Now: SVP, Credit Reinsurance AI · Marsh ReLondon · Credit · Reinsurance · AI
I build the AI that prices credit risk. I lead AI for Credit, Bond and Political Risk at Marsh Re, and I am often asked for a view on enterprise AI, data platforms, and credit and surety.
All layers placed · Structure the uncertainty.
Every layer of a reinsurance tower has a machine learning twin.
Hover a layer to see the pairing.
Live · Open in your browser · No sign-up
Live apps, designed and built end to end outside the day job: a creative studio for seeing risk differently, a private record-matching tool, and pricing engines for credit and creator risk. Click in and explore.
A growing collection of experiments for seeing risk differently: a Contagion Globe where credit risk travels along real trade routes, games that unlock the maths, crisis replays, and an AI you can talk to.
Open the studio
Company name matching and deduplication that shows its working. Links records across systems when names and formats disagree, explains every match, and runs entirely in your browser.
Open MatchFlow
Treaty, facultative and SRT pricing on one structural engine. 10,000 simulated years, 1-in-200 tail capital, and every number traced back to the same simulated economy.
Launch the pricing engine
The first quantitative pricing engine for creator income risk, a risk no insurer prices today. Catastrophe-style hazards, alternative-data vulnerability, live sentiment shocks, Monte Carlo capital and jump-diffusion.
Launch the risk enginePersonal projects. These apps are experiments I designed and coded myself, in my own time, on openly published methods such as the Merton structural model, Monte Carlo simulation and classic record linkage, and on public data feeds. They ship with synthetic or publicly available data only and are not connected to, derived from or endorsed by my employer or its clients. How they work
Structure the uncertainty.
Banu Munnangi
Expertise
People bring me the questions that sit between technology and risk. My answers come from deployment, not theory.
How large organisations adopt, govern and pay for AI: LLM adoption, hyperscaler versus direct model access, copilot tiering and AI pricing.
Snowflake versus Databricks from the customer’s seat, and how Unity Catalog, Genie and Palantir Foundry compare in production.
Treaty and single risk pricing, surety, SRT, portfolio capital, and where AI genuinely changes underwriting.
Experience
Speaking
Talks on AI in reinsurance, agentic systems and the future of underwriting.
Everyone is investing in AI. Far fewer can answer the question the CFO is asking: what did we actually get for it? A practical framework for measuring AI return in insurance, from hard cost savings and automation to decision quality and speed to quote.
Generative AI is not an efficiency tool for insurance, it is a structural shift. From submission triage to treaty analysis, the underwriting workflow of 2030 will look fundamentally different. The question is how fast, and how safely.
Credit insurance is both a financial risk product and a leading indicator of economic stress. How early-warning models can help primary credit insurers underwrite through the cycle rather than react to it.
Writing
Two probabilities of default for the same company, both correct. A one-factor model connects them, and calibrating it properly is where the real work is.
Read Credit risk maths · 5 min readFour numbers that all get called “the PD”. They answer different questions, and mixing them up quietly misprices every multi-year risk.
Read Reinsurance structuring · 5 min readEvery reinsurance programme trades off cost, capital relief and exposure relief. Getting two is easy. A broker earns their fee in the centre of the triangle.
Read