The future of AI in business travel depends on human expertise

We trust Artificial Intelligence (AI) to navigate, but we validate the route. We’ll follow what it gives us, but we keep watching the road, and if the car ends up somewhere wrong, that is on us, not the app.

Even though GPS could be one of the most successful pieces of technology ever built, we still treat it as a suggestion.

We trust it because it’s right almost every time. But every so often, it sends you with total confidence into a construction site or towards a road with so many potholes it’s undriveable (I live in New Orleans). Which is exactly why your eyes stay on the road. You are still the key to arriving safely at your destination.

In the hype around integrating AI into all aspects of business travel management, a lot of our industry lost that instinct. When it comes to their travel programs, I’m seeing plenty of companies talking about handing AI the keys.

So far, the results for letting the AI take the wheel aren’t very encouraging. MIT’s Project NANDA looked hard at this last year, and in a report called The GenAI Divide, found that only 5 percent of enterprise generative-AI initiatives are bringing real financial value. The other 95 percent? No measurable impact on the bottom line. That is tens of billions of dollars spent to move essentially nothing.

Don’t get me wrong, I’m not making a case against AI. But I am pointing out what’s missing in many companies’ AI strategies. MIT was clear on the cause: even though they’re using excellent models, they still fail on what the researchers call the “learning gap,” which is really just the human work of wiring AI into how a team operates, decides, and answers for its choices. The decision layer isn’t there. And there’s a second culprit every travel manager will recognize on sight: the data. Gartner expects companies to walk away from 60 percent of their AI projects through 2026 because they lack AI-ready data, and about two-thirds of them already admit they’re not confident their data is good enough.

I’ve spent most of my career pulling messy, disconnected data into something a team can actually act on. So both of those failure modes, the missing decision layer and the less-than-stellar data, are personal to me. They’re also what shaped how I think AI should be used in the travel industry. Let it navigate, but keep a person at the wheel.

1. Build AI you can trust.

Trust doesn’t only come from data. When AI can be confident, fluent, and flat-out wrong all at once, the question becomes “who’s vouching for that information?” The legal world is finding this out the hard way. In Mata v. Avianca, a couple of lawyers filed a brief full of cases that ChatGPT had simply invented. The court sanctioned them, and there’ve been more sanctions like it since. The takeaway isn’t that the tool is worthless. It is that somebody qualified has to sit between what the AI produces and what you decide, and own it when it’s wrong.

The market already feels this. In a 2025, Harvard Business Review Analytic Services survey, just six percent of companies said they fully trust AI agents to run core processes on their own. But here’s the part that gets misunderstood: adding a human doesn’t automatically make AI better. MIT ran a huge study on this and found that for many tasks, human plus AI teams actually performed worse. The human becomes essential by bringing something that the model can’t provide. You see it in medicine, when radiologists read a scan with AI support, they find more cancers than they do without. One study put it at 18 percent. And they still beat the AI on the toughest-to-read scans. Each covers the other’s blind spot. That is the model we have to build from in business travel.

The foundation for building AI you can trust (but verify) is strong data and processes. At Advito, we train our AI on robust, proprietary data sets. We’ve also invested in building best in class ingestion and consolidation processes. Disconnected systems and data sets are a huge barrier to AI success. The underlying architecture is equally as important as the AI itself, and the human expert that sits between that output and your decision.

2. AI that acts on its own.

Generating information is the easy part now. The complex part – the part that still needs context, judgment, and experience – is turning that information into a decision leadership will actually back. This is where most AI runs out of road. The tools are getting genuinely good, and that’s not the problem. It’s the platform that surfaces a signal, like an increase in program leakage or an odd spend pattern, and then hands the travel manager the genuinely hard question: now what? I’ve sat across from clients staring at a screen full of perfectly good analysis, asking me, essentially, “okay, what do I do with this on Monday?” The value is in making a decision, knowing whether the move is a renegotiation, a push for adoption, or a policy change, and selling that to the people holding the budget. An insight nobody acts on isn’t worth much.

3. Accountable execution and measurable outcomes.

There is a lot of temptation in many companies to integrate as much AI as possible into every workflow, tool and process – often completely eliminating the human who used to do that work. And as several Fortune 500 companies have learned the hard way, that doesn’t usually pan out. Ford learned this publicly earlier this year, when they deployed AI-powered quality checks and hundreds of AI cameras in their system but later admitted that they “mistakenly assumed” that simply feeding design requirements into AI would be enough to produce high-quality vehicles. The tools missed defects that experienced engineers would have caught, many of whom had already left the company. Ford’s response was to bring back roughly 350 “grey beard” engineers to retrain the AI, mentor young staff, and restore quality controls. Ford realized that AI was not enough, and the better solution was to put expert judgement back into the loop and make people accountable for the outcome. In the end, you won’t remember how brilliant the AI recommendation was. You’ll remember what changed for your business travel program, what kind of savings you achieved, how travelers felt, and the results someone was willing to attach their name to.

For a lot of business travel programs, AI just gets bolted onto whatever workflow is already in place. But if that workflow was broken to begin with, automating it only gets you to the wrong answer faster. The more valuable move is to rethink how the work should be done first, and then turn AI loose inside a better process. Automation makes things faster. Reconfiguration makes them better. And the difference between the two is almost always a person who knows the program well enough to take it apart and put it back together.

None of this makes me a skeptic, for the record. AI is already reshaping how our own consultants work, swallowing analysis that used to eat days, freeing our experts for the questions clients genuinely need answered. And the trend line isn’t ambiguous. Phocuswright’s 2026 outlook shows 61 percent of travel companies scaling up their AI investment, and 39 percent of US travelers already leaning on AI to plan trips, up from 28 percent just a year earlier. This is coming, ready or not. The choice was never AI versus human expertise. It is whether you pair the two, or hand the software the wheel and hope.

Managed travel is a difficult road to try to drive hands-free. It runs across every region, every supplier, every traveler, every booking channel, all at once, and the parts that are hardest to automate are usually right where the money and the risk are hiding. So, when you size up anyone’s AI, ours included, run the same gut check you’d run before letting a car drive you around on its own. Who owns this output? Who catches it when it’s wrong? Who’s still at the wheel when it’s time to actually do something? If the answer is just “the AI,” I’d be careful. If the technology is paired with a real person who can walk you through the reasoning and stand behind the result, then you’ve got something you can trust to help optimize your program.