Why your AI travel agent will be only as good as your worst data
Everyone is racing to deploy agents that book and manage trips. The data underneath decides whether they save you money or just make confident mistakes faster.
The most repeated promise in corporate travel right now is some version of this: soon, an AI agent will plan, book, and manage your business trips end to end. The traveler will state an intention, the agent will handle the rest, and the whole apparatus of searching, comparing, and reconciling will dissolve into the background. It is a good promise. I believe most of it. And I think it is about to collide, hard, with a problem that almost no one in the marketing material wants to talk about.
The problem is data. Specifically, the gap between how good the agents have become and how bad most companies’ data still is.
The number nobody puts on the slide
Let me start with the numbers, because they are stark and they are recent. Across enterprises in 2026, the experimental phase of agentic AI is clearly over: by one widely cited Gartner figure, 40% of enterprise applications are expected to embed task-specific AI agents by the end of 2026, up from under 5% in 2025. Adoption, in the sense of trying agents somewhere, is now nearly universal. And yet the same body of research keeps surfacing a second, far less flattering number. Almost four in five enterprises have adopted AI agents in some form, yet only one in nine runs them in production. The distance between those two figures is the whole story.
Why do so many agent projects stall between pilot and production? The research is unusually consistent on the answer, and it is not that the models are not smart enough. IDC research found that 88% of AI pilots fail to reach production, with failures clustering on governance, data-readiness, and observability gaps rather than model quality. Read that again, because it inverts the intuition most people have. The bottleneck is not the intelligence of the agent. It is the state of the ground the agent is asked to walk on.
The single most-cited barrier makes this concrete. 52% of businesses cite data quality and availability as the biggest barriers to AI adoption, and 37% of organizations face data quality problems for AI readiness. The verdict that accompanies these figures has become something of a refrain in the field, and it is worth stating plainly: agents are only as good as the data they can access, and you cannot build AI agents with poor data infrastructure and expect better outcomes.
I want to bring this out of the abstract and into the specific texture of corporate travel, because this is where I have watched the principle play out, and where the stakes are unusually high.
What an agent actually needs to book a trip
Think about what an AI agent actually needs in order to book a business trip well. It needs to know the company’s travel policy, not as a PDF written in legalese, but as structured rules it can apply at the moment of decision. It needs to know the traveler’s history and preferences. It needs live access to fares across multiple distribution channels, because the cheapest option frequently is not the one a single conventional search surfaces. It needs the corporate agreements, the negotiated rates, the cost-center logic, the approval chains. It needs to know which hotels are in policy in which cities on which dates. And critically, it needs all of this to be connected, consistent, and current, because the agent is going to act on it autonomously, without a human double-checking each step.
Now consider the reality inside most companies. The policy lives in one document. The traveler history lives in the booking tool, or several booking tools. The card transactions live in the bank’s portal. The receipts live in employees’ pockets and phone galleries. The corporate agreements live in spreadsheets maintained by someone in procurement. The approval rules live partly in software and partly in people’s heads. None of these talk to each other cleanly. Every handoff between them is a place where data is lost, duplicated, or contradicted.
A confident mistake is still a mistake
Drop a capable AI agent into that environment and here is what happens. The agent does not save you from the mess. It acts on the mess, confidently and quickly. It books against a policy that was out of date. It misses the negotiated rate because the agreement was in a spreadsheet it could not read. It approves an expense that should have been flagged because the cost-center mapping was wrong. And because the agent is autonomous, these errors do not announce themselves the way a human’s hesitation would. They surface later, at scale, after the damage is done.
This is the part of the agentic future that the hype skips, and it is the part that separates the companies that will get value from agents from the companies that will get a faster, more confident version of their existing problems. The leaders in the field have started saying this out loud. McKinsey’s 2026 data shows the benchmark question has changed: “Are we using AI?” stopped being useful, and “Have we redesigned a workflow around AI and can we prove it?” is the 2026 question. Broad usage is cheap. License counts grow and dashboards light up. Scaled usage is rare, because it requires the unglamorous work that has to happen underneath: connecting the systems, cleaning the data, structuring the rules so a machine can read them.
There is a financial edge to this that should concern anyone running a travel program. IDC predicts a 15% productivity loss by 2027 for companies that fail to establish AI-ready data foundations. That is not a missed opportunity cost. That is a projected decline, the price of layering autonomous systems on top of disconnected data and letting them run. The companies that win will not be the ones with the most agents. By one account, the question is no longer if you will use AI agents, but how you will govern them, and the companies that stay ahead will not have the most agents but the best orchestration, balancing the throughput of AI with strategic human oversight.
This is the conviction that has shaped how we built VOLL, and I will be direct about it because it is the whole point of the argument.
The sequence is the thing
We did not start by building an agent and then go looking for data to feed it. We started from the opposite end. We built a platform where the booking, the policy, the expense, the payment, and the corporate agreements live in one connected environment, captured as the activity happens rather than reconstructed afterward. The agents came after, and they came after on purpose, because an agent is only worth deploying once the ground beneath it is solid. When our agents act, they act on data that is already connected and current, which is the only condition under which autonomous action is safe rather than reckless.
I am not saying this to claim we have solved a problem the rest of the industry has not. I am saying it because the sequence matters, and the sequence is the thing most companies get wrong. They want the agent first, because the agent is the exciting part, the part you can demo. The data foundation is the boring part, the part with no demo. And so the natural temptation is to bolt an agent onto whatever data you happen to have and hope the intelligence of the model compensates for the mess underneath. The research is now clear that it does not. The model cannot out-think bad data. It can only act on it faster.
So if you are a leader looking at the agentic future of travel and wondering how to prepare, I would offer a reframing of the question.
The question worth asking first
The question is not “which agent should we adopt?” That question, asked first, leads you straight into the 88% that never reach production. The better question, the one the data supports, is “is our data ready to be acted on autonomously?” Can an agent see your policy as a rule rather than a document? Can it reach your negotiated rates? Can it reconcile an expense without a human reassembling the context from scratch? If the answer to those is no, then the most valuable thing you can do this year is not buy an agent. It is fix the ground the agent will stand on.
There is a version of the next few years where corporate travel genuinely transforms, where the trip really does manage itself, where the traveler states an intention and the rest dissolves into a well-governed background process. I believe in that version. I am building toward it. But it does not arrive by deploying smarter and smarter agents onto messier and messier data. It arrives the other way around, by doing the patient, unglamorous work of making the data worth acting on, and only then handing it to agents capable of acting.
The agent is the part everyone wants to talk about. The data is the part that decides whether the agent saves you money or just makes expensive mistakes with more confidence than you could have managed on your own. Your AI travel agent will be only as good as your worst data. It is worth knowing, before you deploy one, exactly how bad your worst data is.
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.