis. The value pairing is not best practice in healthcare — it is a safety requirement.

One law, many denominators

Stand back and the two metrics are revealed as a single idea wearing two badges. The engine is always the same — the cost of intelligence is now a metered, consumption-based variable cost — and only the denominator changes with the industry. Guest. Patient. The unit you serve.

Call the general form Token Cost Per Unit Served. TCPG and TCPP are simply its first two instances.

This is not a hospitality curiosity or a healthcare one. It is already everywhere, if you look. Dyrda’s reporting notes the tells from far outside both sectors: Uber reportedly capping employee AI use after burning its entire annual agentic-AI budget in the first three months of the year; Google reducing AI costs after token consumption multiplied roughly sevenfold year over year; Salesforce tracking whether token use actually drives business outcomes — which is to say, measuring cost against value, which is to say, doing TCPG without calling it that.

Tokenizing the planet — who is next

Every industry that serves a countable unit will need its own denominator. The pattern is portable, and the names almost suggest themselves.

Retail — Token Cost Per Basket. The conversational shopping assistant runs on every browse, whether or not it ends in a sale.

Banking and financial services — Token Cost Per Customer. Advisory copilots and fraud-triage agents that fire on every transaction, not every account.

Insurance — Token Cost Per Claim. Claims increasingly read, routed, and adjudicated by models — a per-claim meter is the only honest read.

Legal — Token Cost Per Matter. Review and drafting agents whose cost must be set against the matter’s value, not the firm’s enthusiasm.

Education — Token Cost Per Learner. Tutoring and assessment agents that, like the ambient scribe, run on every student interaction.

Contact centres and BPO — Token Cost Per Contact. The sector already meters cost-to-serve and handle time; the token line slots straight in.

Logistics — Token Cost Per Shipment. Routing and exception-handling agents priced against each parcel moved.

Airlines and travel — Token Cost Per Passenger. Disruption-recovery and service agents, metered per traveller.

Government and public services — Token Cost Per Citizen. The case with the highest stakes and the least margin for an unexamined bill.

The denominators differ. The discipline does not.

The discipline beneath the noise

There is a temptation to read all this as a story about cost-cutting. It is the opposite. A meter is not there to starve the intelligence; it is there to keep it cheap and accountable so the value it funds — the welcome, the diagnosis, the judgement, the recovery when something goes wrong — never goes hungry for budget.

We are tokenizing the planet whether we name it or not. The organisations that come through the next two years intact will be the ones who measured early — who knew their cost per guest, per patient, per claim, per citizen — while everyone else was still arguing about whether the magic was real.

The intelligence may be artificial. But the experience — the guest’s, the patient’s, the human’s — is not. The meter is simply how we keep it that way.

Token Cost Per Guest (TCPG) is a metric originated by Terence Ronson of Pertlink. The healthcare observations draw on “Health systems race to rein in AI costs” by Laura Dyrda, Becker’s Hospital Review, June 2026. Token Cost Per Patient (TCPP) is proposed here as its direct translation.