Vijay Vijayasankar, Global Agentic AI Officer at Genpact – Interview Series

Vijay Vijayasankar, Global Agentic AI Officer at Genpact, leads the company’s global productization of Agentic AI solutions, with a focus on shifting enterprise value creation from traditional manpower-based models toward higher-margin, outcome-based frameworks. In his role, he is helping shape Genpact’s approach to scaling AI across the enterprise by building organizational models around both AI Builders, who design advanced orchestration layers, and AI Practitioners, who bring domain expertise into workflows powered by autonomous agents. Before joining Genpact, Vijayasankar held senior leadership roles at IBM, including Managing Partner for Financial Services, CTO of IBM Services North America, and General Manager for Cognitive, IoT, Analytics, and Watson Health in GBS North America, alongside earlier executive roles at MongoDB (MDB ) and SAP.

Genpact is an agentic and advanced technology solutions company that helps large enterprises modernize operations through process intelligence, artificial intelligence, data, and digital transformation. The company works across areas such as finance and accounting, customer care, risk and compliance, sourcing and procurement, supply chain, technology, and trust and safety, while its Data and AI services focus on helping organizations use advanced analytics, AI, and autonomous agents to turn complex data into measurable business outcomes.

You’ve spent more than two decades at companies including IBM, MongoDB, SAP, and now Genpact, where you serve as Global Agentic AI Officer. Looking back across the evolution from big data and analytics to today’s agentic AI systems, what lessons have most shaped your view of how enterprises should approach AI transformation?

I’ve watched this same movie three times now. Big data was the silver bullet for decision-making. Cloud was the instant switch for organizational agility. Now, agentic AI is the cure for operations. In every cycle, the companies that generated value focused on their operating models, their processes, and their people. The ones that struggled, and many still do, treated the technology itself as the transformation.

What’s fundamentally different about agentic AI is its location. It’s sitting at the execution layer, not the insight layer. Older tools generated charts and recommendations. Agents take action. A bad analytics dashboard produces a bad chart that a human analyst catches before it hits the CEO’s desk. A bad agent executes a bad transaction at scale before anyone realizes it’s happened. That changes the risk profile completely, and it forces a level of operational discipline that most companies haven’t needed before. I’ve learned to be suspicious of my own excitement. Agentic AI is fascinating, but that’s exactly when you need to slow down and audit the foundations.

Genpact’s new report with HFS Research estimates that unresolved enterprise debt is trapping nearly $18 trillion in potential value globally. What surprised you most about the findings, and why do you believe so many organizations continue to underestimate the impact of process, data, technology, and talent debt?

What was eye-opening is the scale of the value trapped behind inefficient processes, fragmented data, aging technology, and workforce readiness challenges being measured in trillions of dollars.

What matters more than the total figure is how the problem is structured. Most leaders look at these issues as separate silos. A data problem, a process problem, a technology problem, a skills problem, each treated independently. They aren’t separate. Poor data forces people to invent manual workarounds. Those workarounds create process complexity. That complexity burns out your best people and makes your operations fragile. By the time an AI initiative arrives, it runs into all of that simultaneously. The AI gets blamed for a failure that was baked into the business a decade ago. We talk about enterprise debt as a cost burden, but it’s more accurately a growth constraint. That distinction matters because it changes who owns the problem and what fixing it is worth.

The report found that 85% of executives believe enterprise debt is limiting AI value realization. Why do you think enterprises continue investing heavily in AI while often neglecting the foundational issues that prevent those investments from delivering returns?

Because AI is visible and foundations are invisible. Every board wants to see an AI strategy. Every CEO needs a story to tell. Deploying a shiny new model is a headline. Redesigning a core supply chain process is a thankless, invisible grind until it finally works.

I’ve seen organizations spend six months selecting a model and six days examining the process that model is supposed to improve. That ratio is backward, and they find out the hard way. The bottleneck is almost never the AI. It’s inconsistent processes, disconnected systems, and data that nobody has been accountable for in years. AI doesn’t solve those things. It exposes them faster. The incentive structures don’t reward fixing foundations. There’s no demo for cleaning up data governance. The executives who do it quietly are making a bet that operational health compounds over time, but most boards aren’t measuring that.

Data debt emerged as the largest AI blocker in the research. What practical steps should enterprises take today to move from fragmented, low-quality data environments to data foundations that can support agentic AI at scale?

Stop trying to launch a massive, top-down governance program. That’s a technology answer to an accountability problem. In most large organizations, no one owns the accuracy of the information that runs the business.

The practical starting point is to pick a handful of high-value processes and work backward. What decisions need to be made? What data do those decisions require? Who owns that data? How often is it wrong? That exercise surfaces the accountability gaps immediately. And we must get this right because agents fail quietly. A human analyst sees two systems disagreeing and makes a judgment call. An agent arbitrates between the conflicting data and executes with total confidence. If different parts of your organization define a customer or a transaction differently, the agent is making confident decisions from bad inputs and nothing in your organization catches it. That is a new category of operational risk that most organizations haven’t fully priced in.

Agents don’t handle ambiguity the way humans do. This is also why the success of an agentic solution is less about the intelligence of the models and more about the quality of guardrails put in place.

You have argued that there is “no AI without process intelligence.” Why do you believe process redesign is often more important than the AI model itself when organizations attempt to automate complex business operations?

Because the model performs as well as the process it’s embedded in, not better. Organizations obsess over model selection because that’s where the visible innovation is, but the process is where value is created or lost.

