AI - Changing the future of insurance

Date16th September 2021

More insurers are recognising they need the competitive edge that AI provides

Artificial Intelligence (AI) is unquestionably a hot topic for 2021, but in reality, it is a concept that has been around since the 1950s and the dawn of computing. More recently, AI has broken into popular culture and into all aspects of business – not least insurance.

More and more insurers are recognising they need the competitive edge that technology like AI provides. At Finity, we are seeing more of our clients’ work involve designing AI based solutions to help solve business problems. So, what has led to the rise of AI over the past few years?

AI trends in insurance

The single key trend that underpins growth in AI is the exponential increase in the volume of digital data that has been stored over the past two decades*. Not only are we storing more data than ever before, but the diversity of data that we are capturing and digitising has expanded. Gone are the days of the only 'useable' data being that which is stored in database tables (tabular data). More and more of the data that organisations are leveraging is natural language data (documents, emails, transcribed phone conversations) as well as image data (photos, videos).

As insurance professionals we know the importance and power of data – and this trend isn’t going to slow. So how are we going to capitalise on the opportunities that this rise in data volume and diversity offers us?

The right tools for the job

The rise in these three very different types of data - tabular, language and image - pose different challenges to insurers in terms of extracting valuable insights, and we've seen three distinct sub-fields of AI grow in response.

Predictive modelling

Predictive modelling is at the heart of insurance - we want to use information that we have collected to make predictions about future or unknown events (typically, claims). We’ve been doing this since the very first days of insurance, an almost always using standard tabular data. All that has changed is the amount of data we have collected, (enabled by the rise of digital technology), which we can now store as electronic versions and accelerated by the tools to analyse and make sense of that data (computers as well as the programs/algorithms that they use). As a side note, sometimes the term ‘predictive modelling’ is used interchangeably with ‘machine learning’. They are not the same thing, but often machine learning tools are used to assist with some predictive modelling tasks.

Claims fraud detection is a challenge that many insurers have to deal with and as a result anomaly detection is an AI sub-field that Finity has been increasingly utilising to help our clients. This still is ‘predictive modelling’, where we try to find ‘patterns’ in data and ‘predict something’ but instead of predicting the cost of insurance claims or likelihood of a customer renewing we are predicting the likelihood of a ‘data point’ (claims, quote, policy, customer etc) being ‘unusual’ or ‘suspicious’.

Natural language processing

At Finity, we are seeing more and more language data being recorded in insurance in phone calls, chatbot history, emails, and documents of all kinds. Also, customers are expecting a more ‘conversational’ style of interaction – less so about filling out forms or scrolling through a list of FAQs on a website, and more about question and answering, having a discussion with someone, or asking questions to a chatbot directly and receiving a well thought out response. As such, techniques to process language data are becoming critical to our insurance toolkit in order to understand vast swaths of text data to find and utilise patterns to help us understand and respond to our customers better.

Computer vison

The final of the three subcategories of AI that is becoming increasingly relevant to insurance is computer vision. This is effectively, the ‘technology of tomorrow’, however an area where insurers are only currently dipping their toes into the water. Computer vision covers that range of techniques and approaches for dealing with image data. The amount of image data that we have access to has increased substantially over the last decade, mainly due to the fact that we all carry with us high definition cameras in our pockets making it easy to upload photos and videos. Some examples in insurance is automatic claims assessment for vehicle damage. This has come a long way in the last 2-3 years – but is still far from perfect. Further improvements are likely in the next 5 years, however as more ‘training data’ becomes available the accuracy of these algorithms will likely improve.

The industry's use of AI today

To illustrate just how far we have come, it’s interesting to note that many insurers are using machine learning techniques that many would have considered to be at the cutting edge of AI only five years ago (such as GBMs, model ensembles and natural language processing models). Just like the chess playing programs of the 1990s, these are becoming more and more of the status quo.

Insurers are also making encouraging strides regarding the storage and organisation of data. We have seen a move away from storing data in legacy systems and into more flexible ‘data warehouses’ that allow for easier extracting, transforming and loading of data for reporting and analytics, instead of trying to work with existing monoliths.

Moving onto a more advanced application of AI - The Artificial Immune System (AIS) is a relatively new software product developed at Finity by our AI team. The AIS uses a novel artificial network to model the joint probability distribution of a set of data. This can be applied to a wide range of industries including financial services, insurance, digital advertising and cyber security - to name a few. The network becomes the backbone of an intelligent system able to classify data appropriately identifying issues like insurance fraud, claims leakage, and abnormal customer behaviour. This is a specific application of anomaly detection systems (a sub-field of machine learning) – which we think are going to be used more frequently as ‘firewalls’ to monitor data that enters insurers’ systems from all areas of the value chain to detect and respond to data in real time to reduce risk and ultimately improve customer experience.

So, what’s next ?

Rapid advancement in AI technologies over the next decade will likely lead to disruptive changes in the insurance industry. AI is the critical tool that insurers will need to have in their toolkit to make sense of the vast amount of data in a digital world. Those businesses which will thrive will be the insurers that embrace and utilise AI to create their new products, streamline processes, lower costs and exceed customer expectations.

As you look around at the growing volume and diversity of data you are collecting, how will you use the recent advances in AI technology to solve your clients' problems and grow your business?

Learn more about Finity products and AI capabilities here.

References

* In 2020, one estimate of the amount of data stored in the digital universe globally was 44 zettabytes. (If you're like me and don’t know how many bytes that is, its approximately 40x the number of stars in the observable universe, or 4.4 followed by 22 zeroes. A lot.)