Interviewer: Hello and welcome to The AI Summit London. Today we're joined by Ankit Bharadwaj, Head of Responsible AI at Liberty Specialty Markets. How did you land at Liberty Specialty Markets, and could you share a bit of your background?

Ankit: I’ve worked in AI for almost a decade in different roles. I was working with Liberty for a period, this opportunity came up at the right time, and I went for it. I’ve been here a few months and I’m really enjoying it. Working in AI is incredibly topical right now.

Interviewer: There’s a perceived value gap in AI investments. Many organisations have invested heavily, yet leaders question the return. Why is there such a gap?

Ankit: Many organisations treat AI primarily as a tool or technology, when a large part of success is about operating model and change, how people will actually use it. Leaders often push for more AI, but you also need bottom-up pull by bringing teams along on the journey. AI metrics are frequently framed as usage, how many people are using it, when the conversation needs to shift from usage to value: measurable outcomes tied to the work.

The second issue is failing to reimagine workflows. You can’t just bolt AI onto broken processes or poor data and expect good outcomes. Think in an AI-native way, redesign processes end to end to unlock fundamentally different value.

Interviewer: There’s a common perception that governance slows innovation. Your work suggests otherwise. Can you shed some light on that?

Ankit: Governance is often seen as the function that says “no,” but it should enable innovation responsibly. The shift is toward being risk-aware, not risk-averse. Use a risk matrix that classifies AI use cases as low, medium, or high risk and apply proportionate controls. You don’t treat a low-risk use case the same way as a high-risk one. That means flexible review methods and accountabilities, tailored to risk. That’s the change we’re driving, governance that is adaptive and accelerates responsible adoption.

Interviewer: A big challenge is moving from promising experiments to scalable, production-ready AI. Why do so many initiatives get stuck in pilot, and how can teams bridge the gap?

Ankit: The hype cycle pushed many to spin up in-house LLM experiments quickly. Now there’s a realisation of real costs, tokens and infrastructure, and the focus has shifted to business value: cost savings, revenue impact, and ROI. Organisations are learning to track AI usage economics and wrap them with governance and operating model changes.

To scale, combine three things: maintain an ideation engine, make governance flexible and proportionate, and prototype fast. Traditional approaches, long requirements, big vendor projects, then delivery, are too slow for AI’s pace. Prototype quickly, deliver quickly, bring people on the journey, and keep iterating.

Interviewer: Even the most advanced tools fail if people don’t trust or adopt them. What drives adoption and builds trust?

Ankit: Transparency is key. Be clear about where AI is used, how data is used, and how AI informs decisions. That applies to customers and to employees. Then, upskill users so they understand where AI adds value, how to interpret AI outputs, and how those outputs affect their decisions end to end. Data quality is also fundamental: poor data leads to poor outputs. Combine ongoing communication, change management, and data discipline to build durable trust.

Interviewer: Your session focused on how persistent memory can improve AI reliability and decision quality. How does organisational memory enhance AI trustworthiness and utility, and what roles do governance and privacy play?

Ankit: In the late 2010s, data was the moat. In enterprise AI, memory is becoming the moat, especially with agentic AI. Agentic systems depend on how memory is used and continuously updated, and how feedback from outcomes loops back into memory to improve the next iteration of products and interactions.

Organisations need to design memory intentionally: how it’s structured and stored, how it’s versioned and time-scoped, when it’s relevant, and how it delivers time-to-value. Governance and privacy must be embedded—controls from functions like privacy, compliance, and security determine what is retained, for how long, and for which purposes. Memory is both an innovation engine and a governance challenge; the conversation needs to cover both sides of that coin.

Interviewer: Beyond governance, what should leaders prepare for next to stay ahead?

Ankit: Technology is moving faster than organisational change. We’re seeing trends like sovereign AI, embodied or “physical” AI that understands real environments, and the mainstreaming of agentic AI. On the organisational side, the priority is operating model change and process reimagination. If you get the design and the people aspects right, you can capitalise on these technology shifts far more effectively.

Closing

Ankit Bharadwaj is Head of Responsible AI at Liberty Specialty Markets, where he leads responsible AI adoption across international insurance markets. With over 15 years’ experience across AI, automation, and technology transformation, he helps organisations move from experimentation to trusted, scalable AI. 

Prior to Liberty, he was AI Practice Lead at Gate One Consulting, advising senior leaders on AI strategy, transformation, and governance across financial services, retail, consumer health, and the public sector—enabling responsible innovation by embedding accountability and human oversight. He holds an MSc in Computing from Imperial College London and is a founding member of the university’s AI Network.

Ankit's session at The AI Summit London focused on: Memory at Scale — Building AI That Remembers, Learns & Evolves.