If a workflow has redundant approvals, duplicate work, or unclear ownership, AI doesn’t fix those things. It accelerates them. You get a broken outcome faster and at a greater scale. People often think a thorough process redesign slows them down, but the reality is exactly the opposite. You save the eighteen months you would have spent unwinding a bad deployment. Think of it like a vehicle: you can put the best engine ever built into a car with misaligned wheels, and it’s still going to pull to the left.

Many organizations are experimenting with AI agents, but relatively few have successfully deployed them across core business functions. What separates enterprises that are moving beyond pilots from those that remain stuck in proof-of-concept mode?

Pilots are designed to succeed. You pick the best process, the cleanest data, and the most motivated team. Scaling means hitting the real environment where none of those conditions hold.

The organizations that scale stop creating artificial conditions from the start. They pick a real operational problem, assign an owner who is accountable for the business outcome, not the technology, and bake governance into the design from day one. We have clients running agents in production across accounts payable and transaction monitoring. The productivity numbers are there, but the real change is structural. The end state is a fundamentally different workflow where people are freed up to do the strategic work they never had time to handle before. Scaling agents is a leadership challenge, not a software challenge.

The report highlights that only 6% of enterprises have successfully resolved their major debt challenges. What are these organizations doing differently, and what common mistakes do you see among companies that struggle to make meaningful progress?

They treat it as a business problem, not an IT program. It’s tied to executive sponsorship and measurable business performance. They understand that operational health isn’t separate from growth strategy. It is the growth strategy.

The other 94% are chasing symptoms. They deploy a tool to fix a process problem or launch an AI initiative to compensate for poor data. That produces a short-term bump, but the underlying issue remains and the debt keeps compounding. The leaders who break the cycle are willing to say what most executives won’t: this process has been broken for fifteen years, we’ve built entire operating models around its failures, and we’re going to tear it down and rebuild it from scratch. That admission is uncomfortable. It invites questions about why it wasn’t fixed earlier. The ones who make that call are the ones making real progress.

You have spoken about the distinction between “AI Builders” and “AI Practitioners.” How do you see enterprise workforces evolving over the next five years, and what skills will become most valuable as humans increasingly work alongside autonomous agents?

The most valuable skill will be knowing how to frame a problem. As AI handles more execution, the quality of the result depends on the quality of the question and the judgment applied to the output. You need to be able to recognize when something looks right but is structurally wrong.

Beyond that, I’d highlight three specific things. First is domain depth. General AI capabilities are becoming commoditized quickly, so generalists are losing leverage. The premium will go to deep domain experts who combine genuine functional expertise with the ability to apply AI within that context. Second is exception handling. You need people who know what to do when the agent hits a wall or the financial stakes are too high to automate, which requires institutional wisdom rather than training data. Third is workflow design, meaning the ability to structure work so that humans and intelligent systems are each doing what they are best suited for, and to recognize when that division needs to shift. It’s not a term the industry has settled on yet, but the skill is real and currently undervalued.

There is ongoing debate about whether AI will eliminate jobs or transform them. You’ve argued that much of the concern comes from misunderstanding what AI actually does well. How do you see the relationship between human expertise and AI agents developing as adoption accelerates?

AI is spectacular at tasks that are clearly defined, high-volume, and pattern-dependent, such as processing invoices, monitoring trade anomalies, and drafting routine communications. It does those things faster and at a scale no human team can touch. However, AI cannot hold accountability for real-world outcomes, nor can it handle things like novelty, ambiguity, and messy human relationships, which is why human expertise remains essential.

Most jobs are a collection of tasks, not a single activity, and AI will absorb the execution layer over time. That doesn’t eliminate the need for people, it just changes where they create value. As routine work becomes automated, expertise shifts toward solving exceptions, redesigning processes, and helping organizations adapt as technology and business needs evolve.  Some roles will be change significantly, and leaders have a responsibility to address whether the new work being created is accessible to the people whose old work disappears. The answer depends almost entirely on the choices organizations make about training and transition. That’s a leadership decision, not a technology outcome. The companies I respect most are being honest with their teams about which roles are changing, over what timeline, where human expertise remains extremely valuable, and what realistic paths forward look like. Most are not having that conversation yet.

Looking ahead, what does a truly agentic enterprise look like in practice? If we revisit this conversation five years from now, what changes in business operations, organizational design, and AI deployment do you expect will have become standard across leading enterprises?

We’ll stop talking about AI as a separate initiative, just like we stopped talking about going digital. It will simply be how work gets done.

The structural change is what interests me most. A lot of our current hierarchy exists because information was hard to move. We have approval chains and coordination roles because information had to travel from where it was created to where decisions were made. Agents collapse that distance. When they do, a significant amount of coordination infrastructure becomes redundant. Organizations that recognize this early and redesign around it will have a speed advantage that compounds.

We’re also going to see a shift in how AI is bought and delivered. Software companies build products but don’t take accountability for outcomes. Services companies take accountability but struggle to build scalable products. Both models are inadequate. What we’re building at Genpact is productized AI in specific domains, delivered as an outcome. You buy the result and we’re on the hook for it. I think that becomes the standard because it aligns incentives correctly. The vendor who only sells capability has no stake in whether it works in your environment. The ones accountable for the outcome have every stake. The winners of this era won’t be defined by having the most sophisticated AI models. They will be defined by mastering the structural redesign between humans, processes, and intelligent systems. We are only at day one of that shift.

Thank you for the great interview, readers who wish to learn more should visit Genpact.