Most of the AI conversation in travel is about whether a machine can plan a trip. My read is that the argument is happening in the wrong place — the task worth handing over first is the one nobody is worried about losing.
Ask a travel advisor what AI threatens and you get the same answer: the recommendation. In a TravelAge West survey of more than 200 advisors, 45% said they believe AI-powered OTAs or AI-powered travel agents could impact the need for travel advisors in the future, up from 38% in 2023. That is where the attention sits — on the part of the job an advisor is proudest of.
Itinerary tools versus inbound questions
Now look at where advisors are actually pointing AI today. The same survey found most advisors use AI for itinerary creation and trip proposals, followed closely by website and social media creation, then marketing and lead generation. My read: that is the tool aimed squarely at the two things that need you most — the recommendation you get judged on, and the voice clients chose you for.
The task with the clearest return is the boring one in between, and vacation rental management has measured it before travel agencies have. Evolve, a US vacation rental manager, deployed AI on routine guest support in late 2025 and told Skift it is now deflecting over 60% of those conversations without human intervention. That is an inbound-question operation — the same handful of questions, asked constantly, by people who have not told you anything about themselves yet. So is yours.
Repeatable questions versus judgment calls
The sorting rule that follows is not glamorous, and it holds. A quoted fare, a deposit deadline, a visa requirement, a recommendation you will be judged on — none of that is work where being probably right is good enough, so none of it gets handed over unsupervised. "What's included?", "do you do Japan?", "what's the deposit process?" at 11pm on a Sunday is a different category entirely. The line isn't machine versus human — it's repeatable versus particular.
And start narrower than feels worth doing. Erik Zahnlecker, agent product manager at Sierra, gave Skift the version a small team can use: begin with a use case that is high-volume but contained — reservation confirmations, cancellations, modifications — and start with one channel or a small percentage of traffic before expanding.
The line isn't machine versus human — it's repeatable versus particular.
Tally your last 50 inbound messages
Open your inbox, WhatsApp and DMs, scroll back 50 first messages from new or existing clients, and tag each one with the question it was really asking. You will almost certainly find five or six questions covering most of them — price ranges, what's included, deposit and payment terms, passport and visa basics, availability for dates, "can you do X destination". That list is your handover candidate, in priority order, and it took twenty minutes to find. Everything not on it is the work you're actually paid for.
This is roughly the sequence we ran with a growing travel agency: hand the machine the first contact and the qualifying, keep the advisor on the recommendation — the layer we build as Attensia (https://attensia.ai). Published result on our site: 5× monthly sales — 8 to 40+ in 90 days. The case is here if you want the detail: https://brandrstudio.com/our-work/travel-agency-ai-sales See the case →
Get this every week. One issue, one idea, sources you can check. No pitch.