AI has moved from the edge of the travel program to the centre of the conversation. In its 2026 Business Travel Innovation Research, GBTA found that the majority of buyers want AI to manage disruption, support travelers and simplify booking. The appetite is clear. What has not quite caught up is a consistent way to measure its success, and that, not the technology, is the interesting question right now.
That gap matters, because the pressure is coming from above. “What’s our AI strategy in the travel program?” is a fair question, but it starts from the technology and works toward a use. The programs getting value do the opposite: they start from an outcome and make the technology earn its place.
Look at where AI is actually being applied in travel programs today and a pattern appears. Common uses include customer service, traveler personalization, and itinerary planning, all worthwhile and all hard to put a number against. Meanwhile some of the concerns buyers raise are data privacy and integration. That is the shape of a market experimenting with capability faster than it is defining value. Activity is easy to show. Outcomes are not.
So before any AI capability goes near your program, four questions separate the useful from the decorative.
- What specifically does it do? Not the category, the task. “Rebook a fare when the same fare on the same conditions drops below a threshold, inside policy” is an answer. “Enhance the traveler experience” is not.
- Does it produce an outcome your CFO would recognize? The output has to be a number that lands in finance reporting: spend reduced, leakage closed. If the benefit cannot be counted, it cannot be defended.
- Can it commit to that outcome against a baseline? Pick the number, pick the comparison period, and put the capability on the hook to hit it, in your reporting, by that date. Vague upside is not a commitment.
- Will anyone have to notice it? Ideally not. The best AI in a program is invisible by design: it changes the result without adding a step for the traveler or a task for your team. If people have to think about it, it is spending attention you will not get back.
The questions are deliberately strict. That is the point. They filter a crowded market down to the few capabilities that actually move what your program reports each month.
So when the AI-strategy question gets asked, there is a better one to ask yourself: what does AI need to produce in our program for it to be worth running?
Pick the number first, then make every option earn it. The conversation stops being about which tools are exciting and starts being about which ones deliver.
This reframe does two things. Upward, it turns a vague mandate into a defensible position: you are not resisting AI, you are holding it to the same standard as every other line in the budget. Outward, it changes what you accept. A vendor who can commit to a result is offering you one. A vendor who can only name a category is offering you the category.
The programs that come out ahead over the next few years will not be the ones that adopted AI first, or talked about it most. They will be the ones that decided what it had to deliver, however it reached their program, and then checked that it did. Quiet under the hood; loud in the line items. Start with the number you would need to see. Everything else is a demo.