Using machine learning and external data to improve underwriting decisions

Underwriters often receive incomplete submissions, making it difficult for them to make confident and efficient pricing decisions. External data and machine learning offer insurers a significant opportunity to optimise the underwriting process, driving down loss and expense ratios.


In a rate depressed environment, the time taken to quote matters more than ever. When an underwriter receives an application for insurance they usually have a small window of time to beat their competition in answering one fundamental question; should we write a policy for this particular customer or exclude them from our book? 

To maintain efficiency and keep underwriting expenses down, the traditional approach is to create a set of risk selection criteria - a set of rules and conditions beyond which the risk will not be written. Based on certain attributes or ‘red flags’ in the application, the underwriter will quickly reject or accept a risk, and rules and conditions are adjusted over time based on experience. 

Risk selection is a delicate balancing act. Accepting risks indiscriminately will lead to increasing loss ratios. Alternatively, being too selective can limit growth and lead to declining premiums. 


There are two key issues with the traditional approach to risk selection. The first is that risk selection criteria is often derived from subjective claims experience and only updated when a loss occurs, making it an inaccurate proxy for any risks - future or current - that are not identical to what has been written in the past. 

The second problem is that even the most basic risk selection criteria can be difficult to verify if a submission is incomplete. Both direct and broker submissions often lack the level of detail required to make an accurate pricing decision. 

The absence of granular, up-to-date information not only makes it difficult for underwriters to distinguish between attractive and unattractive risks, it makes it difficult for insurers to define them in the first place.


Gaining access to accurate, high-resolution information at the point of underwriting is critical in driving down loss ratios and underwriting spend. Cytora uses machine learning algorithms to continuously extract and analyse billions of data points, enabling insurers to manage loss ratios driven by a population-scale view of commercial risk. Crucially this data reflects risk in the present and is not reliant on experience from the past.

Based on a combination of loss datasets, total potential insured data from external sources and predictive rating factors, the Cytora Risk Engine is able to rank every available risk in the market according to loss ratio and pre-compute a technical price for each risk.

Cytora eliminates total reliance on information from brokers and application forms enabling confident, efficient and accurate underwriting decisions. To learn more please contact us.