The evolving role of AI in business travel

If AI has taught us anything, it is that the pace of change is
accelerating.

If we turn the clock back to
the 2024 Business Travel Show Europe, executives in the sector
told PhocusWire they felt positive about the ability of artificial
intelligence (AI) to improve the traveler experience and enhance
efficiency and productivity behind the scenes.

In the year since, we’ve seen a torrent of AI-related announcements
and launches, both in the wider world and in the business travel sector
specifically.

The world’s first comprehensive AI law—the European Union’s AI Act—was
drawn up and entered into force. The act outlines rules for the use of AI based
on risk levels and bans manipulation of vulnerable groups by AI, social scoring
and biometric identification and facial recognition.

A raft of new models have been launched in the interim too—OpenAI’s
GPT-4.1, o3 and o4-mini, Google’s Gemini 2.0 and 2.5 and Claude Opus and Sonnet
4 from Anthropic. The launch that caused the greatest waves was that of China’s
DeepSeek, which arguably showed low-cost models that did not entirely rely
on the computing power of Nvidia’s GPUs were possible.

It has been a busy year for AI in business travel as well.

A recent Serko/Sabre
survey on AI in corporate travel revealed that 44% of corporate travel
managers believe AI will “have a
significant effect on their programs” over the next five years, while 22%
believe it will be “transformative, reshaping the industry at its core.”

Change is already happening

Last July, Altour announced its AI-powered suite of tools, including booking,
travel disruption management and a natural language interface for travel
managers.

Amex GBT, which launched a dedicated AI initiative in early 2024, has
recently announced AI enhancements to the Egencia platform. It has enhanced the
virtual agent it launched in 2020 with new AI-powered capabilities and has also
added natural language query abilities to its Egencia Analytics Studio as a
beta. The full release is anticipated in 2026, allowing travel managers to
access travel program data by asking questions in plain language.

Meanwhile, in early May, HRS launched its AI-powered
Copilot platform, which combines Anthropic’s large language model (LLM) with
HRS Labs’ own specialized language model to allow travel and procurement
managers to manage, interrogate and optimize their hotel programs. 

HRS has been using AI and machine learning capabilities for many years
to manage and optimize lodging and meetings programs for its clients.

“While many people are
talking about the booking experience, we saw the biggest benefit in the
corporate travel space would be in managing the program,” said Martin Biermann, chief product officer of HRS. 

“We thought about how
we can innovate the entire approach. What’s the point of running this exercise
of program optimization once a year when the world is changing so fast? It is
no longer just savings; it’s sustainability, satisfaction and safety—all take
a role in this.”

Biermann believes that the real power of AI will come from boosting
adoption of the program rather than optimizing the program itself.

While many people are talking about the booking experience, we saw the biggest benefit in the corporate travel space would be in managing the program

Martin Biermann, HRS

“You can optimize your program with AI, but the bigger lever, from a
corporate point of view, is to drive adoption to your program, and for this, you
need to attune your program to what your travelers ultimately need. But it’s a
multidimensional problem; you need a lot of data in order to get it right, and
AI can help us with that.”

The HRS analytical data model looks at different markets, supplier
structures, customer business units, traveling personas, employee profiles,
spending behaviors and loyalty.

The model is optimized in three dimensions: reducing unmanaged spend,
saving time for the procurement or travel manager and maximizing compliance to
sustainability, rate availability rate correctness, price correctness and
supplier compliance.

Corporations are still concerned about AI risks

According to Biermann, data security and privacy pose the biggest questions.

“We made sure that have a model trained in our own infrastructure and our
own cloud environments. We’re not sending data client data anywhere. We provide
transparency so the action log and the reasoning are fully transparent for the corporation,”
he said, adding that this allows them to see—at any time—the actions users have
taken, decisions that were made and which recommendations were accepted and declined.

Keesup Choe, CEO of PredictX recalls that the first questionnaire
on AI was two pages long.

“The latest one is now a book. So, if you’re a vendor
wanting to provide these models for business travel, then the biggest hurdle
now is getting that IT security approval.”

Some corporations know
they want AI but are unsure about how that will play out in reality, according to Gray Dawes Group’s chief technology officer Sophie Taylor.

The travel management company (TMC) holds regular innovation meetings
including senior executives and customers.

“Customers all say, ‘We want AI.’ When you ask them what they mean by AI,
the room goes deadly silent,” Taylor said.

Choe agrees. “The travel managers and end users, for all sorts of
reasons, have not been using AI enough in their own work to be able to guide
vendors on what to build. Vendors, on the other hand, if they’re being honest,
have not been working in the field long enough. So, for both reasons, it’s been
a little bit slow.”

There are also voices being raised in corporations against quicker
adoption of AI.

“Despite this appetite for AI from travel managers and bookers who
believe AI is going to enhance the product and service, their information and
security teams want to be uber-cautious here,” Taylor said.

There are also very public concerns around accuracy and “hallucinated”
answers in particular.

“Even a 1% hallucination rate is unacceptable. There are lots of
strategies to reduce the hallucination rate to virtually zero. We figure it can
go to get to 99.9999%, which is pretty much perfect, but that’s a lot of
work and special expertise required to achieve this,” Choe said.

I think TMCs will exist long into the future because it’s not about when things go right, it’s about when things go wrong

Sophie Taylor, Gray Dawes Group

But, the developments on the horizon are “massive,” according to Choe.

“There’s a new whole new set of models that will be released. I played
with one recently, which is actually different than the current transformer
models,” he said. “Diffusion models, for creating an image for example, kind of
come from the ether and then get sharper and sharper. Instead of generating
things linearly from start to finish, the whole thing comes at once, and that
is going to be transformative.”

He also believes that AI giant Google will make waves.

“I would be shocked, especially with Google’s presence in Google Flights
and so on, if Google doesn’t announce a Gemini agent on your phone. You speak to
it, it knows your calendar, it knows your preferences. All they need to do is add
that last little button to connect to your travel policies and your booking
codes that you’ve that you’ve negotiated with all the providers. It will know
your itinerary. What more do you need really at that point? And then you could
argue, well, why do you need a TMC?”

Despite being bullish on AI, Gray Dawes’ Taylor does not see it as an
existential threat to TMCs.

“I think TMCs will exist long into the future because it’s not about
when things go right, it’s about when things go wrong,” Taylor said.

“We’ve just had the Heathrow shutdown, not to mention COVID, hurricanes
and volcanoes—all these things when you need a TMC where AI doesn’t work.
That’s where TMCs come into their own.”