This insight summarises Put AI on the P&L, by Terence Ronson, Founder and Managing Director of Pertlink Limited, published on Hospitality Net on 5 October 2026.
Hours saved are not money. After two years of measuring AI by adoption and hours saved, P&L results are thin. A Canary Technologies survey of 404 hotel IT decision-makers found only 37% of AI users reported higher room revenue and only 20% lower operating costs. Hyatt's Pat Nestor asked the question boards will ask: "Where does that live on a P&L sheet?" Hours saved become money in only three ways: buying fewer hours, selling more, or avoiding cost.
A proof point. Wyndham's AI voice concierge, built on Salesforce Agentforce, targets a real leak: 25% of inbound hotel calls were being dropped before AI. Now 85% of callers stay with the AI agent without asking for a human, AI-handled calls convert 15% better, and calls transferred to a live agent carry 16% higher ADR.
The P&L landing test. Before approving or renewing any AI tool, answer five questions:
- Which P&L line moves? Name it. "Efficiency" is not a line.
- Which conversion applies? Fewer hours bought, more sold, or cost avoided.
- What is the baseline? Measure it before go-live, over enough time to mean something.
- Who owns the number? A department head, not the vendor or IT.
- What is the full cost? Licence, tokens and human review time.
Then set a check date and a scale, hold or stop decision.
Oversight is the hidden cost. Human review is a third cost column most budgets miss. The AI override rate, how often people overrule the machine, is both a cost and a quality signal.
A 90-day plan. In the first 30 days, list every AI tool, apply the five questions and find revenue leakage. By day 60, set baselines, assign owners and cost the oversight. At day 90, run the first scale, hold or stop review, stopping at least one tool, then report to the board quarterly.
Why this matters for hotels. Ronson is not anti-AI. He is against "unaccounted AI: spending justified by enthusiasm, renewed by habit, and defended by anecdote." The tools that survive this test are the ones worth scaling.