Jitendra Putcha, COO of Tredence – Interview Series

Jitendra Putcha, COO of Tredence, is a seasoned technology and operations executive with nearly three decades of experience spanning artificial intelligence, data analytics, digital transformation, delivery, and enterprise technology services. Before joining Tredence in April 2026, he served as Chief Delivery Officer at Mastech Digital, where he was part of the executive leadership team overseeing strategy, transformation, P&L responsibilities, and service delivery. He previously spent several years at LTIMindtree and LTI, ultimately serving as Senior Vice President and Global Head of Data & AI, with responsibility for enterprise AI adoption, generative AI, AI agents, data and analytics services, product engineering, and platform development. Earlier, Putcha spent nearly 18 years at Cognizant, progressing through senior leadership positions including Vice President of AI & Analytics and Hyderabad Center Head, where his work covered global data and analytics delivery, AI-driven transformation, and the scaling of analytics capabilities across industries. His earlier career included roles at Wipro and DSQ Software, giving him a career foundation that spans data warehousing, business intelligence, analytics, and large-scale enterprise delivery.

Tredence is a global data science and AI solutions company focused on addressing what it calls the “last-mile” problem in AI—the gap between generating insights and turning them into measurable business outcomes. Founded in 2013, the company combines industry expertise with data engineering, data science, machine learning, generative AI, AI agents, cloud technologies, and proprietary accelerators to help enterprises operationalize AI at scale. Tredence says it has more than 4,200 employees and works with companies across retail, consumer packaged goods, technology, telecommunications, healthcare, travel, industrials, and other sectors. Its approach emphasizes embedding analytics and AI into real business processes rather than stopping at models or dashboards, with capabilities spanning enterprise-ready generative AI, multimodal and agentic systems, responsible AI, data modernization, and AI governance.

You bring more than 28 years of experience spanning enterprise data, analytics, AI, and large-scale global delivery, including leadership roles at Cognizant, LTIMindtree, and Mastech Digital before joining Tredence as Chief Operating Officer in 2026. What attracted you to Tredence at this particular stage of its growth, and what do you see as the biggest opportunity to reshape how an AI-focused services company operates and delivers value as enterprises move from experimentation to AI at scale?

What drew me to Tredence was how grounded the company is. In my career, I’ve seen plenty of technology shifts, but it’s rare to find an organization that connects deep AI expertise directly with real business outcomes. Tredence doesn’t just talk about AI for the sake of it; it builds solutions that solve actual enterprise problems.

Right now, the industry is at an amazing inflection point. It is moving past the testing phase and asking the tough question: How do we embed AI into our daily operations? This is what we call the “last mile” of AI: bridging the gap between generating insights and making sure those insights reach the specific person, process, or decision where they can produce a useful result.

The opportunity here isn’t just about deploying cooler tools. It’s about solving an execution problem, helping organizations think differently, and seeing measurable value. As COO, my job is to protect that edge. I want to make sure we scale up our operations without losing our agility or client focus, turning high-level AI ambitions into real-world returns.

We continue to see enterprises run promising AI pilots that never become meaningful production systems. When you look at the projects that stall, what are the most common failure points, and which of those problems are executives still underestimating?

Honestly, most AI pilots don’t fail because the tech fails. They fail because we treat AI like a powerful engine but forget to build the actual car around it. Think of it this way: A world-class engine is completely useless if you don’t give it clean fuel, a steering wheel, and a driver.

Your AI model will sit stranded in the garage the same way if you underestimate three fundamentals. Trusted data is your clean fuel. If your data is a messy, unorganized baseline, your engine clogs up and your model feeds back unreliable answers. Strong governance is your steering wheel. Without clear rules and safety guardrails, teams are simply too afraid of the risk to put the car on the road.

Finally, clear business ownership is your driver, and it’s the piece executives underestimate the most. If a business leader doesn’t take the wheel, champion the tool, and embed it directly into their team’s daily workflows, the project struggles to deliver results. We must escape the “demo trap.” Success isn’t about launching lots of pilots or gathering test engines. It is about locking down your data, setting clear guardrails, assigning real accountability, and embedding AI quietly into how your business operates every day.

Tredence has long described the “last mile” of AI as connecting insights to action. Agentic AI potentially takes that much further by allowing software to execute decisions. How does the last-mile problem change when AI moves from recommending an action to taking that action autonomously?

In the past, dealing with the last mile of AI was a lot like driving with a GPS. The screen told you exactly where to turn, but you still had to physically move the wheel. Agentic AI completely flips that. Now, the car is essentially driving itself.

