AI & agentic products
Product vision, orchestration, human/agent interaction, controls, evals, guardrails, and productionisation.
AI that ships, not AI that demos.
I take agentic AI from pilot to production with insurers and pension providers: operating models, controls and products that self improve and evolve continuously.
Applied AI Head: UK insurance analysis and product management at a global technology services firm, Edinburgh.
- 15 years in insurance, pensions and wealth
- 400 schemes · 70% of pension calculations automated
- Live underwriting workflow · risk-function sign-off
- Defect rate down 25% across five delivery teams
- £1M premium in year one · blockchain insurance product to market
Where I operate
Insurance
London Market, broking, underwriting, treaty reinsurance, and insurance operating models.
Pensions & wealth
Retirement journeys, guidance and advice, cash-flow modelling, and complex regulated product propositions.
Product & transformation
Product strategy, discovery, operating-model design, business architecture, and multi-team delivery.
Case record
Agentic broking operating model, design to pilot
AI product lead
An agentic operating model for broking: specialised agents on domain-driven boundaries handling submission triage, market matching, and follow-up, with evaluations, guardrails, and explicit human decision points designed into the workflow rather than bolted on.
A production-grade platform design proven against real placement workflows. Client and outcome details are confidential – available in conversation.
AI-enabled underwriting workflow
Product lead
An AI-enabled underwriting workflow: agents prepare, enrich, and pre-assess submissions inside hard domain boundaries; underwriters keep the decision, with the system’s reasoning and evidence attached to every recommendation.
Adopted into the live underwriting process with the risk function’s sign-off. Volume and turnaround figures are confidential – available on request.
Pension calculation engine
Product lead
An AI engine that classifies each calculation against the scheme’s rule base, executes the routine cases automatically, and routes genuine edge cases to a human with its reasoning attached.
Automated 70% of pension calculation routines across 400 schemes. The remaining 30% now arrive at analysts pre-triaged.
Enterprise transformation
Product and delivery lead
A coordination layer of shared quality gates and a single defect taxonomy, so problems surfaced once and were owned once.
Defect rate fell 25% over the programme while delivery pace held.
Ideas I'm working on
From workflows to agentic operating models
Automating a workflow keeps the org chart; giving agents work changes it. What that means for how insurers design operations.
Why domain boundaries matter more when agents can act
DDD was good architecture advice. Once software can act, it becomes risk management.
Human approval isn’t an AI strategy
A human in the loop is a control, not a design. What meaningful oversight of agentic systems actually requires.
What insurance product teams need to understand about agentic systems
The gap between an AI pilot and an agentic product, seen from the product seat rather than the lab.
Deciding what an agent may do without a human in your business?
I have taken that decision through a risk function and into production. Tell me the workflow and the control that worries you, and I will reply within two working days.