Insider Lessons on Getting AI from POC to Production in Regulated Banking with Sean Harkin

In this episode of Boss N’ Data podcast, Joshua Smith is joined by Sean Harkin, PhD, Data Science & AI Lead at Lloyds Banking Group, for a practical conversation about what it really takes to ship AI in a regulated environment. Sean and his team are putting GenAI and agentic solutions into production for life insurance, going beyond the POC bottleneck that traps so many AI initiatives. You’ll hear👇 How to control LLM hallucination at production quality (ontologies, prompt engineering, LLM-as-a-judge) Why human-in-the-loop is non-negotiable in high-stakes decision recommendation Domain-specific LLMs vs general models with careful prompt engineering Making AI reasoning transparent for regulators, risk and reinsurers The cost-per-token reality check on agentic architectures POC to production: what actually unblocks deployment in a bank Measuring real AI value (and a 20-day → 1-day customer journey result) Chapter Markers👇 00:00 Intro: Josh introduces Sean Harkin (Lloyds Banking Group) 00:01 Sean’s path: research science → risk modelling → data science leadership 00:02 Solving the hallucination problem: ontology, prompt engineering & LLM-as-a-judge 00:05 Human-in-the-loop in underwriting: AI as decision support, not decision-maker 00:07 Domain-specific vs general LLMs: why Gemini + prompt engineering won (for now) 00:10 Explainability for regulators: translating across stakeholders 00:13 Beyond magical thinking: the real timeline for AI value (and the electrification analogy) 00:17 The hidden cost of agentic systems 00:19 POC to production: what actually unblocks deployment 00:22 Measuring AI value (20 days → 1 day customer journey) 00:24 Shipping AI in a regulated environment: model risk, data privacy, cybersecurity, reinsurers 00:29 Data quality, drift & risk management as the same thing as good science 00:32 What banking has absorbed from tech — and what it can’t 00:33 Wrap-up + where to find Sean About the Guest Sean Harkin, PhD, is Data Science & AI Lead at Lloyds Banking Group, working in the insurance arm. He has 18 years’ experience in financial services across startups, banks, and consultancies, with a research science background spanning biochemistry and biophysics and a PhD in applied statistical modelling focused on causal inference. Sean now leads a team of data scientists, ML engineers, and other technical professionals putting GenAI and agentic solutions into production , cutting through the POC bottleneck that holds many firms back. 👉 Connect with Sean on LinkedIn:   / sean-harkin-phd-22776674   About the Host Joshua Smith has been a Data and AI talent specialist since 2011. He is Sales Director at MBN Solutions, a leading Data & AI consultancy building teams across the UK, Europe and the US. Follow MBN Solutions on LinkedIn:   / mbn-solutions   Disclaimer: The views expressed by the guest are their own and do not reflect those of their current or former employers. AI in insurance, life insurance, regulated AI, banking AI, GenAI, agentic AI, LLMs, prompt engineering, RAG, hallucination control, knowledge graph, ontology, model risk, model governance, data quality, model drift, POC to production, MLOps, data leadership, data strategy, customer experience, Lloyds Banking Group, Boss N’ Data Podcast, MBN Solutions