Across the travel industry, large consultancies are helping clients roll out AI agents. Everything from customer service bots to booking assistants and operational copilots is now on the table.
There’s plenty of excitement, and rightly so. Done properly, the impact should be significant.
But the reality isn’t quite matching the expectation.
Many of these agents are underperforming. Some fail outright. Others appear to work, but quietly return the wrong answers.
So what’s going wrong?
It’s not the AI itself. The models are strong. They’re fast, capable, and improving all the time. The real issue sits elsewhere. It’s the data feeding them.
A consultant we recently spoke to described it quite well. It’s like going to the gym every day but living on junk food. The effort is there, but the inputs are working against you.
Travel data, in particular, is difficult to manage:
- Inventory updates constantly
- Pricing moves all the time
- Policies differ by supplier, geography and context
- Legacy systems hold fragmented and inconsistent records
- Pricing comes from multiple sources including GDS, OTAs, metasearch and brand sites
When agents rely on stale batch data, poorly structured APIs, or inconsistent schemas, the result isn’t just minor errors. They misinterpret, hallucinate, and lose credibility.
Once trust goes, it’s hard to recover. Agents get pulled, rebuilt, or worse, left running while quietly giving incorrect responses.
Why Traditional Data Pipelines Fall Short
Most enterprise travel systems weren’t built with AI agents in mind. They were designed for human users, dashboards, and scheduled reporting.
That creates three clear problems:
- Latency – the data is already out of date
- Ambiguity – fields lack clear meaning
- Friction – too much processing is needed before the data is usable
For AI agents, these aren’t minor issues. They’re deal-breakers.
The Move Towards Agent-Ready Data
Consultancies are now shifting focus to what’s being called agent-ready data.
In simple terms, it’s data designed specifically for machines to use in real time.
It has three key qualities:
- On-demand – fetched exactly when needed, not preloaded in batches
- Semantically structured – fields are clearly defined with embedded business logic
- Machine-ready – no cleaning or interpretation required
The data flows straight into the agent without needing extra work.
Why It Matters
As more travel companies scale AI agents, this becomes a business issue, not just a technical one:
- Incorrect bookings damage trust
- Mispricing affects revenue
- Policy mistakes create compliance risk
At this point, it’s less about how good the model is, and more about whether it’s making the right decisions.
A Practical Example
One consultancy working in car rental shared results from a recent project where they shifted a client onto agent-ready data.
They didn’t change the AI model. Only the data layer.
The results were clear:
- Task success rate improved from 62% to 91%
- Error and hallucination rates dropped from 28% to 6%
That’s a material difference. Not just in performance, but in confidence and usability.
Final Thought
The travel industry doesn’t have an AI problem.
It has a data problem.
And as adoption grows, the advantage won’t sit with those using the most advanced models. It will sit with those who have the data infrastructure to support them properly.
