Gee Mann is the founder and CEO of TravlrID
In the past 12 months, virtually every significant platform in travel has announced an AI agent. Airlines are building conversational booking assistants. OTAs are shipping personalisation engines trained on billions of search signals. Hotel chains are deploying AI concierges. Corporate travel platforms are rolling out autonomous agents that can search, book, rebook, expense and report – all in natural language, all without human intervention.
The investment is real. The ambition is genuine. And almost all of it is built on a data foundation so broken that the agents will systematically underperform, frustrate the travellers they were designed to help, and quietly recreate – at far greater scale and with far higher stakes – the exact fragmentation problem the industry has been trying to solve for 30 years.
The travel industry does not have an AI problem. It has a memory problem. And nobody is talking about it.
_____________________________________________________________________
This is a two-part opinion piece by Gee Mann, the founder and CEO of Travlr ID, a travel profile infrastructure company. Part two contains what some readers might consider commercial language, which is disallowed in BTN’s policies for opinion contributions. The BTN editorial team has cut a considerable portion of this section. It has determined the remainder is illustrative and the topic is of critical importance to the industry.
______________________________________________________________________
What an AI agent actually needs to work
An AI agent is only as useful as its context. Strip away the natural language interface and what you have is a recommendation and execution engine – one whose output quality depends entirely on how well it understands the person it is serving.
For a travel AI agent to perform well, it needs to know who the traveller is right now. Not who they were when they last updated their profile. Not a probabilistic guess based on aggregate behavioural patterns. Right now: their current loyalty status across carriers and hotel chains, their valid travel documents and visa entitlements, their live policy parameters, their stated and revealed preferences, their upcoming trip context, their cost centre as of this month – not last quarter.
That data exists. The problem is that it lives in 17 different places, none of which agree with each other.
The average managed traveller has some version of their identity held by their employer’s HR system, their online booking tool, their GDS profile, their expense platform, their travel management company, their airline frequent flyer programme, their hotel loyalty account, their car rental preference and whatever risk management platform their company uses to track duty of care obligations. Each silo maintains its own record. None of them sync in real time.
A traveller who changes departments, acquires a new passport, earns top-tier status on a new carrier, or simply updates a seat preference faces a cascade of manual updates across systems that may or may not actually propagate.
We tested what happens when you give an AI agent access to a unified, accurate profile versus forcing it to work from fragmented data. API calls dropped between 60 and 80 per cent when agents had real profile context. Not because the model improved. Because it finally knew who it was helping. The noise went away. The recommendations became relevant. The look-to-book ratio collapsed.
The inverse is also true. An agent working from stale, incomplete or conflicting profile data will hallucinate preferences, surface out-of-policy options, fail to apply entitlements the traveller has earned, and produce results that feel generic – because they are. The traveller learns quickly that the AI doesn’t really know them, loses trust, and routes around it. And the industry wonders why adoption rates disappoint.
The fragmentation crisis the industry keeps rediscovering
Profile fragmentation is not new. It has been an operational headache in managed travel for decades, and a frustration for leisure travellers for nearly as long. What is new is that the consequences of getting it wrong have just become existential rather than merely inconvenient.
Consider what happened during the recent escalation of conflict in the Middle East. Companies with large corporate travel programmes – global enterprises with sophisticated risk management systems and dedicated travel managers – could not locate their employees. Not because they lacked tracking technology. Because the data feeding that technology was incomplete: bookings made directly with suppliers that never reached the TMC, hotel stays not captured in the itinerary, travellers who booked off-channel because a direct rate was better, itinerary changes made at the gate that nothing recorded. When the crisis hit, organisations were in “scramble mode” – finding out an employee was in a danger zone only because that employee called to say so.
This is the same fragmentation problem, expressed in its most consequential form. The AI agent that cannot find a traveller’s loyalty number and the risk platform that cannot locate an employee in a crisis are downstream symptoms of the same upstream failure: There is no single, trusted, current record of who this person is and where they are in the world.
The irony is that AI, which was supposed to solve these problems, will accelerate them if the underlying data infrastructure does not change. Every agent that produces a bad recommendation because it lacked profile context trains the traveller to distrust the agent. Every duty-of-care failure that traces back to incomplete data is a liability that no amount of interface sophistication can offset.
The walled garden trap
The industry’s response to the data problem has been, predictably, proprietary. Every platform is solving fragmentation for itself, in isolation, in a way that makes the aggregate problem worse.
OTAs are deepening loyalty programmes to keep travellers inside their ecosystem, where their AI can learn from repeat behaviour. Airlines are investing heavily in first-party data to reclaim the customer relationship they ceded to intermediaries over two decades. Hotel chains are pushing hard on direct booking for the same reason. Corporate booking platforms are training AI models on proprietary spend data and positioning that as a competitive moat.
Each of these is a rational move for the platform making it. Collectively, they are producing a more fragmented, more locked-in, more opaque data landscape than the one they inherited.
The data governance practices running underneath this are starting to surface as a serious issue. Several major platforms now opt corporate clients into AI training on traveller data by default, burying the opt-out in settings that many customers do not know exist. At least one major platform has drafted terms of service granting itself irrevocable rights to use customer inputs and outputs to train and improve its AI systems – with no sunset clause and no compensation. These are not edge cases or startup oversights. They are deliberate architecture choices by large, well-resourced companies.
The individual traveller – the person whose preferences, biometrics, movement history, visa records and behavioural data are at the centre of all of this – has almost no visibility into what is being collected, almost no ability to correct errors across systems, and almost no leverage to take their data with them when they change platforms. The question “who owns the traveller?” used to be a distribution argument.
In the AI era, it is a data sovereignty argument. And the answer being written into terms of service right now is: not the traveller.
______________________________________________________________
Read on for part two of Gee Mann's opinion piece here.