This changes the entire conversation. We’re no longer just trying to convince a manager to trust a chart on a dashboard; we’re building systems we can trust to make decisions on our behalf. The real challenge isn’t getting people to look at data anymore; it’s figuring out how much control we’re comfortable handing over to software.

For routine, low-risk tasks, like updating inventory logs or cleaning up data files, letting an agent run on autopilot is a no-brainer because it cuts out the usual corporate delays. But the second a decision affects a major financial transaction, a sensitive customer complaint, or a legal boundary, you need a human ready to step in and hit the brakes.

At the end of the day, the companies that win with agentic AI won’t just be the ones with the smartest technology. They’ll be the organizations that draw crystal-clear boundaries around where the software can run solo and where human judgment needs to take back the wheel.

You’ve argued that an agentic enterprise requires domain context, governance, and intelligent execution to work together. Why isn’t it enough for companies to simply place an LLM or AI agent on top of their existing data stack, and what architectural changes are required?

Putting an LLM on top of enterprise data is like hiring an intelligent employee and giving them access to every document in the company on day one. They are smart and capable, but they still have no idea how your systems, processes, internal rules, and regulations work.

Businesses face a similar challenge because their environments are naturally complex. Data exists across multiple systems, and business rules have quietly evolved over many years. On top of that, compliance obligations vary by industry and geography. For agents to function effectively, they need context, not just information.

That’s why we think about enterprise AI as a three-layer architecture. The first layer is a trusted data foundation. The second is domain intelligence, which captures business context, relationships, definitions, and workflows. The third is the experience layer and closed-loop execution, where agents can reason and act within governed boundaries.

Many organizations have invested heavily in the first layer, and some have built elements of the second. However, the real opportunity lies in connecting all three.

Without context and governance, even the most capable model can generate plausible answers that simply don’t align with business reality. Enterprise AI doesn’t need more intelligence. It needs more context.

“AI agent” has quickly become one of the most overused terms in the industry. From a technical perspective, what separates a genuinely useful enterprise agent from a chatbot connected to a few tools and workflows?

A genuinely useful enterprise agent is defined less by how smoothly it talks and more by what it can responsibly accomplish. Right now, a lot of what people call “agents” are just chatbots plugged into a couple of APIs. They can answer a basic question or trigger a simple, rigid workflow. That is handy, but it isn’t transformative.

A true enterprise agent doesn’t just follow a script; it understands an end goal, reasons through different options, works across multiple legacy systems, adapts when things change, and stays strictly within your corporate governance lines. I look at it as the difference between a search engine and a trusted operations manager. One hands you information; the other takes responsibility for getting a job done.

To do that, agents need deep domain context. A retail inventory agent, for example, can’t just look at a spreadsheet; it has to understand replenishment cycles, demand patterns, service levels, and margin implications all at once. What separates a real agent from a sophisticated chatbot isn’t the chat interface. It is the domain understanding, the memory, the reasoning, the backend orchestration, and the built-in accountability. Ultimately, businesses don’t need another interface to look at. We need systems that can execute work reliably and at scale.

Tredence places significant emphasis on industry-specific AI, particularly in areas such as retail and consumer goods. As foundation models become more capable, where does domain expertise still create an advantage, and how much of the intelligence ultimately needs to come from understanding the workflows, terminology, constraints, and economics of a particular industry?

Foundation models have become incredibly powerful, but domain expertise remains the difference between raw intelligence and actual usefulness. It is like handing the exact same high-powered medical textbook to a surgeon, a corporate lawyer, and a retail manager. The information in the book is identical, but it is only valuable to the person who actually understands the patient, the operating room, and the stakes of the job. 

Every industry has its own invisible rules, unique economics, strict regulations, and daily operational constraints. In retail, for example, an AI recommendation that technically improves your forecast accuracy but accidentally causes massive out-of-stock situations on shelves isn’t helpful—it’s a liability.

When we apply deep domain intelligence to general-purpose models, the business outcomes are massive. For example, at Tredence, we have deployed domain-specific solutions that delivered incredible, verified results for our clients. A supermarket forecasting program produced close to $200 million in value and pushed replenishment match rates above 90 percent. A manufacturing supply chain platform drove $10 million in first-year savings. A retail workflow modernization reduced analyst effort by 70 percent, while another program achieved a $100 million loss reduction.

As foundation models become a commoditized utility, like electricity, the real competitive advantage shifts entirely toward who has the best proprietary context and workflow understanding to turn generic intelligence into precise business outcomes.

