How should travel companies measure AI success?

Some big companies have recently been crowing about how much of their code is generated by artificial intelligence (AI).

In Google’s first quarter earnings call this year, CEO Sundar Pichai said that the use of AI internally had been “transformative.”

He said that that “well over 30%” of code involved people accepting AI-suggested solutions, up from the 25% he announced in the Q3 2024 earnings call last October.

Google is not alone.

In an April 2025 fireside chat between Microsoft boss Satya Nadella and Meta CEO Mark Zuckerberg, the former said: “I’d say maybe 20%, 30% of the code … and some of our projects are probably all written by software.”

Why is the metric so important?

The answer is productivity and, by extension, profitability.

A study of 4,867 developers at Microsoft, Accenture and an anonymous Fortune 100 company found that there was a 26% increase in completed tasks among developers using AI tools and that junior developers enjoyed even bigger productivity boosts.

The rise in adoption is far from unexpected. According to Stack Overflow’s Developer Survey in 2024, 61.8% of developers are using AI in the development process, up from 44% in 2023; a further 12.8% of developers said they intend to use such tools soon.

Should travel companies be following Google and Microsoft and sharing this metric alongside more traditional metrics?

Jon Pickles, founder of Sygnifiq and chair of the Travel Technology Initiative does not believe companies should be using such a metric since there is no precedent to do so.

“Enhancements, refactorings, auto-completions, code linters, code review tools, macros, templates, etc. have all been used; none of these were required to be declared,” he said. “Many of them assist development, change style, speed work, but we didn’t have a norm of publicly quantifying them. AI is in many ways just a new form of assistance,” he said.

John Morhous, CXO and former CIO of Flight Centre Travel Group, said, “There is no universal metric that says writing X% of code with AI makes you good or bad, but we use the number to demonstrate our commitment to using innovative technologies (like AI) to do things more efficiently, which ultimately translates to greater shareholder value.”

He said that AI-enabled software development has enjoyed clear success across all industries, not just in travel.

“We are not replacing engineers with AI robots, however, just using the tools to help make our engineers more productive at what they do. That may be through writing some code for specific things that are quite repetitive or using them for portions of the software development lifecycle (like regression testing) where it makes the process a lot faster.”

He added, “Like many roles, software developers have a portion of what they do which is very repetitive, or syntax oriented, which is a perfect use case for AI tools to assist. It is not replacing the critical thinking required to solve customer problems, but it can definitely help make parts more efficient.”

Travel management company Gray Dawes Group (GDG) says it has a “pragmatic and exploratory” approach to AI. It has given developers access to several AI tools and asked them to investigate use cases where it might help increase productivity.

GDG’s SVP of global IT change and product Antoine Boatwright does not believe it is a meaningful metric for many reasons.

“Generating code does not mean good code has been generated,” he said. “AI could generate less efficient code so measuring lines of code generated would not reflect efficiency aspects like code size, maintainability, code speed and modularity for scalability purposes. It may also add testing overhead and reduce productivity.”

Boatwright believes there are better metrics that could be tracked, such as feature throughput, code quality as measured by bugs, application scalability and, depending on the code base, tech debt backlog.

In July when Indian online travel agency Ixigo released its Q1 2026 earnings figures, the company’s CTO and co-founder Rajnish Kumar said the metric is “fundamentally flawed if we’re trying to measure the real impact of AI in software engineering.”

He wrote: “Coding itself is not the main bottleneck in software development. In most real-world engineering workflows, writing code only accounts for about 20% to 30% of a developer’s time. The remaining 70% to 80% is spent on far more cognitively intensive and collaborative tasks, like system design, architectural planning, writing detailed documentation, thinking through edge cases, defining interfaces, creating test strategies and setting up CI/CD pipelines. In that context, even if an AI assistant can generate 80% of your code, that’s still just a productivity boost on a small slice of the overall effort. Mathematically, it amounts to about a 15% to 20% efficiency gain at best, and that’s assuming near-perfect AI-generated code, which often still requires review, debugging and refactoring.”

He added, “Code generation has been getting incrementally easier for years—whether it’s via autocomplete, StackOverflow copy-pasting or more recently with tools like Copilot or Cursor. So, while these are impressive evolutions in productivity, they are evolutionary, not revolutionary.”

The risk of the reveal

Revealing these percentage figures also has potential pitfalls, argues Jon Pickles.

“Showing dependence on AI generation might be interpreted by investors, partners, competitors as a weakness. On the other hand, claiming a high percentage might backfire if downstream code quality, maintainability or security issues surface,” he said.

As in many other uses of AI, using it to generate code raises questions over intellectual property.

“Revealing that a large fraction of code is generated by AI raises questions about where the training data came from, whether any licensed or copyrighted code was inadvertently reproduced or derived,” Pickles said

As in many other uses of AI, using it to generate code raises questions over intellectual property.

“Revealing that a large fraction of code is generated by AI raises questions about where the training data came from, whether any licensed or copyrighted code was inadvertently reproduced or derived.”

Ixigo’s Kumar believes that rather than asking companies what percentage of their code is AI-generated, a better question would be what percentage of the end-to-end engineering process is now autonomously handled or significantly accelerated by AI?

He said, “That’s the metric that will truly reflect how AI is transforming software development, not just in quantity but in quality, velocity and scale. [That number] … is currently north of 40%.”

To reveal or not reveal, that is the question

There may be a bigger question—not everyone believes in AI’s productivity gains.

GDG’s Boatwright said recent articles by Gartner and MIT show “we have hit a ‘trough of disillusionment’” and that “95% of gen AI projects are failing.”

Another study focusing on 16 developers with moderate experience found that they took 19% longer to complete a series of 246 complex tasks using AI tools than without.

If AI Is not making development faster and cheaper, then why promote the metric?

The cynical among us might argue that Google and Microsoft’s willingness to reveal what proportion of code is AI-generated is not so much about productivity but more about promoting their own AI-powered coding tools.