How Genpact is building an autonomous enterprise: Vidya Rao, CITO, on scaling AI from within
Under the leadership of Vidya Rao, Chief Information & Transformation Officer, Genpact has shifted focus from pilots to platforms—re-architecting data foundations, embedding AI directly into business workflows, and putting governance frameworks in place to support speed without sacrificing trust. The effort has helped the company move from dozens of experiments to more than 100 AI agents in production, supporting over 100,000 users and handling millions of interactions annually
Some edited excerpts:
The State of AI in Business 2025 report highlights that less than 20% of organizations have scaled AI beyond pilots, with legacy systems as a primary culprit. Drawing from Genpact’s own internal digital transformation, what are the top three architectural shifts you’ve implemented to dismantle these legacy roadblocks and operationalize AI at the speed of business?
As CITO, my primary focus is to transform Genpact into an autonomous enterprise—a company that operates with minimal manual intervention by integrating AI, data, and technology.
To overcome legacy barriers and scale AI at the speed of business, Genpact has focused on three critical architectural shifts:
1. Modernizing the Data Foundation through Microsoft Fabric
We are moving away from fragmented legacy data estates and building a unified, AI-ready architecture on Microsoft Fabric. Its native LLM-powered functions allow us to automate enrichment tasks—such as sentiment analysis, classification, and entity extraction—with minimal engineering effort. This shift creates a single, intelligent data backbone that dramatically accelerates AI deployment.
2. Democratizing AI Development with the Genpact AgentBuilder
We introduced a low-code/no-code platform powered by reusable accelerators and APIs, enabling teams across functions to rapidly build and scale enterprise-grade AI agents. This reduces dependency on specialized engineering resources and ensures AI solutions can be operationalized directly within business workflows. This is to further strengthen integration by offering a single place to design, build, deploy, and monitor AI agents.
3. Establishing Agentic Interoperability Through MCP
We are exploring the adoption of the Modular Coordination Protocol (MCP) to enable secure, seamless agent-to-agent communication. This foundational step will support dynamic orchestration across multiple AI systems—critical for achieving autonomous, scalable, and business-aligned AI operations.
Enterprises often grapple with fragmented AI experimentation across silos. How has Genpact designed its internal architecture—perhaps through your global ERP program or partner ecosystem collaborations—to foster seamless integration and prevent ‘AI sprawl’ in large, complex organizations?
AI sprawl is one of the most underestimated barriers to enterprise adoption. When employees encounter multiple disconnected tools, inconsistent interfaces, or AI systems that deliver conflicting insights, the user experience deteriorates quickly. Confusion replaces confidence, adoption stalls, and the promise of AI-driven productivity never materializes. Preventing this sprawl requires intentional architectural design—not just more AI.
At Genpact, we’ve addressed this challenge through three strategic interventions:
1. Establishing a Unified Experience Layer with Scout—Our “UI for AI”
Instead of exposing users to a fragmented landscape of agents, Scout serves as the single, intuitive interface for interacting with AI across the enterprise. Whether an employee needs insights from financial systems, or support from HR workflows, Scout orchestrates multiple underlying agents and models, ensuring a consistent, frictionless experience. This approach prevents proliferation of point solutions by centralizing access through one intelligent gateway.
2. Designing Enterprise Platforms to Minimize Sprawl—Guided by Our Global ERP Program.
Our ERP modernization has provided the architectural discipline needed to avoid siloed experimentation. By embedding AI capabilities into process workflows—rather than building AI “around” them—we ensure that use cases scale within enterprise platforms, not outside them. Central data governance, shared integration frameworks, and reusable accelerators allow us to deploy AI in a way that is modular yet coherent, reducing duplication and enabling cross-functional intelligence.
3. Evolving Our System of Engagement Strategy—With Workday (our ERP) as a Critical Anchor
We are rethinking the role of Workday as a core system of engagement to ensure AI does not sprawl across HR and people operations. Instead of disparate HR chatbots or standalone assistants, we are exploring ways of integrating AI natively within Workday workflows while enabling Scout to surface Workday-driven insights and transactions. The goal is to render employee experience with a unified, purposeful set of AI capabilities—consistent, governed, and grounded in enterprise-wide standards.
Genpact’s MyTwin, AgentBuilder, and AI Guru are redefining workflows in finance, talent, IT, and employee services. Can you walk us through a specific ‘before and after’ example where one of these tools transformed a routine process, and what lessons on agentic AI deployment can other enterprises apply?
At Genpact, our focus is on strategically applying AI to deliver tangible value at scale. To achieve this objective, we leverage a range of AI technologies and develop purpose-built AI solutions to address specific needs and drive measurable outcomes. Our approach is deeply integrated, serving both our internal operations and our clients. For example, the Genpact AP Suite combines agentic AI with deep domain knowledge in finance and accounting, allowing us to deliver precision at scale for clients while also making the work more meaningful for our teams. These task-oriented AI agents operate within business context to reimage accounts payable, improve cashflow, and enhance supplier relationships.
MyTwin is another example of what happens when AI becomes part of the daily flow of work. Previously, leaders would spend considerable time preparing for reviews, tracking open actions, or scanning through emails to understand where things stood. MyTwin changes that by using models trained on the user’s digital footprint and by bringing together information from email, calendars, Teams, Salesforce, ServiceNow, and other systems. It highlights where leaders are spending their time, surfaces risks that might otherwise stay buried in inboxes, and helps them quickly regain momentum after time away by listing pending actions and even suggesting appropriate delegation.
