Last week I attended the Deep Learning Finance conference organized by RE•WORK. As the event was smaller compared to many AI conferences I have attended, it had a more casual vibe which allowed for more open conversations about how people are using AI, some of their data challenges, and their overall goals.
The insurance industry is transforming through the use of AI. Huma Lodhi from the Direct Line Group shared the various ways that deep learning in insurance is having a huge impact. Deep learning, for the record, is a subset of machine learning focused on identifying data patterns and classifying information. The insurance industry is data-rich and based on rules that are centuries old. Lodhi explained that most insurance companies have data living in various silos, including text, image, and voice, but by extracting it and integrating it into a data lake, firms can use deep learning to automate processes that have traditionally relied on human involvement.
Direct Line is one of many examples demonstrating the successful integration of deep learning in insurance. The company uses deep learning to improve four key areas of their business:
1. Claims Management
Automating the claims management process with machine learning allows claims to process much faster than with traditional methods. According to a report from April 2018 by McKinsey entitled “Insurance 2030 — The impact of AI on the future of insurance,” machine learning in insurance companies will allow them to reduce headcount aassociated with claims management by 70-90% in 2030, even though it will remain a primary function for carriers. With processing time shrinking from days to hours, insurance firms will increase customer satisfaction while also creating internal efficiency. Some firms are already seeing this in action after deploying chatbots to interact with customers, and using machine learning to assess the total loss and process the claim. In some cases these technologies can remove the need for any human interaction whatsoever.
2. Fraud Detection
Fraud detection is another important application for machine learning in insurance. With the amount of payment channels on the rise leading to rapid growth in the number of overall transactions occurring worldwide, machine learning algorithms are used to develop automated fraud screening systems that are faster and more accurate than systems that rely on transaction rules combined with human reviews. Machine learning distinguishes between normal and fraudulent behavior, and adapts over time based on variations of fraud patterns in the data. This is the true power of deep learning as compared to traditional analytics methods — the ability to detect types of fraud that are similar but not identical to existing patterns, as well as the ability to spot completely new types of fraud altogether.
3. Pricing
Machine learning in insurance companies is also used to assess customer risk when it comes to pricing, as well as optimize price based on customer segments. Algorithms are used to determine cost vs. risk based on the past behavior of each customer segment, which then helps to determine more accurate pricing. Since machine learning models can calculate risk premium estimates in real-time based on large amounts of policyholder data, firms that adopt this technique can offer more competitive pricing to their customers in a fiercely competitive landscape.
4. Customer Analytics
Machine Learning — along with predictive analytics — is also being used to improve retention and reduce churn. Insurance firms use these techniques to identify and optimize customer loyalty behaviors around retention, advocacy, and purchase behavior. From this, they can develop more sophisticated marketing programs with targeted offers and upgrades for high-value customers.
The Importance of a Data Strategy
With the advances in machine learning for insurance purposes, it is imperative that firms create a machine learning strategy to stay competitive. A critical component to this strategy is ensuring that your machine learning algorithms are trained with large volumes of high-quality, human-annotated data. The right training data will ensure that your machine learning-based solutions perform with a much higher level of accuracy so that your firm can more quickly improve internal efficiencies.
