Speaking at an aviation event recently, Riccardo Vittoria raised an issue that has long been simmering within the travel industry: the hidden costs of artificial intelligence (AI).
AI is being sold to the travel industry as a near-universal solution to lower costs, improve efficiency and delight customers. But as more companies move beyond tests and pilots into deployment, a murkier picture is emerging.
Vittoria, co-founder and CEO of AI platform Acai, said onstage at the Airline Distribution 2026 conference in March: “In servicing, AI can be a disruptor; in bookings, we need to see. I say that because agentic AI has a lot of cognitive power, it can do so much, it can do things better than humans lightning fast, but it’s still expensive."
Paying the innovation bill
At another recent event focusing on hospitality, tech leaders were equally honest.
"Whenever something is new, you're having to pay for the development of the person that created it," said William Connors, international retail systems manager at Starbucks, at Hospitality Tech360 in London.
Many entrepreneurs also complain about “subscription creep,” where the cost of using an AI service rises over time, either to support growth or unlock new features.
Alex Ragin, founder and CEO at software development company Zoftify, has seen this first-hand and draws a parallel with the trajectory of other cloud services.
"As your app grows, as you get more users, you have to pay a huge premium on top for this kind of a convenience, rather than if you were to build it yourself," Ragin said.
Energy prices are also a factor, because at a basic level, the foundations of AI companies are built on tokens and computer processing, with data centers consuming vast amounts of electricity. More cost hikes are expected due to the situation in Iran, which is disrupting global energy markets.
Meanwhile, a report from Gartner predicts more than 40% of agentic AI projects will be canceled by the end of 2027 due to escalating costs. It also warns about so-called “agent washing.”
“Many solutions labeled as AI agents are actually extraction or classification tools or workflow automation with simple rules,” said Romy Abbrederis, CEO and co-founder of LobbyAI.
Cost and context
AI is being implemented across a range of applications, including to increase efficiency. But the travel industry is now asking, "What exactly does 'expensive' mean?" The answer depends on what you’re comparing AI to, and in what context.
According to Acai’s Vittoria, the economics of AI vary dramatically depending on what you are replacing.
"When we are replacing human work, the unit economics are extremely effective," he told PhocusWire. "We can make AI-driven solutions that can cost a twentieth of a human depending on where the human is based."
Acai focuses on post-booking scenarios, such as flight changes, cancellations and disruptions. Deploying a human agent to handle a complex rebooking might cost $30 to $50; an AI-powered equivalent can handle the same task for a few dollars, Vittoria said.
But the math changes when AI is used to replace what Vittoria calls “deterministic technology,” or the established tech systems that underpin (for now) much of the travel industry’s booking infrastructure.
"The cost is always higher and sometimes too high," he said. His concern is that airlines, TMCs and OTAs are rushing to build agentic booking experiences without understanding the unit economics.
An AI agent handling a new booking through a call center interface could cost anywhere from 80 cents for a simple transaction to $10 if it involves complex back-and-forth with a GDS, Vittoria continued. Compare that to the same ticket made via an online booking tool: "Operating that new booking on the OBT [online booking tool] is a few cents; operating on the agent layer is a few dollars," Vittoria said.
Procurement insights
Accommodation platform HRS has been running the numbers too.
“It's not free, for sure,” said chief product officer Martin Biermann.
HRS has invested in AI over the past couple of years, launching its AI-powered Copilot platform to help businesses manage their corporate lodging and meetings programs. Biermann said that even if clients do not take advantage of AI, the company still has running costs.
“You have your license costs for infrastructure, for running RFPs, for contracting, for auditing. You have a lot of people globally deployed to that. You have regional travel managers, local travel managers. You have procurement folks,” he said. “When we say what is the cost of AI, we need to put this into context of what kind of work you are trying to achieve.”
Where Biermann sees the biggest gains is speed, mostly by running 30 to 50 additional AI-driven optimization cycles per year, covering rate caps, policy calibration and expense data scanning. AI generates an additional 10% yield, he said.
Biermann also calculates internal headcount efficiency ratios. While it previously took four employees to deliver 1% savings for one customer, using AI slashes that to one fifth of an employee, equivalent to 20x improvement in “personnel intensity,” he said. The savings matches Vittoria’s efficiency figure.
Infrastructure advantage
Of course, larger companies like HRS have the means to invest in technology. Startups might have the advantage of building “AI-first” but can face the costs mentioned earlier.
At an enterprise level, many international organizations have been migrating their data to Amazon Web Services (AWS), Azure and other cloud infrastructure providers in preparation for the AI revolution.
Wyndham Hotels & Resorts, for example, has invested $425 million in modernizing its technology stack over the past eight years, moving all its operations to AWS. It has a “93% cloud optimization score,” according to Mike Mahar, senior vice president and head of commercial technology, which he said is above the industry average of 70%-85%.
The company recently spent $100,000 connecting its data to several LLMs—a cost CEO Geoffrey Ballotti described as nominal—so guests can use ChatGPT and other tools to interact with hotels.
"The MCP (Model Context Protocol) itself is not an expensive lift," Mahar said. "There is a very purposeful approach. How do we add value for our owners and how do we offset that cost with some cost savings in our organization?"
Value comes in the form of platforms like the AI-powered guest messaging tool Wyndham Connect, which was rolled out to hotels for free. The business case is built around offsetting call center volume, a cost that Wyndham covers.
Hotels also benefit from smarter upselling through AI-powered tools. Mahar said at the top end, 5% of its hotels save an average of $61,000 a year. One hotel has managed to save $120,000.
The employment equation
However, for independent hotels it’s a different story. “The hidden cost of AI in hospitality isn’t the technology, it’s the work to get ready that operators underestimate,” said Nicola Longfield, chief commercial officer, global accommodation and payments at Access Hospitality. Its “AI in the Hospitality Sector” report reveals 58% of hotel leaders in the U.S., for example, worry about sharing data with AI tools.
Sandrine Zechbauer, chief marketing officer at RMS, said many hoteliers know they need to adopt AI but struggle to understand how to implement it. “They either need to build this knowledge themselves, invest time in understanding how modern hotel tech stacks work or hire people who can bridge that gap, which for many businesses is an added cost.”
She said she was starting to see the emergence of a new kind of professional, the “hospitality engineer,” who can blend an appreciation of service and guest experience with a practical understanding of technology, integrations and data.
Yet hiring that knowledge costs more money too. HRS’s Biermann said AI-related talent costs are rising at about 30% each year, and people are increasingly hard to find. “Recruiting specialist firms say these in-demand roles now take 50 times longer to recruit,” he said.
In search of a "beautiful world"
Overall, the travel industry is being asked to handle AI with caution. Design and deployment are key, and Acai’s Vittoria argues a blanket, fully agentic design is the wrong approach, with hybrid models the way forward for now.
The vision for the future, though, is for an interconnected, agentic travel ecosystem in which systems across the supply chain, such as airlines, hotels, TMCs and OTAs communicate directly with each other.
“We need to land into this beautiful world where travel works for booking, for post booking, for disruptions," he says. "But we need to be careful how we do it—not to shoot ourselves in the foot and stop an initiative because we didn't design it right from the beginning."
He added he has been trying to make this argument for the past 18 months. “We cannot make the mistake of thinking AI is going to solve everything. Then we get a $50 million electricity cost from AWS and shut it down, just because we designed the wrong solution.”