A travel agency should price AI by the cost of the outcome, not the unit: Mohsen Ghasempour warned at Travel Weekly's Future of Travel Conference that travel businesses risk a surge in operational costs as AI providers move to unit-based pricing. For an agency, the outcomes to count are the accepted quote and the confirmed booking, not the message.
Key facts
- More than 238,000 message-and-reply pairs could be rebuilt from the record of the desks I ran (the author's operations archive).
- AI providers are moving from subsidised models to unit-based pricing, which risks a surge in operational costs (Mohsen Ghasempour, via Travel Weekly UK).
- Providers charge separately for what goes in and what comes out, and background tasks add hidden metering of their own (Travel Weekly UK).
The news. Speaking on an AI panel at Travel Weekly's Future of Travel Conference on 10 September, Mohsen Ghasempour, chief AI officer at B&Q and Screwfix parent Kingfisher and non-executive director at Hays Travel, warned that travel businesses risk a surge in operational costs as AI providers move from subsidised models to unit-based pricing. The true cost of deploying AI, he said, is about to show as limits on compute bite, and businesses are deploying generative AI without calculating the unit costs underneath. The article explains the meter: every interaction is broken into units, providers charge separately for what goes in and what comes out, and background tasks add hidden metering of their own. Ghasempour noted that the big AI providers have heavily subsidised usage so far, and that limited server compute will force usage prices up. He urged travel firms not to deploy AI on low-value tasks such as generic email summarisation, where the operational cost outweighs any productivity gain. As he put it: "It is not about the cost of the unit of AI power; it’s about the cost of the outcome." Deloitte partner Andy Gauld advised firms to set clear governance frameworks before experimenting across multiple AI platforms, and warned that without strict guardrails on how staff use AI, costs will spiral out of control.
Why is most of a desk's traffic not the answer?
More than 238,000 message-and-reply pairs could be rebuilt from the record of the desks I ran. That is the traffic a per-unit meter would count: every question, every chase, every correction. The client acts on a few of them: the price, the confirmation.
Run AI on every message and the meter runs on every one of them, whether or not it carries the answer the client acts on. Ghasempour's example was generic email summarisation. A desk has its own version: AI running across the whole conversation, when what the client acts on is the price.
So before worrying about the price of the unit, cut the number of messages the meter runs on.
What outcome should an agency count?
Ghasempour's line is the right test for an agency, because a desk already knows its outcomes. A quote the client accepts. A booking that is confirmed. An agency earns on the booking, not on the message.
Those are the units to price AI by: its cost per accepted quote and per confirmed booking, not per message sent or summarised. If AI prices rise, as he expects, that figure still tells you which uses to keep.
Gauld's governance point belongs next to it. Decide where AI runs, and what it is measured against, before each team starts trying its own platform.
Why does it matter for the agency?
On Ghasempour's reading, subsidised pricing is ending and prices will rise. An agency that watches the unit price will see a bigger bill and not know which part of it was worth paying for. One that counts the cost per accepted quote and per confirmed booking will know. Stop running AI on every message. Put it on the work that ends in something you can count.
Source: Travel Weekly (UK) — Tech experts tell trade to stop deploying AI on low-value tasks.