For Julen Mohanty, the problem with enterprise AI is rarely the technology. It is what companies expect the technology to accomplish. As VP (IT Automation) and CIO at Gallagher, Mohanty oversees technology operations for the insurance broking major’s India organisation, which has close to 19,000 employees across Mumbai, Bengaluru, Pune, Kochi, Kolhapur, Shimoga, Chandigarh, Visakhapatnam and Gurugram. The India operation primarily supports Gallagher’s businesses in the US and UK, with a growing domestic role following its acquisition of Edelweiss General Insurance.
That gives Mohanty an unusually large laboratory for thinking about automation. Gallagher carries decades of insurance data, processes and systems, but his approach to modernising them is less about replacing legacy technology and more about identifying where technology can produce a measurable business outcome.
His starting point is blunt. Around 80% of AI projects, he estimates, fail to deliver what organisations expect. The reason, Mohanty believes, is that companies often begin with the technology rather than the problem.
Start with the business case
Mohanty’s test for an AI initiative is straightforward. Does it increase revenue, reduce cost or manage risk? “If it doesn’t do any of these three, don’t implement it,” he says.
That runs against the way many enterprise AI programmes are being conceived. The rapid availability of generative AI and increasingly capable agents has created a temptation to look for places where the technology can be inserted rather than first asking whether a process actually needs it. “If you have a hammer in your hand, you start searching for nails everywhere,” Mohanty observes.
Headcount reduction, he believes, is equally the wrong starting point. Automation may eventually change the amount of human effort required, but that should be a consequence of improving the process rather than the objective. “Nobody’s thinking that by using AI, I will make this process run faster, error free,” he explains.
That approach is visible in Gallagher’s insurance processes. Proposal generation once involved pulling information from more than a dozen systems. Automation has compressed that work dramatically, but the output still goes through human review before reaching the customer.
Claims processing follows a similar model. The objective is not to remove people from the process at any cost, but to eliminate repetitive work while keeping accountability where it matters.
The human stays accountable
Mohanty had arrived at this idea well before generative AI became a boardroom priority. Almost a decade back he coined the term “Humbotai”, combining human, bot and AI. The concept was based on a simple observation that has become more relevant as autonomous systems have grown more capable. A machine may execute an action, but responsibility still belongs to a person.
“You cannot have accountability to a bot,” he points out. “You cannot fine a bot, you cannot penalise a bot.” For Mohanty, the future employee is therefore less a standalone human worker and more a human working alongside software and AI systems.
The distinction is particularly relevant in insurance, where decisions can affect customers, claims and financial outcomes. At Gallagher, AI can take on repetitive work, but human oversight remains part of the design.
The same principle applies to customer service. If a customer has already completed verification while requesting an address change, an agent should not have to make the customer repeat the same exercise simply because another system cannot recognise what has already happened.
The objective, Mohanty stresses, is not automation for its own sake. It is removing unnecessary friction.
Data becomes valuable only when it is used
Mohanty is equally sceptical of another familiar technology metaphor. “Data is not the new oil, data is a new cash machine,” he says.
His distinction is important. Data becomes valuable only when an organisation knows what it contains, connects it to a business problem and can act on it.
That matters for Gallagher, where large volumes of historical information sit across established systems. Mohanty does not believe legacy systems should automatically be discarded because newer technology exists. A system that has worked reliably for two decades does not become obsolete simply because an AI platform has arrived.
The question, he notes, is what the business needs from it. That thinking also shapes his view of AI governance. Data governance, in his assessment, has to come first. Without clarity around data, ownership and permitted use, AI governance becomes difficult to enforce.
Measure before automating
Gallagher’s approach therefore begins with measurement. The organisation looks at the strength of a process, identifies which steps can be automated and establishes how long the existing process takes. It then measures what changes after automation.
The results vary significantly. “There are cases where we have seen it is less than 30%, but there are also cases where we have seen it is more than 70%,” Mohanty shares, referring to time savings across different processes.
The variation matters because it challenges the idea that AI produces a standard efficiency gain that can simply be applied across an enterprise.
A process that is already efficient may offer little room for improvement. Another involving multiple systems, repetitive decisions and manual intervention may present a much larger opportunity. For Mohanty, that discipline is central to getting AI right.
From delivery centre to innovation engine
Mohanty also has a broader view of India’s GCC evolution. Gallagher’s India operation began around 2006-07 as a contact centre before expanding into technology support and wider business operations. Yet Mohanty questions whether the term “Global Capability Centre” accurately describes what many such operations actually do.
In his assessment, a significant number remain delivery organisations supporting overseas businesses rather than genuinely global centres of product innovation. He estimates that fewer than 5% of GCCs in India undertake meaningful product innovation or file patents.
India has established itself as a formidable technology delivery base, he observes, but the next step is to create more intellectual property rather than primarily execute work defined elsewhere.
He compares the tendency to solve immediate requirements with Bengaluru’s Outer Ring Road, infrastructure designed for an immediate need but now carrying the consequences of years of growth. The lesson for technology organisations is similar. Solving today’s problem is not the same as designing for the next decade.
The technology follows the culture
At Gallagher, Mohanty believes the differentiator is less likely to be a particular AI platform than the organisation’s culture.
The company refers to this as the Gallagher Way, with an emphasis on collaboration rather than authority. Mohanty recalls a recent interaction with Gallagher’s global CEO, who recognised him by name during a visit. For him, that moment reflected the organisation’s culture.
Technology transformation, he believes, depends on that kind of environment because AI inevitably changes processes, roles and decision-making. That also becomes important as AI systems become more autonomous.
Asked about the possibility of AI agents escaping controlled environments and compromising other systems, Mohanty compares cybersecurity testing to checking a drum for leaks. You inspect it first, then put it under water and look for bubbles.
Security controls, he notes, must become progressively more rigorous as systems become more capable. Attackers will continue looking for weaknesses, making resilience an ongoing exercise rather than a destination.
His conclusion is simple. “You cannot stop the rain,” Mohanty says. “You have to carry an umbrella.”
For a CIO navigating the current AI cycle, it is also a useful definition of the job. Technology will keep changing. The harder task is deciding where it belongs, proving that it works and ensuring that someone remains accountable when it does not.