The 'Token Cost Per Guest' Framework Hotels Need Before Pricing Shifts Again
A new 'Token Cost Per Guest' (TCPG) framework argues hotels should route commodity guest tasks to $15/million-token open models and reserve $50/million-token frontier models for judgment calls — not lock into one vendor for everything.
Terence Ronson’s new framework gives hotels a way to price their AI stack by task, not by vendor loyalty: cheap, open-weight models — he cites Kimi K3 at roughly $15 per million tokens — for high-volume commodity work, and premium frontier models around $50 per million tokens reserved for judgment calls.
Ronson calls it “Token Cost Per Guest” (TCPG). The commodity tier handles guest FAQs, review summarization, and translation — high-volume, low-stakes tasks where a cheaper model is indistinguishable from an expensive one. The premium tier is reserved for judgment-intensive work: orchestrating multi-step workflows, maintaining brand voice, compliance-sensitive decisions. The “Camry” metaphor is deliberate — a reliable, economical option for everyday needs, with a premium option held in reserve for what actually requires it. His warning: properties locked into a single expensive model for every task are the most economically exposed over the next 12 months, while properties architected for model-agnostic routing stand to benefit most as token pricing and capability keep shifting.
The stakes of routing the wrong task to the wrong tier aren’t abstract. A recent piece on AI-mediated hotel reputation describes how AI assistants now answer “tell me about this hotel” with a single synthesized verdict rather than a list of links, and warns independent properties are more exposed because they have fewer sources for that verdict to draw from — exactly the kind of brand-voice-sensitive, judgment-heavy task Ronson would route to the premium tier, not a $15 commodity model. By contrast, the Kenilworth Hotel case — a QR-code web app moving routine guest requests off the front desk — is commodity-tier automation working precisely because the task was low-judgment routing, not guest-facing brand voice: the kind of high-volume, low-stakes work TCPG assigns to the cheap tier.
The practical test for any GM: does your current AI stack have a lever to move a task down-tier when it’s commodity work, or are you paying premium-model prices for FAQ answers because switching vendors felt like too much friction? Ronson’s argument is that the operators building in that lever now are the ones who won’t get caught flat-footed the next time token pricing moves.
Source: Terence Ronson, Pertlink — 'The Camry Guest Experience' Auto-generated brief — verified before publishing.