The biggest barrier to enterprise AI adoption is no longer model capability but access to trusted, well-governed data that can provide context, according to Joydeep Mukherjee, executive vice-president and global head, data & analytics, digital and creative services, Infosys. In an interaction with Poulomi Chatterjee, he spoke about AI-led compression of analytics timelines, the need for a value office to track returns on AI investments and why security cannot be bolted on after AI deployment. Excerpts:
Q. What can enterprises do to prepare their data for AI adoption?
Data preparedness has become a business imperative. Our Infosys Knowledge Institute research found only 13% of organisations are confident in their data foundations. The gating factor for AI at scale is not model capability but the richness, reliability, trust and real-time availability of context.
Enterprises need accessible, well-governed data foundations to move beyond pilots. We are seeing strong demand for data modernisation and AI-readiness projects as companies explore AI agents and agentic workflows, whose effectiveness depends on high-quality enterprise data and clear governance.
Q. How is AI changing traditional data analytics processes?
AI is compressing several stages of the analytics lifecycle that traditionally involved significant manual effort, including data discovery, pipeline creation, documentation, insight generation and reporting. Enterprise-grade LLMs can automate metadata generation, summarisation and natural-language analytics.
Infosys Topaz Fabric, for instance, connects fragmented enterprise data and creates governed, AI-ready context by harvesting, unifying, relating, contextualising, publishing and operating data. This is making analytics more responsive to business requirements.
Q. Which enterprise workflows are best suited to AI agents?
AI agents deliver the most value when embedded in business processes and operating on trusted enterprise data. Use cases include claims processing in insurance, customer onboarding and fraud management in banking, supply-chain exception handling in manufacturing, and price-pack optimisation in consumer goods.
Our Autonomous Data and AI Estate Solution applies this to real-time intelligence across plant floors and supply chains. Agents can gather information, recommend actions and execute routine tasks while employees retain oversight.
Q. How important is an enterprise-wide data and AI strategy?
Organisations need to treat data and AI as an enterprise transformation agenda rather than a collection of individual initiatives. A strategy should align business priorities, data investments, modernisation, LLM and agentic AI adoption and governance around common outcomes.
It is equally important to establish a value office that links every AI initiative to a business metric such as revenue, cost or productivity before funding it, and tracks value realisation as it scales. This makes AI an accountable P&L conversation rather than simply a technology expense.
Q. What should companies prioritise to prevent AI-led cyberattacks?
Cybersecurity must be part of a broader data and AI governance framework. As AI systems consume more sensitive data and become more autonomous, attacks can also become faster and harder to detect. The five cornerstones are trust, ethics, privacy, security and compliance.
Our research found 86% of enterprises expect heightened AI risks, while only 2% meet responsible AI gold standards. Governance built into the data and AI architecture is therefore more effective than security measures bolted on later.
Q. How will Human + AI models improve outcomes?
The strongest deployments augment rather than replace human judgment. AI can process large volumes of data and identify patterns quickly, while people provide business context, domain expertise and ethical judgment. In one engagement with a large US technology company, AI agents reduced root-cause analysis from five days to one day.
Engineers then applied their judgment to interpret the findings and decide on action. Agentic AI can narrow the signal quickly, allowing people to focus on higher-value decisions while remaining in the loop on consequential outcomes.