I’m the portfolio director for Ki and I’ve been asked to share some of my thoughts on how our algorithmic approach could have an impact on Lloyd’s of London market. For those who are unfamiliar with Ki, we are an algorithmic digital syndicate operating within the Lloyd’s of London market (https://www.ki-insurance.com/ ). 

I’ll discuss how: 

• We are taking a data-first approach to understanding risk, digitising the decisioning processes for follow underwriting whilst circumventing some of the unconscious biases and behaviours exhibited in traditional syndicates. 
• We are building something that is fast and efficient, even at scale. This means we are building a software company as much as an insurance company. 

This post introduces these topics in a bit more detail – and we’ll go into more depth in future articles. 

A data-first (or data-only!) approach to Risk

A contextualised data asset

Data as a means of understanding risk is pretty old-school in insurance terms, with the fact that the industry pioneered this way of thinking being well-documented. These approaches have improved over the course of the past few years, and the proliferation of open source software has meant that the ability to experiment with the cutting edge of data science, machine learning, deep learning (or whatever the buzzword du jour is) becoming increasingly easier and less of a competitive advantage. 

Of course, the precise techniques and approaches we use are important, but in our mind the way we create value is combining these, increasingly available techniques with: 

• Our access to Brit’s data asset of 20 years’ worth of policy and claims data across a variety of classes of business, coupled with the data we have been collating since we went live at the end of 2020. 
• Deep experience of the Lloyd’s of London market. It would be naïve to think we could just throw an open source algorithm at a dataset and come up with an end to end solution. The reality is much messier and we have worked, and continue to work, extensively to contextualise the data asset we have and string together an algorithm, often leveraging open source components, which will actually work in the market. 

We couple this specialty insurance knowledge with a healthy dose of challenge from other industries – drawing inspiration for our design from industries as varied as quantitative finance to autonomous vehicles. 

Data first

Ki’s quotes are processed by our algorithm and this offers a few advantages in comparison with a more traditional syndicate of underwriting teams: 

• First, it is much easier to take decisions at a portfolio level – because at any point in time we have full visibility of what has been written and quoted to our partner brokers. There are no weekly forums or conversations that need to happen – the Ki algo has full visibility of our portfolio at any point in time – whether it contains 1 risk or a million, making it easier for us to automatically think of things like portfolio balance. 
• The Ki algo isn’t susceptible to hopes, dreams, fears and doubts that single individuals might have when they are quoting business and taking on risk – we can be as bullish as the data tells us to be, without the human discomfort of career longevity worries when taking on bigger risks, or premium targets when reducing our line size. 

From the very first release of Ki, we have thought of it is as a data system first and foremost. This means it is easier for us to ingest new data sources in a way which can directly influence the way we quote and write business, allowing us to increase the breadth of data we have beyond the data asset we have from Brit and are building within Ki. 

Of course, it is not all digitally perfect today, a data first approach introduces its own challenges – whilst it is easy for our algorithm to synthesise large volumes of data – it is tricky for it to consume unstructured information. Given this is still quite endemic within the market this is something we are working on with our colleagues in Brit’s Innovation team. 

Whilst we aim to be data-first, we are not data-only and for specific classes of business we ensure our algorithm has a layer of human oversight, something often referred to as “human in the loop”. 

Scale

Scale – digital products scale easily

Our data first approach is only half the reason Ki’s algorithm is critical to our commercial success. The second is the fact that, being a digital component, it can scale easily. The incredible premium upside of going from 100 to 100,000 quotes comes at a negligible cost to Ki, meaning our profitability automatically improves with scale (all else being equal). A digital platform also means that, as we grow the number of users leveraging Ki and the product set we offer, the journey and experience remain as seamless and fast for brokers using our system. This is currently a 10 second quote journey for our brokers, a far cry from the typical 2 or 3 week lead time for getting quotes in the market. 

We’re forcing the insurance world to think about (software) product development

There is a world of difference between a data scientist experiment with a machine learning approach on their desktop, and creating a production grade algorithm that can operate and scale in the way we have just described. 

Doing this requires us to think about Ki as a software product for insurance – much like Spotify is a product for listening to music or Instagram is a product to share photos and videos (and adverts). Thinking about what we are doing in the context of a data driven software product is arguably as big a step-change for the Lloyd’s of London market as our algorithmic approach. If you look at any decent insurance textbook, machine learning is a natural progression to power actuarial processes, but software development with the end to end scale and mindset that we are undertaking at Ki has arguably never really been seen as a focus in the London insurance market. 

In most places, the emphasis is on the front office underwriters who meet the brokers and write the risks. IT and tech are seen as back office functions keeping the business moving but not a source of true value or growth. 

As we build out the algorithm we see these 2 worlds much more intimately connected. Our underwriting approach is powered by technology, and a sound understanding of underwriting and the market drives the design of tech. The only way to do this is to put the two aspects on an equal footing, thinking about our overall product, with its insurance and digital aspects as a single thing – a well trodden approach in the digital world (with hundreds of books on the subject written), but certainly less common within the Lloyd’s of London market. Our algorithm team consists of actuaries, data scientists and engineers – and we speak with our Portfolio Underwriting team on a daily basis, valuing the communication and learning that comes via osmosis when we work in such a multidisciplinary fashion. This is especially important as there are very few individuals who simultaneously have both deep specialty insurance market knowledge as well as experience building out software products. 

Feel free to reach out if you want to discuss and I’m looking forward to exploring these topics in more depth with you.