Background

During some of my misadventures at Sachiv (circa mid 2024), at the doorsteps of a well-known VC, I ran into the (extremely brilliant, might I add) founders of a very early stage West Coast based insurtech startup, also seeking their first cheque. As an aside, at the time - I may have, at some point, tried to poach one of them (🫣). While my founder journey saw a co-founder pull out and some of the funding plans crash with it (😔), they've had great success with raising. Later in the year, they gave me an opportunity to do some data science work for them.

The Brief

We started with some sample data for a few hundred customers and assumed claims data. The idea was to build a risk model leveraging a variety of traditional and some novel signals, and come up with a suggested pricing model to ensure motor vehicles (EV's as a starting point). This really fascinated me as I have always been interested in modelling drone and new-image mobility businesses.

I took things a (few) step(s) further and also computed the re-insurance parameters, and the scale of data needed for varied levels of risk tolerance for us as a business, and to develop greater confidence in our bands for a sustained business.

The Output

This was a fantastic opportunity to review Linear Models, Bayesian Models (including SCM's and BN's), tail-dependent risks and fit models with Gradient Boosting techniques over a week. We also went back and forth to generate some synthetic data with monte carlo simulations to test assumptions and variable relationships and distributions. The net outcome was a deeper understanding of the business, our data/funding needs, level of risks involved and the sensitivity to certain parameters.

In my archives are a few thousand lines of Python and R code, and in my phonebook - a couple of really smart people I feel lucky to call friends.