The insurance industry runs on data, and its business model is based on analyzing data to evaluate information and take appropriate decisions. This makes it one of the best candidates for big data disruption. Welcome the new kid on the block, Big Data Insurance.
Gone are the days when the said data was limited to what applicants would enter in a form. Insurers today have a large amount and diverse spectrum of intimate data that allow them to better understand risk, price their policies, and help their clients with risk management during a policy year.
There are a number ways by which big data insurance is impacting the sector.
How do you price a product without knowing the costs involved in manufacturing it? This is the same problem that the insurance industry is facing every day. Most industries know the cost of labor and raw materials, and calculate a profit margin to decide the price of the product. Insurers, on the other hand, do not know the cost of the product when it is sold. The exact product price might not be known until the claims have been paid.
So insurers – specifically actuaries- have relied heavily on analyzing historical data to predict future behavior for premium rate creation so they can price products. Today, insurance firms are using advanced machine learning techniques such as deep learning to improve profitability.
Let’s take the example of the multinational insurance firm, AXA, a vanguard for big data insurance.
Out of the 7-10% of their customers that cause car accidents every year, only about 1% are involved in “large-loss” accidents, which involve $10,000 in payments. It’s imperative for AXA to identify these clients in order to optimize their pricing for policies.
AXA’s R&D team in Japan started researching how they could utilise machine-learning to predict which drivers are likely to commit these large-loss accidents during the policy period. Initially, they focussed on Random Forest, a popular machine learning technique that uses multiple decision trees. For example, exploring the possible reasons as to why a driver might cause large-loss accidents. Though otherwise effective, the prediction accuracy of Random Forest was only 40%.
So, they created an experimental deep learning model through Cloud Machine Learning Engine, and the prediction accuracy increased to 78%! This improvement could give AXA a significant advantage for optimizing price and cost.
AXA is still in the early stages of this approach, but it is a great demonstration of how advanced machine learning techniques can provide immense business value.
The Coalition Against Insurance Fraud, America’s anti-fraud watch dog, estimates that nearly $80 billion in fraudulent claims are made in US, annually.
This indicates the inefficiency of current fraud detection techniques deployed by insurers. The ability to analyze only historical structured data is one of the main reasons for the under-performance of the existing fraud detection techniques. Analysis solely based on historical data tends to lose its predictive power beyond a certain point.
Also, historical data analysis is inefficient if insurers want to detect sophisticated fraud cases such as staged accidents and deliberate fire-raising (or arson) of commercial premises where profits are down.
Today, fraud analytics has evolved from building models and providing fraud scores based on analyzing historical data, to real-time fraud detection that efficiently and effectively processes huge amounts of structured and unstructured data that is available inside and outside the organization. This is a shot in the arm for the new players, big data insurance companies.
It is necessary for insurance firms to adopt real-time fraud detection techniques to identify sophisticated frauds.
Analysis of internal and external unstructured data helps companies to uncover complex fraudulent activities, which are difficult to find through analysis of structured data.
- Investigators were able to uncover fraudulent claims by scouring through social media feeds of claimants. For instance, a woman claimed to have lost her wedding rings in the ocean, but investigators found a picture on social media where she was wearing her “lost” wedding ring.
- When a person raises a claim saying that his car caught fire, the story that was narrated by him indicates that he took most of the valuable items out prior to the incident. Analysis of the person’s claim documents will indicate that the car was torched on purpose
- Applying data science on claimant’s social network data might reveal his connection with people who are/ were involved in fraudulent activities.
Real Time Risk Analysis
Previously, insurance companies were depending primarily on historical data for actuarial calculations. Now, big data insurance firms can access new data sources in real time. Because of this, they are now able to be more responsive in a highly volatile risk environment.
Earlier, climate data used to be relatively stable so that insurer did not have to monitor it on a regular basis. Today, the climate is changing regularly and insurance companies need to monitor these changes to predict future trends.
Insurance companies can also update weather information by collecting data from countless environmental sensors to understand wind speed, barometric pressure, temperature and changes in the jet stream.
The connected car is transforming the way firms view automotive insurance. Vehicles send data about driving behavior to servers in real time, detailing everything from their speed, acceleration, location, braking and time of day. In the next few years, vehicles will be able to communicate the road conditions on a real-time basis. This data will be incredibly valuable to big data insurance firms as they will be able to make real-time decisions that manage risk. In fact, they can use telematics and automated-driving technology to advise drivers on the least dangerous route to take.
Big data and data science are already revolutionizing the insurance sector. For example, Ford’s Driver Score app tracks driving behavior. The app uses machine learning algorithms to interpret data from the vehicle. The algorithms learn more once it gets more data. Drivers also get daily scores based on their behavior. They can use the app’s Discount Zone feature to share their driving data with insurance carriers based on which they can receive personalized quotes.
Even Big Data Insurance companies have more work to do
Most of the data is scattered within and outside the organization and is in an unstructured form, which makes it difficult for an insurer to gain insights.
They need to have robust data management capabilities to be able to cleanse and integrate the external data with internal data and obtain a 360-view of the policyholder and his/her actions.
All this requires investing heavily in technology, acquiring new talent, and perhaps even transforming the company culture to better cultivate the sort of attitude and thinking required to prioritize which kind of data to gather and how to best exploit it.
But one way or another, big data insurance is here to stay. What remains to be seen is which companies will lead the charge and which will be left on the sidelines.