Two years ago, Sreekandh Balakrishnan was at an all-time professional low. At 45, he’d built a successful career as a technology leader, becoming technology director at Travelopia, a global travel company, in 2020. But the AI tidal wave looked ready to make him irrelevant.
“A lot of senior tech leaders felt the ground shifting under their feet. We’d spent decades perfecting our craft, but suddenly here was this AI that could generate in minutes what might take our best engineers hours. The instinctive response was fear,” says Balakrishnan.
After initially dismissing AI as a fad, Balakrishnan realised his fear was making him more vulnerable. “One day my CFO is going to come into the office and ask me how we can use AI to reduce our team by 50%, and either I won’t know the answer, or I can tell them that I’ve figured out how to use AI for more than just that,” he says.
Experiment and learn
Balakrishnan set out to learn any AI tool he could get his hands on. Rather than being told how AI boosted productivity, he wanted to experiment with and develop a practical understanding of the technology. Only then would he decide what to do with it.
Settling on an ‘experiment-and-learn’ approach calmed some nerves – people tend to get nervous about tools that might replace them. He was also careful to establish guardrails. Balakrishnan agreed with the company that he could take two full-time developers out of their current team to spend three months playing with AI.
You can’t lead your organisation from the back. You need to inspire people to learn along with you
Next, Balakrishnan selected six real business initiatives and put the new, AI-enhanced team to work on the projects in parallel with the traditional, human team, to compare their performance.
Initial progress was slow, as the engineers struggled to understand how to use the AI tools. They used inconsistent prompts and productivity was uneven. But Balakrishnan believes that chaos made a great classroom: “That’s when we started to learn where AI accelerated value and where human intuition reigns supreme – context, quality and storytelling.”
Balakrishnan explains that projects ran smoother when a human remained firmly in the loop, treating AI as a thinking partner, not an autopilot. The job of the human was to review the AI’s suggestion, improve prompts iteratively and understand that AI is fundamentally a process, not just a tool.
After six weeks, Balakrishnan gave both teams a ticket to process and the difference became clear: “The traditional team said they needed two weeks to make those changes, but the AI team requested just one hour. That was a lightbulb moment.”
How to set and track metrics
Another early lesson was that small projects might seem easier, but they often fail because they don’t have enough impact. Instead, Balakrishnan says it’s crucial to choose priority projects that really matter to the business. “This should be something that, if it doesn’t go live, the CEO is going to call me to ask why,” he says.
One of the first projects the AI team tackled was rewriting a module that was used by the business to capture leads and assign them to the best salesperson. “I wasn’t confident the AI could do it, but I knew if it worked, we’d immediately deliver hard revenue to the business,” says Balakrishnan.
These early projects helped define the process for running and evaluating AI. Everything the AI teams built was set live and monitored for a month. If the existing team could learn it, manage it and fix at least one production issue, then the project was deemed a success. If not, it was a learning opportunity.
Travelopia tracks new projects over six weeks, collecting metrics such as time from idea to delivery and total hours to completion. The company also monitors value-added versus non-value-added work. “Default [platforms for planning and monitoring projects] weren’t giving us enough information, so we tweaked them and started to track if work is mitigating risk, reducing cost, generating revenue or improving service. We then add sub-metrics showing if AI was used,” explains Balakrishnan.
These metrics track 25 squads and early results show dramatic reductions in delivery time. A complex module that might have taken six weeks to build can now be done in three days.
Challenges in AI integration
There are still significant challenges, such as the unpredictable costs of AI. Balakrishnan estimates that the company pays up to $200 (£153) per month, per developer, to access multiple AI models. This represents a $250,000 (£191,330) investment annually, which must be factored into ROI calculations.
Don’t be the person who doesn’t know what to say when the CFO knocks on your door
In some cases, Travelopia has chosen to upgrade after realising ‘pro’ models would expose its data to training models, or the AI vendor would own the company’s code. Travelopia also pays for AWS-hosted private LLMs when needed to protect commercially sensitive data.
Then there’s the sheer pace of change. “Every two weeks, Copilot gets better, ChatGPT gets better, Gemini gets better. And every time it gets better, some feature starts working better, but another feature breaks,” Balakrishnan says.
He admits that change management has been extremely challenging. He spent four weeks having ‘heart-to-heart’ conversations with employees about the company’s intentions and outcomes around AI. Some people were discouraged and wanted to quit, but Balakrishnan explained that the goal was learning – and even if the company learned their team could do more using AI, that would be a positive outcome.
Redeploying resources
After a year of experimentation, has AI enabled Travelopia to reduce headcount or costs? Balakrishnan says it’s not that simple.
Certainly, developers can work more quickly. A team of eight people used to be split into four pair programming teams, but now there are eight teams, with each person paired with an AI. These teams work much faster, but this means the business needs more quality-assurance engineers and more product owners. Balakrishnan has on many occasions weighed slowing the velocity, or increasing the budget and hiring more people.
So far his approach has been to redeploy people to new roles, taking steps to slow the work rate. “In a regular team, you have time to take a break and maybe you get a good night’s sleep and wake up with a new way to solve a puzzle. AI works at such a pace that there’s never a break, and I worry people will get burnt out.”
For tech leaders just beginning their AI journey, Balakrishnan’s advice is to embrace it. “You can’t lead your organisation from the back. You need to learn the ecosystem and inspire people to learn along with you,” he says. “Above all, don’t be the person who doesn’t know what to say when the CFO knocks on your door in another two years’ time.”
Two years ago, Sreekandh Balakrishnan was at an all-time professional low. At 45, he’d built a successful career as a technology leader, becoming technology director at Travelopia, a global travel company, in 2020. But the AI tidal wave looked ready to make him irrelevant.
“A lot of senior tech leaders felt the ground shifting under their feet. We’d spent decades perfecting our craft, but suddenly here was this AI that could generate in minutes what might take our best engineers hours. The instinctive response was fear,” says Balakrishnan.
After initially dismissing AI as a fad, Balakrishnan realised his fear was making him more vulnerable. “One day my CFO is going to come into the office and ask me how we can use AI to reduce our team by 50%, and either I won’t know the answer, or I can tell them that I’ve figured out how to use AI for more than just that,” he says.
