We use cookies to improve this site and enable full functionality. You can change your cookie settings at any time using your browser. Our cookie policy.

Breaking through the issues preventing AI adoption in insurance

Breaking through the issues preventing AI adoption in insurance | Insurance Business

Breaking through the issues preventing AI adoption in insurance

Advanced analytics has always been the cornerstone that the insurance industry was built upon. Maths and statistics have long driven the industry’s ability to calculate and effectively transfer risk. The industry has always looked for, or developed, more effective analytic techniques. However, with the sudden and comprehensive development of so many different advanced analytic methods, new techniques and methods are actually creating a higher barrier to entry.

Traditionally insurance professionals have found machine learning unfamiliar, difficult to explain and complicated to implement. Data science requires specialist knowledge that cannot easily be self-taught, and insurance companies have struggled to build in-house data science organisations as a result. The process requires actuaries and quantitative analysts to go ‘back to school’ to learn new methods in automated statistical analysis, programming techniques to capture and re-organise large volumes of data and new technologies. 

With scarce resources and a very slowly evolving technical infrastructure, insurance companies find it hard to hire enough trained data scientists to harness the full potential that machine learning offers. The end result is that for every success a data science team achieved, there were as many as 10 unmet requests. Fortunately, the desire to become an AI-driven enterprise with a clear understanding of how machine learning can support a data-driven business, has won through in many cases.

AI, without doubt, has the potential to help underwriters better understand their customers, improve risk differentiation (low from high risk and the many levels in between) and establish substantially more accurate pricing. In fact, those who cannot differentiate risk levels, will unfortunately find themselves subject to adverse selection - where they win clients who are likely to be unprofitable, and at the same time lose the profitable ones. 

The development of Automated Machine Learning (Auto ML) has emerged as a much-needed solution here.  Replacing the first generation of machine learning, Auto ML uses artificial intelligence to automate each step; incorporating guidelines and best practices to ensure results are consistent and accurate. This new level of automation brings a further unexpected benefit - actuaries, business analysts and IT staff can use automated machine learning with only a few days of training.  The lowered technical barriers enable business domain specialists to participate and significantly increases the organisation’s capacity to solve problems. It is this democratisation of machine learning that will ultimately make the most significant contribution to improved decision making in the insurance industry.

Implementation becomes the next challenge but adopting a straight through deployment process can eliminate months of IT effort to rebuild models, create additional programs to organise and prepare data. It can simplify the integration of machine learning models into existing business applications, such as underwriting systems, policy administration systems and claims management systems. 

Maintain a clear and determined will to drive change, and the benefits of machine learning will not only increase operational effectiveness, it will enhance the overall customer experience and ultimately improve the bottom line.

The above article was an opinion piece written by Neal Silbert (pictured), Insurance GM at DataRobot. The views expressed within the article are not necessarily those of Insurance Business.