Ask today’s travel managers whether or not they want AI in their business travel programmes and the answer is a near-universal yes. AI minimises manual tasks, automates reporting, saves time, increases efficiency and allows employees to book personalised, policy-compliant itineraries with ease.
What then are the barriers to adoption? Some are practical. AI tools come with costs that need to be justified, and it takes a dedicated team to keep pace with how quickly the technology is advancing. Training matters too: an AI assistant is only useful if travellers and bookers actually use it, and adoption inside a business takes deliberate effort.
But speak to the people who sign off on these decisions and a deeper hesitation emerges, one that has less to do with time, budgets and expertise and more to do with confidence: how safe is our data, and what stops the AI from making things up? Anyone who has interacted with generative AI (like ChatGPT, Gemini or Claude) has likely watched it invent a fact, a statistic or a source with complete confidence, and a healthy wariness now follows the technology into the workplace.
Generative AI is non-deterministic by nature: ask the same question a hundred times and you can get a hundred variations of the answer. In personal, consumer settings, that’s a quirk. In corporate travel, where policy compliance, expense eligibility and duty of care are on the line, it’s a serious problem. A confident but wrong answer about visa requirements or an out-of-policy hotel recommendation creates work and risk rather than removing it.
“The hesitation we hear from South African companies is almost never about whether AI is quick enough or clever enough,” says Mummy Mafojane, General Manager of FCM South Africa. “It’s about trust and accountability. If an AI assistant gives one of your travellers the wrong answer – that’s your traveller stranded and your problem to fix.”
So, what should trustworthy AI in managed travel actually look like? The emerging consensus points to a handful of markers, and none of them are glamorous.
The first is controlled information. A general-purpose chatbot draws on the open internet, which is precisely why it can be confidently wrong. AI built for managed travel should work differently, drawing its answers from defined, verified sources: the company’s own travel policy, its booking data and its traveller profiles. The answer to “can I fly business class on this route?” should come from the organisation’s policy document, never from a general source, because the question isn’t really whether a traveller can. It’s whether they may.
“Of course, there are different intelligence layers,” says Mafojane, referring to Sam, the AI solution (or intelligent travel companion) built into the FCM Platform, which went live for customers in more than 90 countries in June. “If your traveller wants to know the current temperature in Munich, plus a packing list, Sam draws on general sources. If a travel arranger wants the most up-to-date visa requirements, Sam will draw on specialist sources, and if it’s a question of policy, Sam will draw on pre-approved and pre-loaded documents. The guardrails are in place, and Sam is built with accuracy, security and privacy at its core.”
The second is governance. Travel managers should be able to configure how an AI assistant responds to specific types of queries, so that spend thresholds, approval workflows and supplier preferences are enforced automatically in every interaction. A traveller who isn’t entitled to a premium fare should never be shown one.
“This is where the conversation has matured,” says Mafojane. “Two years ago, companies asked what AI could do. Now they ask how it’s controlled. Can I decide which sources it uses? Will it apply my policy consistently across the board? Those are governance questions, and any provider serious about corporate travel should welcome them.”
The third marker is data privacy, and for Mafojane it’s never been more important.
“Where is company data stored? Is it ever used to train public AI models? Who can see it? In a trustworthy setup, the answers are specific: data is held in a private, secure environment, never fed into public models, with access – and information – restricted to the right people.”
The final marker is the one most often overlooked: knowing when the technology should step aside. Disruption, complex itineraries and emergencies still call for human judgement, and well-designed AI recognises the limits of what it can resolve accurately.
As Mafojane explains, the test is what happens at the point of escalation. If a traveller has to re-explain their situation from scratch to the consultant who picks up, the system has failed them at the moment it mattered most.
“Humans and AI have to work hand in hand. The technology handles the hundredth policy question so our consultants can spend their time on the situations that need experience and care. As an example, when Sam escalates a request to one of our travel experts, the consultant already has the full context in conversation. The traveller never starts over. That’s what builds trust, one interaction at a time.”
For travel managers weighing up AI this year, the practical advice is to interrogate rather than admire. Ask any provider where its AI gets its answers and how accuracy is assured. Ask how your travel policy is enforced within each conversation. Ask where your data lives and whether it ever touches a public model. And ask what happens, in detail, when the AI reaches its limit and a human needs to take over.
Providers with precise answers to those questions are ready for the realities of corporate travel. And that, rather than any dazzling demo, is what earns AI a place in a travel programme: information you can rely on, controls you can prove and people exactly where you need them.
