Ravi Vasantraj, Global Delivery Head at Mphasis, tells the story of a $40 million yacht that would not start. Technicians charging $5,000 per visit came and went without fixing the problem. Then a mechanic from the garage arrived with a hammer, knocked once on the engine, and got it running. His bill: $400,000.
In an exclusive conversation with AIM, Vasantraj uses the story to explain his view of enterprise artificial intelligence. The model is the hammer, and every vendor now sells one. “But where do you apply it?” he asks.
Models from the US, China and elsewhere are locked in an arms race, he says, but “the arms race will always end up for the benefit of the consumer.” Mphasis CEO Nitin Rakesh made a similar point to analysts in July, saying models would “get commoditised fairly quickly”.
In February, Vasantraj had argued that AI enthusiasm was running ahead of enterprise reality. His focus now is on tangible results. “I’m a big believer of impact; outcomes,” he says. New technologies tend to be overhyped in the short term and underestimated in the long term, and enterprises are “still in the early stages of adoption”.
What has changed, he says, is the level of acceptance. AI is now an assistant at every stage of software development, with humans validating requirements first.
AI-led work made up 63% of the $461 million (~₹43.6 billion) in new total contract value (TCV) Mphasis won in the June quarter. A year earlier, AI-led work represented 68% of $760 million (~₹66 billion). AI-led deals have grown from 12% of the pipeline to about 70%, Rakesh told analysts.
Operating margin stood at 14.8%, down 60 basis points from the previous quarter. Management attributed the decline to ramp-up costs on new deals, lower utilisation ahead of expected growth and acquisition costs. It retained its FY27 guidance of 14.75% to 15.75%.
Mphasis will announce its September-quarter results on November 5, with its earnings call scheduled for November 6.
Why AI Pilots Stall on the Way to Production
Vasantraj draws a distinction between demonstrations and production deployment. Only 5% of AI sandbox projects have made it into production, he says, because the question was never whether AI could write code, but “is that code fit to put into production?”
At Mphasis, he estimates productivity gains at about 20%: the same people can deliver 20% more work, where the alternative would have been to add 30% more people. “Is it giving us 2x acceleration? Not yet.”
Legacy systems are where the pressure most sharp. IBM shares fell 13% on February 23, their steepest one-day drop since 2000, after Anthropic said its Claude Code tool could automate much of COBOL modernisation. Indian IT stocks also fell, and Anthropic’s announcements have rattled the sector more than once this year.
Vasantraj says AI works well where outcomes are measurable, such as converting COBOL end to end, but is less effective when the task involves rethinking something as complex as digital banking.
He puts global tech debt at about $2.8 trillion (~₹265 trillion), citing McKinsey & Company and Nasscom estimates. Converting even 10% of that, he says, represents a $280 billion market, which he considers a conservative estimate. Nasscom leaders, using a different yardstick, put the modernisation opportunity at $1.6 trillion in February.
Mphasis is itself converting 40 million lines of code, a task Vasantraj calls 70 years of tech debt, over more than five years.
Where Context Lives in Talent and Training
If models are cheap, delivery capacity is becoming less of a differentiator, too. “Cost arbitrage is a given,” Vasantraj says. What buyers want to know now is whether a provider understands their business well enough to hold “the keys of my kingdom”.
Mphasis stopped campus interviews for delivery roles about two years ago, he says. New recruits now intern on client projects for three to six months, after which client leaders can make offers on the spot.
“You can’t train people on prompts,” he says, because the quality of those prompts depends on context, which varies between industries such as retail and banking. He expects Mphasis’ headcount to rise, although not in a straight line. “I am still recruiting.”
Org Debt and the Shift to Buying Outcomes
Vasantraj uses the term “org debt” to describe the structural complexity that slows large enterprises down. Over decades, organisations build structures in which infrastructure, applications, operations and cybersecurity report to different people, making it difficult to coordinate work across functions.
Mphasis, which has about 25,000 people in delivery, merged its service lines three years ago. Such a move reflects the agility that mid-sized firms have long pitched as an advantage over larger rivals. At a company of 6,00,000, he says, breaking down these silos is “like Titanic”.
That, he argues, is why enterprises buy from outside. “They won’t buy AI service from you. They will buy outcomes which are using AI,” he says, and he expects that distinction to land within 12 to 24 months. Where outcomes can be measurable, such as monitoring or vulnerability fixes, Mphasis uses fixed pricing.
Tria and the Test of Defensibility
The bet is Mphasis Tria, launched on May 27 with three layers: enterprise knowledge, decision intelligence and agentic execution. Vasantraj says it turns ‘Insight’ into ‘Foresight’ and then ‘Expressions’.
One use case is an insurer seeking to raise net interest income by rebalancing its asset portfolio. Mphasis customises the platform once, then earns annual recurring revenue (ARR) from support, with forward-deployed engineers (FDEs) available when needed.
Rakesh has told analysts that these engineers work in pods of three to five and hand over to scale delivery, with the larger prize being a deployed platform.
Vasantraj’s case for defensibility rests on Mphasis’ position as an external partner. As an outsider, he says, Mphasis can bring infrastructure, applications, operations and cybersecurity together to deliver an outcome. “The owning of the IP itself is meaningless,” he says. “The application of those to solve specific context is where the game is.”
Mphasis’ platforms are designed to capture and retain that context. Chief Solutions Officer Srikumar Ramanathan has said clients own the Ontosphere built for them. The company has said it funds such platforms by redeploying margins.
Risks Outside the Platform
The risks, he says, sit largely outside it. Oil at $150 or $200 a barrel could push clients into recession.
His second worry is AI going rogue. In July, Hugging Face disclosed that an autonomous AI agent system had driven an intrusion into its production environment end to end. A plurality of models, he says, is the best safeguard, because intelligence stays distributed.
He stops short of declaring victory. “I’m not saying that we figured it out,” Vasantraj says. Some clients are validating the approach, he adds, describing the progress as a “good start but things to do”.
