The mistake big tech already made
Meta, Amazon, OpenAI, Microsoft, Disney, Paramount, and Uber all spent the last cycle pushing internal "tokenmaxxing" — leaderboards, performance reviews, and incentive structures that rewarded engineers for consuming as many AI tokens as possible. Meta's internal leaderboard reportedly logged 73.7 trillion tokens in roughly a month before the program was quietly reversed. The bills arrived. The business outcomes didn't.
Terence Ronson's argument in Hospitality Net is that hospitality is now lined up to make the same mistake — except with one critical difference. Hotels don't build AI; they buy it. So the misaligned incentive doesn't sit inside the org chart, it sits inside the vendor contract.
Why hospitality is more exposed, not less
Most AI vendors landing in hospitality price on consumption: per token, per call, per query, per "AI minute." That model rewards the vendor when usage rises, regardless of whether a guest noticed. Price wars make it worse, not better — when tokens get cheaper, the rational vendor move is to encourage operators to use more of them.
Ronson points to a GitHub study where coding agents produced 741% more code output but only a 20% increase in actual software releases. That is tokenmaxxing in one statistic: enormous measured activity, modest real-world result. Hospitality's version of that gap would be glossy AI dashboards next to a P&L that quietly bleeds.
Six questions to take onto the HITEC floor
Ronson hands operators a checklist for vendor conversations:
- Does your pricing reward token consumption or outcome delivery?
- Can you show me cost per guest, not just raw usage metrics?
- Do you route routine tasks to cost-effective models, or to billing-maximizing ones?
- Who absorbs the cost when the underlying model or tokenizer changes?
- Can you show evidence of falling per-guest cost alongside improving service?
- Can your outputs map cleanly into USALI accounting standards?
His framing for the answers is sharp: "A vendor who can answer those is selling you a tool — a vendor who can't is selling you a meter."
The metric: Token Cost Per Guest (TCPG)
The proposed antidote is operationally simple: take total AI token spend, divide by guests served, and watch the trend. TCPG converts a vendor-favorable input metric into an operator-owned outcome metric — the same discipline hospitality already applies through energy cost per occupied room or cost per cover. It re-centers the conversation on guest value instead of input volume.
The takeaway for AIHA members heading to HITEC
The mood on the show floor will be confident, and most vendor demos will be impressive. But the questions that matter are not about features; they are about whose interests the pricing model is structured to serve. Walking the floor with TCPG in mind — and with Ronson's six questions in your back pocket — is the cheapest insurance an operator can buy against a generation of consumption-priced AI contracts.
Source: Hospitality Net — Going to #HITEC? Ask about Tokenmaxxing!