From pilot to platform: ​Why AI must run through the enterprise

AI will create little value in companies if it remains trapped inside labs, and isolated proof-of-concept projects. Value comes when workflows are reimagined with AI, and true transformation comes when AI’s embedded into the enterprise fabric as part of a CEO-led cultural change. That was the message from a webinar we held last week.

Sangeeta Gupta, senior VP and chief strategy officer at Nasscom, said too many companies are still approaching AI as something that can simply be inserted into existing workflows. “If you have to get the value of AI and agentic AI, you have to rethink workflows endto-end,” she said.

Tapan Sahoo, senior executive officer for digital enterprise at Maruti Suzuki India, provided a striking example of that. Most of us don’t read our vehicle owner manuals. And when something goes wrong with the vehicle, we rarely look up the manual. In collaboration with a startup, Maruti built an AI tool where a customer can photograph a dashboard warning or a faulty part and get step-by-step guidance on solving the problem. That changes the workflow from manual lookup to AI-guided troubleshooting. “It’s perfect,” Sahoo said. The solution has been deployed in India and could go to Japan next.

Sanjay Chalke, CEO for India at Capgemini, put it more bluntly. “AI has to stop being something we experiment with and start something we run the business on,” he said. “It’s not about proof of concept, but it’s about proof of impact.”

That means embedding AI into the core workflows of pricing, customer service, supply chains and decision-making, not leaving it to a small team of data scientists or technologists. It also means leadership has to own the shift. “It starts top down,” Chalke said. “It’s actually a CEO-level cultural shift. It’s not a technology upgrade.”

Sahoo offered a memorable way to think about it. The question, he said, is whether AI is treated like an application or “like oxygen – silent, invisible, but indispensable.”

For Sahoo, AI is truly embedded only when it moves from the shopfloor to the boardroom. It must not just improve quality or efficiency in pockets. It must create value across the organisation. Gupta underlined that point, saying the focus must be on making sure AI is used not just for efficiency and cost savings, but also for decision-making backed by human intelligence.

Enterprise discipline needed

All agreed that the biggest barrier is not lack of excitement around AI, but lack of enterprise discipline. Gupta said the corporate world is now split between leaders and laggards. In the first group, CEOs are personally trying to understand what AI can do for their business. In the second, AI is left to a head of data science or a small AI team and treated as a side project – with many often worrying about issues like AI hallucinations.

That is where AI governance becomes critical. Sahoo argued that governance is often misunderstood as something that slows innovation. In fact, he said, it is what makes speed possible. “In a car, the brake is there for you so that you can accelerate faster,” he said. And emphasised that with another analogy from automotive design: Safety, he said, is built into the chassis, not added after an accident. The same logic should apply to AI. Guardrails should be designed in from the start, not bolted on later.

At Maruti, not every AI use case is treated the same way. Drafting an email is very different from a model predicting safety failures. Different applications need different levels of control. Sahoo said the company has built a risk evaluation matrix and tool through which all AI projects pass before deployment.

The data problem also remains central. Enterprises are cluttered with legacy systems and siloed data. A pilot in one business unit may work well, but the moment a company tries to scale AI across functions, poor data quality and fragmented access begin to show.

Chalke said companies need robust foundations. He broke AI readiness down into what he called seven levers, including infrastructure, process architecture, talent, performance metrics, business data, decision authority and risk controls. Good data matters most. “If data is poor,” he said, “you can use any platform, any tool, you will never be successful.”

As for talent, they need today to be AI native, not simply be skilled in AI tools. “We can always say that everybody is trained,” Chalke said. “But what’s more important for us is, how many of them are AI native. They’re not just specialists, but people who are comfortable working with AI.”