AI in Corporate Travel: BCD Focuses on the Plumbing, Not the Flash

Skift

6 hours ago

AI might be the future of corporate travel, but for now, it's clearly in pilot mode where the real work is taking place out of view.
Much of the artificial intelligence work in corporate travel isn't happening in slick user interfaces — it's buried in the data infrastructure.
At BCD Travel, the world's second-largest travel management company, AI development is focused on cleaning up messy, fragmented data before it can be turned into anything actionable.
The foundational work in AI at the Netherlands-headquartered company centers on improving data importation, quality, and reconciliation with the aim of delivering better insights to clients.
'There are four key areas we look at when it comes to using AI for data analytics,' said Ajay Singh, BCD's Vice President of Payment and Data Science. 'Data ingestion, improving data quality, enriching data through reconciliation, and finally, delivering insights back to clients.'
.
This supports the backbone of how corporate travel managers make decisions about spend, compliance, traveler behavior, and supplier negotiations.
Each step also reflects a real pain point in the travel data landscape, one that Singh describes as 'highly fragmented' because of information flowing from global distribution systems, travel management companies, online booking tools, card providers, and suppliers.
One clear use case is AI-assisted matching of credit card transaction data with booking records, a task that once required tedious manual review. Another is parsing unstructured data, like comments, PDFs, and voice notes, to pull out meaningful travel insights.
'AI can help us move from reactive reports to proactive recommendations,' he said. 'It's about making the data work for each client, in a way that's relevant to their specific needs.'
Personalization Is Key
As for where AI will have the greatest impact in the next few years, Singh argued it won't be in automation, but in personalization.
In other words, enabling clients to query the data, generate their dashboards, and set predictive alerts based on what matters to them.
'Every client is different. Even within a company, different markets and stakeholders care about different outcomes,' Singh said. 'The goal is to democratize data, to let every user, not just analysts, become a data user.'
Despite the potential of AI, Singh is candid about where the industry stands: 'If this is a seven-inning game, we're still in the first inning.'
One reason? The data isn't ready yet.'AI is like oil; it needs refining. If your data isn't clean, AI can't do much with it,' Singh said.
Amex GBT Helps Agents
Other corporate travel businesses are likewise using AI behind the scenes, and are also experimenting with more visible forms.
As detailed at Skift's recent Data + AI Summit, Amex GBT, the world's largest travel management company, has deployed an AI assistant to help travel agents swiftly surface policy rules tailored to each client.
That's a clear efficiency gain, especially when agents juggle multiple accounts, Marilyn Markham, vice president, AI strategy and automation at Amex GBT, said at the summit.
The company also uses a chatbot to handle frequently asked questions, though it admits that it is only accurate about 60% of the time, and falls back to human agents when things get complicated.
Amadeus Tests Natural Language Questions
Amadeus, meanwhile, is betting on 'agentic AI,' a system of small, task-oriented agents that work together to fulfill goals like booking flights or pulling data, Rodrigo Acuna Agost, head of research and the AI Center of Excellence at Amadeus, said at the conference.
The company has rolled out a voice-to-voice assistant in limited internal use and is testing tools that let users ask natural language questions like 'What are my top 10 customers?' without needing database knowledge.
But even Acuna Agost admitted that a large-scale rollout is years away.
Right now, the AI assistants can answer basic questions, said Acuña Agost. But full end-to-end automation? That's still aspirational.