Here are some lessons I keep coming back to whenever colleagues at other firms ask how to land agentic-AI programs in the real world.
1. Anchor everything to a bold, measurable “North Star.”
2. Pair the big goal with crystal-clear accountability
3. Governance is the glue—treat it as a product, not paperwork: We instituted an Agent Development Life Cycle (ADLC) with stage gates for data privacy, risk and ROI. The discipline lets teams build safely at speed and gives execs confidence to scale pilots enterprise-wide
4. Lead with an augmentation narrative: From day one we positioned agents as complimentary—“MyTwin” helps, it doesn’t replace. That framing reduced role-anxiety and keeps change conversations constructive
5. Solve upstream process debt, not just surface tasks: Many early proofs failed until we rewired policies and data flows
Your agentic finance solution is driving faster billing cycles and improved visibility. What were the key hurdles in deploying this at scale across Genpact’s client operations, and how did you address data privacy and compliance in a multi-vendor ecosystem?
Enterprises are moving away from siloed applications toward living, learning systems that combine data, AI, and domain expertise. Multi-agent orchestration demands monitored workflows to ensure accuracy and reduce operational drag. We’ve learned that execution should be built around traceability, explainability, and accountability, and outputs need validation against domain policies and risk frameworks. Human oversight is critical to shape objectives, filter inputs, and implement ethical safeguards. For context-aware augmentation rather than unchecked automation, it’s important to invest in systems and teams that can steer model outputs, synthesize diverse insights, and iterate workflows at speed.
Our finance teams are already using the agentic Genpact AP Suite for accounts payable to work faster and with more precision. The biggest hurdle was the variation in underlying processes, which is why clearing legacy debt first was so important. Our Client Zero initiatives gave us proof and perspective from our own transformation, helping us understand what needed to change before agents could deliver value. Strong governance helps ensure that compliance, access control, and auditability are maintained across different environments. With these foundations in place, deploying agentic solutions at scale becomes more predictable and trusted, even in complex, multi-vendor ecosystems.
Genpact’s Client Zero model treats your own operations as the ultimate proving ground for AI transformations. How has this approach accelerated Genpact’s shift to intelligent ecosystems across core functions, and what advice would you give CIOs wary of ‘eating their own dog food’ in legacy-heavy environments?
Our Client Zero model has accelerated our transformation because we use our own operations as the test bed for everything we build. By applying AI first principles internally across HR, finance, IT, and employee services, we gain rapid feedback, learn from what does not work, and refine solutions before deploying them externally. This gives us credibility with clients and builds internal confidence, because teams see the benefits firsthand.
By becoming Client Zero, we scaled from 20 pilot agents to 100+ in production, now handling 5 million interactions a year and serving more than 100,000 active AI users. This drove enterprise-wide productivity gains, freeing up leadership hours every month and cutting G&A even as functions continued to grow. Process velocity improved across the board—billing cycles accelerated by 25%, HR leaders now pull live analytics instantly instead of using spreadsheets, and IT reduced incident detection and resolution times dramatically. Culturally, AI became embedded in the flow of work for 140,000+ employees, while citizen developers began building agents in hours, compressing idea-to-impact dramatically. The result: a living proof point that intelligent ecosystems unlock systemic value, turning legacy complexity into a competitive advantage and building credibility My advice to CIOs who hesitate to apply AI internally is that perfection should not be the starting point. We learn far more by implementing AI within our own operations, addressing the shortcomings head-on and scaling once the foundation is ready.
As Chief Information and Transformation Officer, you’ve emphasized reimagining tools and infrastructure with an AI-first lens. How does Client Zero integrate with your world-class data office to turn siloed operations into predictive, self-optimizing ecosystems—and what’s one pivot you’ve made based on internal learnings?
Client Zero and our Data Office are tightly connected because our biggest early challenge wasn’t a lack of AI tools—it was data scattered across transactional systems, dashboards, and spreadsheets. That fragmentation slowed decisions and limited our ability to operate with confidence.
By unifying this information through a standard business data model and enabling teams to “chat with their data” via intelligent data agents, we moved from manual extraction to real-time, conversational insights. This has turned previously siloed operations into more predictive, interconnected, and self-optimizing ecosystems.
One key pivot we made from these internal learnings was recognizing that eliminating data debt had to come before scaling AI. Once we established a clean, consistent semantic layer, every function—from Finance to HR—saw immediate gains in speed, accuracy, and responsiveness.
This reinforced a core principle of our AI-first strategy: empower human judgment with instant, contextual intelligence, and make the entire enterprise more agile and future-ready.
Looking ahead to 2026, beyond the current 20% scaling benchmark, what emerging trends—like agent orchestration or decentralized compute—do you see as game-changers for enterprises? How is Genpact positioning itself for the future?
With 2026 around the corner, two trends will fundamentally reshape enterprise AI: agent orchestration at scale—where networks of agents collaborate to run end-to-end workflows—and decentralized compute, which brings intelligence closer to where work happens for greater speed and resilience.
Genpact is preparing for this shift by expanding our agentic solutions portfolio and designing architectures that support distributed, interoperable intelligence. Our standardized data foundation ensures agents operate on trusted, consistent information—essential for orchestration at scale.
We’re also investing heavily in talent, with over two million hours of AI learning completed, so our teams can build, govern, and continually evolve these next-generation enterprise agents.
These moves position Genpact to help clients move confidently into a future of intelligent, increasingly autonomous operations.