How travel companies are approaching agentic AI

As travel brands test and deploy agentic artificial intelligence (AI), they are defining where the technology fits best, from internal reporting and workflow automation to customer service and disruption management.

Some are encouraging broad experimentation across teams. Others are taking a more controlled approach. 

But as deployment continues, many companies are drawing firmer boundaries around where AI agents can act independently—and where human oversight remains essential.

Travel companies need to deploy AI agents or risk being left behind, said Accor’s Amro Khoudeir, senior VP of digital marketing for the premium, midscale and economy division, MEA and Asia Pacific.

“I'm highly encouraging my team to step away from doing anything they can automate,” he told Phocuswire.

Since last year, this mindset has been embedded not only in how the company operates but in its incentive structures. 

“One of our unified bonus objectives across the team was for each person to identify and fully offload at least one end-to-end process to AI. Not just use AI as an assistant—but to own a workflow, redesign it around AI and remove themselves from the repetitive parts of it entirely,” he said. 

“Tying it to performance objectives was deliberate. It signals that this isn't optional experimentation—it's a core professional expectation.”

Khoudeir primarily uses Anthropic’s Claude for its “strong reasoning and context retention,” but he is not restricted to a single platform. 

“I use what performs best for a given task, and I've held paid pro subscriptions with at least two platforms simultaneously for a couple of years. Most recently, I've been experimenting with OpenClaw.”

When it comes to building an agent, he invests time upfront to define the role, the constraints, the format, the audience and the expected output. 

“For recurring workflows, I've essentially built prompt templates that encode institutional knowledge—brand standards, market nuances, strategic priorities. That's where real leverage comes from: not one-off queries, but repeatable, scalable workflows,” he said.

Khoudeir cites two recent use cases. 

“I built a crisis keyword monitoring system that tracks travel demand disruption and recovery signals across 11 destination markets using Google Trends data, producing structured outputs my team can deploy directly,” he said. 

“Separately, I built a share-of-voice reporting workflow that aggregates and structures competitive visibility data across our brand portfolio—work that would previously have required significant manual effort and analyst time.”

Fliggy is also embracing AI agents built using parent Alibaba's Qwen model family. 

Building agents works best for tasks that involve multiple steps, require frequent calls to external tools, or demand dynamic decision-making, said spokesperson Chen Zhang.

“These are the kinds of tasks that a single prompt simply can't handle well. On the other hand, for tasks with relatively straightforward logic and low latency requirements, the cost-benefit of building an agent often doesn't quite hold up,” she said.

Chen said, “We encourage every employee—engineers, product managers, designers, business development managers, HR managers—to use AI as much as possible in their daily work. People tend to surprise themselves with what they can build. The barrier to entry is much lower than most people expect.”

To encourage adoption, Fliggy provides employees with free resources, course materials, frequent workshops, lectures and hackathons. 

John Lyotier, CEO of TravelAI, said the company has built its own agentic AI designed specifically for travel, but uses other platforms too.  

“Our team is constantly experimenting with Anthropic's Claude, OpenAI's Codex and the latest models to build custom agents to optimize their jobs or workflows,” he said. “Anything that is a repeatable task is one that you should be able to create an AI agent to do.”

“Everyone, top to bottom, is a tinkerer at heart and willing to share their findings with others. We had vibe-coding training internally recently where our HR and finance teams were learning from our product managers on how to build apps to automate job functions. So much innovation comes from this 'play' that can then be in turn used in optimizing the outcomes for travelers.”

According to Lyotier, companies wanting to build their own agents need to think carefully about prompting.

“We are probably getting cavalier in our prompting,” he said. “Relying on deeper organizational memory and workflows that pull from these memories does make a big difference in context adherence and thus the quality of the output.”

Where AI agents need guardrails

Travel technology company Amadeus says it is embracing the technology, but with safeguards, taking an “enterprise approach to AI that prioritizes responsible use, evaluation and compliance, so AI can be deployed safely and integrated into real workflows.”

Gaelle Bristiel, SVP of engineering at Amadeus, said, “Our focus is less on any single model and more on how we connect AI to trusted, dynamic travel data and deep industry business logic, supported by robust evaluation and our responsible AI framework.”

Bristiel said the best processes for AI agents are “repeatable tasks with clear rules, good data sources and a well-defined end outcome.”

She said, “In practice, these are often well-scoped, specialized ‘micro-agents’ that handle defined tasks such as drafting and summarizing, knowledge retrieval (for example, Q&As based on approved content), ticket/incident triage and creating first-pass resolution notes, before a person reviews and finalizes.”

A typical agent might support domain workflows like search/shopping, servicing and disruption management, pulling the right context, checking relevant policies and proposing next best actions, she said. 

“We encourage teams to explore how small, well-scoped agents can help with day-to-day work, as long as it’s done within clear governance and security guardrails,” said Bristiel. “The goal is to help teams spend less time on repetitive tasks and more time on higher-value work.”

She said, “In all cases, we keep a human accountable for the final decision and any customer-impacting action.”

Not everyone is yet ready to use agentic AI, including business travel management company Gray Dawes Group, although its use has not been ruled out.

The company’s vice president for global IT change and product, Antoine Boatwright, said, “Our approach is intentionally focused on non-agentic, controlled use of generative AI rather than autonomous systems or customer-facing decisioning. 

“The emphasis is on responsible adoption, starting with well-defined use cases, validating value through pilots and then scaling if and where appropriate. With key partners we are using limited AI for productivity improvement such as Github CoPilot.”

Accor’s Amro Khoudeir is clear on the benefits of using AI agents.

“The companies that will win in travel over the next five years won't just be the ones using AI—they'll be the ones that have embedded it into how they operate.”