Episode Details

In Episode 22 of Season 7 of Driven by Data: The Podcast, Kyle Winterbottom is joined by Chris Pearce, Chief Data and AI Officer at Ageas, where they discuss why so few organisations manage to get AI out of proof-of-concept and into live production. Chris makes the case that this is a structural and operating model problem rather than a technical one, and that the businesses which crack it are the ones that understand their own commercial engine intimately enough to know exactly which decision they are trying to change.

Drawing on a 250-person function spanning data engineering, data science, AI engineering, infrastructure and governance, Chris walks through real deployments into Ageas's contact centres, how the value of those deployments is measured and attributed to the P&L, and why the risk conversation with a board is far more winnable than most data leaders assume.

They also discuss:

• Why rolling out Copilot licences bears no resemblance to putting LLMs into front-end production systems touching customers in real time.
• What the full cross-functional cast actually looks like, from SRE and infrastructure to UX, middleware developers, AI engineers, business SMEs, risk, legal and compliance.
• Why AI delivery is fundamentally a structural problem, with the necessary skill sets scattered across different leaders, agendas and backlogs.
• How building AI capability in isolated pockets of the ecosystem guarantees you never leave POC land.
• Why the first question on any piece of data science work should be how you intend to measure it, and why nothing starts until that's answered.
• How Ageas used LLM summarisation at the chatbot-to-agent handover to remove friction for customers already losing patience.
• Why after-call work was worth attacking, and what shaving minutes off every call does to backlogs, concurrency and demand.
• How A/B testing capability across 50 agents against another 50, de-biased for tenure and experience, produces evidence a board can't argue with.
• What it takes to build a genuine culture of experimentation in an environment as dynamic as a contact centre.
• Why "my job is to help people" is where most value conversations begin, and how to move past it.
• How to trace the decision chain that follows once the phone goes down, and why that's where the financial link is found.
• Why brilliant technical analytics is squandered without the work of presenting it visually and narratively.
• What has to be true for a change in decision-making to be logged, monitored and made someone's accountability.
• Why any organisation asking for an AI strategy should be asked about its business strategy first.
• How to uncover a business strategy that isn't written on a wall or neatly captured in a PDF anywhere.
• Why starting with low-hanging fruit builds the patterns, the track record and the appetite for bigger bets later.
• What the doom loop of perpetual proof-of-concept does to credibility, investment and the perception of ROI.
• Why perfect temples of data platforms get built over four or five years and then fail to land.
• How the risk conversation changes when you demonstrate the operational, technical and information security controls that already exist.
• Why hallucination rates deserve to be compared with how often humans under pressure get things slightly wrong.
• What is missing from every AI maturity framework Chris has encountered, and why counting models in production is activity rather than maturity.
• Why software development skills are becoming essential for data scientists, and how AI engineering mirrors the data science unicorn boom of fifteen years ago.
• Why the technical barrier to entry has never been lower, and why adaptability is now the trait Chris values most.
• Why every practitioner needs a degree of commercial nous, and what happe