There is a tension between building reusable AI platforms and tailoring systems to the specific data and processes of each enterprise. How do you determine what should be standardized versus customized, and can excessive customization eventually become its own barrier to scaling AI?

Finding the right balance between a standard platform and custom features is always a tough call. Personally, I look at it like urban planning. You obviously have to standardize your roads, power grids, and water lines, but that doesn’t mean every single home in the neighborhood needs to be a carbon copy.

The same rule applies to AI. Things like data security, governance, and monitoring must be completely standardized. They should basically become invisible enterprise infrastructure. To make this efficient, I would recommend companies build an internal marketplace for reusable agents so teams aren’t constantly reinventing the wheel.

Your actual business workflows, however, need to stay flexible, because that is where your unique competitive advantage lives. Right now, the biggest risk isn’t under-customization, it’s over-customization. When every single AI deployment turns into a highly unique, isolated engineering project, scaling becomes a nightmare and maintenance costs skyrocket. The goal should be to build a repeatable capability. You standardise a rock-solid foundation and then give your teams the blocks they need to adapt it quickly to their own needs. The ability to harness and reuse AI capabilities becomes a key differentiator, accelerating adoption and delivering faster time-to-value and stronger ROI.

As agents gain the ability to make decisions and execute tasks, how should enterprises determine where autonomy is appropriate and where human approval should remain mandatory? Are there particular business functions where you believe organizations are moving too quickly or too cautiously?

When it comes to giving AI autonomy, risk should always dictate the rules, not just what the technology is capable of handling. Simply because an agent can execute a process from start to finish doesn’t mean you should cut the safety cord entirely. Routine, everyday tasks like logging inventory or syncing back-office files are perfect for full automation because they eliminate massive operational bottlenecks. But the second a decision involves strict regulations, major financial spending, or high-stakes customer issues, you absolutely need a human in the loop.

Right now, I see companies being way too cautious in supply chain and back-office operations. They are dragging their feet on automating routine procurement and invoice matching, where agents could safely save time today. On the flip side, they are moving way too fast in customer service and legal compliance, rushing autonomous agents into unpredictable customer disputes or automated contract reviews without real guardrails. The goal here is responsible automation, not maximum automation.

As frontier models become cheaper and more capable, the role of AI services companies is likely to change. Does value increasingly shift away from building models toward data architecture, integration, evaluation, orchestration, governance, and redesigning business processes around AI? What will differentiate the AI services firms that remain valuable?

We are witnessing a natural shift in where value is created. A few years ago, raw computing infrastructure was a major competitive advantage. Today, it’s largely a utility, and AI models are quickly heading in the same direction.

Enterprises aren’t searching for bigger models anymore. They want faster decisions, higher productivity, and clear business outcomes. Because of that, the value has moved entirely to the ecosystem surrounding the model—things like clean data architecture, tight governance, smart orchestration, and completely rethinking workflows to make AI actually practical at scale. The winners won’t necessarily be the companies with the most sophisticated technology, but rather those that can turn technology into tangible business results.

We’re also seeing enterprises take a portfolio approach to AI. Frontier models, open-weight models, and smaller specialized models all have a role to play depending on the use case, cost, latency, and governance requirements. As AI matures, success will be determined less by who builds the engine and more by who knows how to put it to work inside a business. What I call “execution rhythm” will ultimately separate the winners from the rest.

Looking several years ahead, if enterprises successfully overcome today’s data, governance, and integration problems, what does an AI-native large company actually look like? Do you expect agents to remain tools used by employees, or will we increasingly see entire workflows and business functions operated by coordinated networks of AI agents?

An AI-native enterprise won’t simply be a traditional company with a few additional software tools. It will be a fundamentally reimagined enterprise. We’ve seen similar shifts during the Internet and digital eras, when technology changed not just how businesses operated, but how they were structured. These companies will function very differently. 

Today, businesses are organized around people performing manual tasks supported by software. In the future, entire workflows may be orchestrated by coordinated networks of AI agents, with humans taking on an increasingly supervisory and strategic role. It’s much like how electricity transformed factories. It didn’t eliminate the factory; it completely changed how it was designed and operated.

In these future companies, agents will continuously monitor operations, flag opportunities, and handle routine execution across different departments. This doesn’t mean humans disappear, our jobs just become far more strategic, shifting entirely to judgment, innovation, and relationship-building. 

The winners won’t be the companies that deploy the most agents. They will be the ones that build the strongest mix of trusted data, domain intelligence, and human oversight. The future isn’t about fully autonomous enterprises; it is about intelligently orchestrated enterprises where humans and agents work together to hit goals that neither could achieve alone.

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