The industry is debating the wrong question
Walk into any corporate travel conference this year and the argument you will hear is about autonomy. Should an AI agent be allowed to rebook a flight on its own? Change a hotel? Approve an expense under a certain threshold? Entire panels are built around where to draw that line.
It is a reasonable debate. It is also premature by several years for most of the companies having it, and a survey published this year makes that uncomfortably clear.
GBTA research conducted in March 2026 with 269 travel buyers across the United States, Canada and Europe found that only 12 percent of travel programs have a consolidated view of their own program data from a single source. The same study found that 58 percent of buyers say AI has had little or no impact on their programs so far.
Those two numbers are not independent findings. The second is a consequence of the first.
Corporate travel data consolidation is not a prerequisite the industry has been slow to address. It is the thing the industry has skipped entirely while moving on to a more interesting conversation about agents.
What the same survey wants
The gap becomes stark when you look at what those buyers say they want from AI.
In the same GBTA study, 92 percent expressed interest in predictive analytics for spend forecasting.
Another 89 percent wanted automated disruption management and rebooking. Eighty five percent wanted AI-powered traveler support, and 83 percent wanted conversational booking.
Every one of those capabilities depends on the system having a reliable, current, unified picture of the program. Spend forecasting on fragmented data produces a forecast of the fragments. Automated rebooking without knowing a traveler’s actual itinerary, policy tier and preferences produces a rebooking somebody has to fix manually.
The study also found 63 percent citing lack of consolidated reporting as a specific pain point, and 61 percent saying managing travel across regions is a challenge. Buyers are not unaware of the problem. They are describing it accurately and then asking for capabilities that require it to be solved first.
Where the foundation actually cracks
If the data foundation is thin, you would expect the tools built on it to work well in the simple case and fail in the complicated one. That is precisely what a second study found.
Christopherson Business Travel surveyed 123 US travel management, procurement, operations and finance professionals for a white paper called The Booking Breakpoint. Only 20 percent said their booking tool supports their company’s travel needs very effectively. Just 11 percent said more than three quarters of trip modifications get handled successfully through self service. And during unplanned disruption, the moment a travel program most needs to work, only 10 percent said their tools support travelers very effectively.
Read those three numbers in order and they describe a specific failure shape. The tool works at the point of purchase and degrades from there. Booking is a single transaction against clean inputs. A modification requires knowing what was already booked, what policy allows, what the traveler is entitled to, and what alternatives exist. Disruption requires all of that at once, under time pressure, across multiple suppliers.
Christopherson’s CEO Josh Cameron framed the underlying issue in a way I think is exactly right: travelers experience one connected trip, while the systems serving them treat air, hotel, car and support as separate transactions. The traveler’s experience is continuous. The data is not.
The loop this creates
Here is where the two studies stop being separate observations and start describing a cycle.
When a tool fails a traveler during a modification or a disruption, the traveler does not file a complaint and wait. They solve their own problem, using whatever channel works.
The Amadeus Tailored Horizons survey put a number on how widespread that has become: 62 percent of business travelers already use their own AI tools outside the channel their company provides.
Every one of those trips leaves the managed channel. Which means the program loses the data from it. Which makes the consolidated view thinner. Which makes the tools worse at handling the next modification. Which sends more travelers outside the channel.
The industry calls the result a shadow travel stack, and usually frames it as a compliance problem. It is worse than that. It is a self-reinforcing loop where each rotation degrades the thing everyone says they want to build AI on top of.
The part nobody connects
There is a second consequence of travelers leaving the managed channel, and it has nothing to do with data quality.
I wrote recently about a breach at a group of UK airports in which the records of roughly 8.7 million customers were accessed. The detail that mattered was where the data came from: not a booking tool, not a travel management company, but airport WiFi sign-ups, car park reservations, lounge access and fast track purchases. Every one of those is a transaction a traveler makes personally, outside any system their employer contracted or reviewed.
So a traveler who goes around the managed channel is doing two things simultaneously. They are removing a record from the program’s data foundation, and they are creating a record somewhere the company has no contract, no retention schedule, and no visibility.
Data quality and data security are usually owned by different people, budgeted separately, and discussed in different meetings. The shadow stack is one problem producing both, and it is not being managed as one problem anywhere I have seen.
What doing this properly looks like
There is a useful counterexample, and it is worth studying precisely because it is unglamorous.
When Business Travel News named Moderna’s Jennifer Steinke its 2026 Travel Manager of the Year for building an AI system her travelers use daily, the coverage focused on the AI. The more instructive part of the story sits earlier in the timeline. Before any vendor was involved, in October 2024, she manually uploaded a year and a half of program data into spreadsheets. Only after that foundation existed did she bring in a data specialist to normalize it and feed it back through an API.
She made herself part of the 12 percent, deliberately, by hand, before the interesting part started.
Evan Konwiser of Amex GBT made a related point in a Forbes interview with Jeff Fromm: the breakthrough interface gets the attention, but the durable value usually sits in the plumbing underneath. Proprietary context, clean data, and the infrastructure to act on both in real time.
There is also a security argument for getting the foundation right rather than layering agents onto a fragmented one. Researchers from Stanford and Georgia Tech showed in work presented at ACL in 2025 that AI agents completing web-based tasks, including travel booking, can be manipulated by content designed to fool a machine rather than a person. An agent operating on incomplete context has fewer internal checks against being wrong, and fewer still against being deceived.
The commercial reading of all this
I want to note something about the three studies I have cited, since it affects how they should be read.
The GBTA research was produced in partnership with a booking platform, a hotel group and a travel management company. The Christopherson white paper concludes that booking technology and service need to work together, and Christopherson is a travel management company that sells service alongside technology. Each study’s conclusion lands comfortably on its sponsor’s product category.
None of that makes the underlying data wrong. Vendor-funded research is a large share of what this industry knows about itself, and refusing to read it would leave us knowing less. But it does mean the useful move is to take the numbers and set aside the conclusions.
I should apply the same standard to myself. I cofounded VOLL, a corporate travel and expense management platform, so I have a direct commercial interest in companies concluding that their data foundation needs work. I have written before about how the real cost of a travel program hides in the places that never reach an invoice, and this is the largest example of that I know.
The regional gap in all of it
One last observation about all three studies.
The GBTA sample covers the United States, Canada and Europe. The Christopherson sample is entirely American. The Amadeus sample is American. Latin America appears in none of them.
That means the consolidation figure for this region is simply unknown. It could be better than 12 percent, and there is a reasonable case that it might be: Latin America has repeatedly adopted new travel and payment technology faster than the markets that invented it, often skipping intermediate stages other markets are still working through.
Newer systems tend to be less fragmented than older ones, because they were built after consolidation became an obvious requirement rather than before.
But that is an argument, not a measurement. The most useful thing I took from three days at a global industry convention this year was how much of what the industry treats as settled turns out to be a US finding that traveled. Somebody should measure this one directly, here, before the region inherits a diagnosis that was never about it.
About me
I am an entrepreneur with over 20 years of experience at the intersection of tourism and technology. I am co-founder and Chief Business Officer of VOLL, the largest mobile-first corporate travel and expense management platform in Latin America, and a recognized reference in the development of the corporate travel industry.
A Marketing specialist from Fundação Dom Cabral, I serve on the Tourism Council of FecomércioSP and on the Executive Council of the Latin American Association of Corporate Events and Travel Management (Alagev). A frequent traveler and close observer of human behavior in motion, I write and speak about innovation, digital transformation, entrepreneurial leadership, and the future of corporate travel.