The insurance industry has been using insights and analytics as key competitive strengths. It has also seen significant success in building analytics capabilities over the past few years. However, the industry has been slow to adopt machine learning (ML) in an era where organizations from other domains are increasingly collaborating with machines to build better business models, mitigate risk, enhance competence, and bring sharper competitive benefits.
In essence, ML is an application of artificial intelligence (AI) that involves algorithms using historical data to predict current or future outcomes. As an example, banks use ML algorithms to detect fraud or other inconsistent activity on customers’ credit and debit cards.
Machine learning will also transform the way insurance companies do business. Traditionally, the industry has been following processes through which it:
Identifies a pool of customers who can be assessed for risks
Targets these customers and assesses the risk for each sub-category
Sells differently priced policies spreading risk across the entire pool of buyers
Retains buyers by keeping prices low on long-term contracts
With big data and sources such as telematics/sensors, external data pools, social media and online search trends, the opportunity to use machine-learning techniques in different areas of insurance operations has never been more attractive.
Machine Learning Applied to Insurance Data
Most insurance companies today focus on three principal goals—better compliance, better cost structures, and more competitiveness. With ML they can have at least partial solutions to all of these:
Better Compliance: Modern machine learning algorithms and techniques can be used to review, analyze, and evaluate information in images, videos, and voice chats. One direct benefit is the ability to better observe and comprehend interactions between customers and policy sales executives. This helps in getting stronger control over miss-sale of products.
Better Cost Structures: With a major portion of an insurance company’s operating costs assigned to human resources, any shift towards artificial intelligence will help in increasing cost-savings. By applying machine learning, insurers can potentially bring down their claims processing time from weeks to a few hours or possibly minutes.
Competitive Edge: Along with reduced operating costs and more efficiency that will help insurers gain competitive edge, their product, service, and process innovation will also help them take a lead.
As against conventional statistical methodologies, ML uses the power of data analytics and can compute apparently unrelated data sets that may be structured, partially structured, or unstructured. If we consider an example, predictive models based on ML take into account:
Structured Data: Kind of incident, intensity of loss, vehicle make and model etc.
Text: Notes, Police reports, coverage details, invoices
Spatial, Graphical Data: Location of incident, vehicle history (claimants, repair facilities)
Time Series: Sequence of events, date of loss, claim date, time lapse between event and action
Can Machine Learning Completely Take Over a Sector that Has Been Dependent on Human Judgment?
Opinions as regards the value of AI and ML in insurance industry are divided. One school of thought feels that this technology can fluidly adapt to new issues and process huge quantities of data to generate accurate results in minutes. The other view talks about machine learning being at a stage that still needs lot of human intervention. According to this view, the automation level for most publicly available ML technologies is not yet one that warrants unsupervised use in financial analytics.
It’s true that ML is not error-free and that it cannot corroborate itself or its solutions to be predictive for the future. Nevertheless, evidence does support its ability to predict certain situations that do not significantly change from the past and therefore don’t excessively defy the assumptions of methods employed to fit the past data. Even so, the humans, as the owners of the process, have the ultimate responsibility of making a decision.
Finding a Middle Ground
The first thing actuaries working in insurance companies should understand is that ML is not a conflict between humans and machine. The focus has to be on an evaluated and scientific approach to building up ML capacities and over time arriving at new ways to incorporate ML into ever-evolving aspects of the insurance business.
Source: Defense Advanced Research Projects Agency (DARPA)
Actuaries will need to step in and interface with ML, albeit to a smaller extent. Their laborious involvement in problems will be reduced because heavy computations can be outsourced to the machines. Human intervention will still be needed in setting them up and validating the quality of results generated. Considering the cultural and risk challenges that the insurance sector faces, the actuaries will have to manually develop ‘proof of concept’ models that may be tested and adapted safely in a risk-free environment.
Acknowledging that machines do well at routine tasks or automation and that algorithms start working better with time, actuaries can focus their early efforts for ‘proof of concept’ on processes that are better understood and add less value. As machines keep making decisions and analyzing growing data, they will be progressively more prepared to take on complex tasks and decisions. The categories of applications include advanced claims handling, smart marketing, allocation of resources, underwriting, and fraud detection.
The actuaries and underwriters will gradually have more time to look into more business cases that can be resolved through machine learning. The analytics and research teams will be freed up to work on other issues such as market price reforms and establish new opportunities for the company that help its customers to stay safe. In addition, at leadership levels there will be greater scope for business growth along with better governance, monitoring, and insights.
Coforge has worked with several insurance companies to help them develop functional solutions such as policy admin, claims processing, compliance automation etc. We have vast experience in optimizing underlying insurance business processes and solutions, along with leveraging potential role of ML and AI in these systems. We know that this ‘transformation with automation’ has already kicked-off in the sector. Those who fail to use it in time will only find their competitors using innovative ways to drive more efficiency and value by harnessing the power of machine learning.