To understand the coming tsunami in insurance business, we have to understand how insurance business works in the first place. Buying insurance is not like buying a mango. You can buy a mango and find out quickly if it was worth the money you spent on it by tasting it. On the other hand, when you buy an insurance policy, you have no idea how good it is. All you get is a contract with a promised payment under certain conditions. There are fine prints. Unfortunately, most people do not read the insurance contracts. As a result, when it comes to claims, they find that they are ineligible to receive the promised compensation. Take life insurance, for example.
Life insurance policies often have a pandemic exclusion clause. In these Covid times, many beneficiaries are finding out that exclusion clause the hard way.
An insurance policy covers a low probability specific event where the loss is high (in monetary value). If you buy a car insurance for a year, most often you do not have an accident and not make a claim. As a result, you do not find out if the policy would have paid anything at all – after all it is a low probability event. When you buy life insurance policies, the policy may last for decades. For these reasons, regulators monitor insurance business closely. They want to make sure that the company actually pays the compensation when the time comes. The insurance company cannot simply close shop and not pay if it incur losses.
The consequence of it is this: Regulators (like the IRDA) do not insist on a maximum retail price (what is called a premium of the policy) for an insurance policy. Instead, it stipulates a minimum price! No other business regulation works like that. If the insurance company does not charge high enough for policies, it may lose so much money that it could go bankrupt and leave the policy holders nowhere to go. It does not happen because premium incomes of an insurance company is kept separate from other incomes of an insurance company. This is the reason for charging a minimum price (premium) for a policy.
How does an insurance company set the premium? First, it has to calculate the average loss incurred for a specific policy. Let us take a concrete example.
A life insurance company is selling life insurance for 25 year old healthy (meaning no obvious disease such as cancer or heart defect or smoker) males (females will have a separate premium).
The probability of dying for that individual in India is 0.0017. In other words, out of 10,000 such people 17 would die in a given year. For one rupee (per year) premium, the company can pay Rs 588 (=1/0.0017) life insurance benefits on the average if the company breaks even on that product.
But, if it sells such policies at that premium, it will lose money half the time. If the company sells that policy for 10 years for 10,000 men of that age, it will lose money 5 of the 10 years. Such a policy is not viable in the long run. This is precisely why the IRDA will not permit the company to charge such a low premium. In fact, no company will sell a policy of paying Rs 250 for a Re 1 premium per year for that age group of males.
For females of the same age group, the probability of dying is 0.0013. The same calculations yield a breakeven premium of Re 1 policy to produce benefits worth Rs 769.
There are two relevant observations here. First, females of all relevant ages of buying life insurance have lower mortality rates than their male counterparts of the same age. Hence, life insurance policies always have lower premiums for women. Second, the insurance company is allowed to discriminate against customers based on their age and sex (but not other factors like location). No seller of mangoes can do that legally.
In technical terms, this is a pure term life insurance policy. Most insurance companies are reluctant to sell such policies because they want long term customers who will keep renewing their policies to make a bigger profit.
This example can be used for calculating premium for any other kind of insurance policy. Once we calculate the probabilities accurately, we can calculate a level of premium based on those probabilities. This process is known as ratemaking in insurance parlance.
The trickiest part of ratemaking is to put an individual to the right class of risk. One omnipresent problem in risk classification is attracting the people with higher than average risk. For example, if I know my parents have died of heart attacks, and my grandparents died of heart attacks, I will have a higher than average risk of dying from a heart attack. I might be buying a life insurance policy precisely because I have this knowledge and the insurance company does not. This problem is called an adverse selection problem. The other problem is that I might become less careful about the underlying covered risk if I know I have an insurance policy. I may not be less careful of dying if I have a life insurance policy. But, I might be less careful driving my car if I have comprehensive car insurance with no deductible or coinsurance clause. This problem is known as a moral hazard problem. [It is precisely this problem that car insurance is never sold with zero deductible or coinsurance clauses.]
Insurance policies are sold through agencies – specifically with agents. Agents (and underwriters) assess the risk of the potential buyer of a policy. If the agent signals a potential bad risk, the policy will not be sold.
Once the policy is sold, the risk of the buyer can be constantly monitored. For example, we know that the longer a person drives at a stretch or the higher the speed at which a person drives, the risk of an auto accident goes up. If we can monitor those parameters of driving, we can assess the changing risk of the driver.
This is one area where AI comes into play. Today it is possible to monitor a driver through a GPS in real time to measure the speed and driving duration cheaply.
Over the next decade, the agency model to assess the risk of a potential insurance buyer can be entirely replaced by AI/ML agents. Automation will replace human judgment in the process making it far more uniform.
The AI/ML domain will also be very useful in assessing the risks. Consider the adverse selection problem of attracting the wrong people. AI/ML methods can be used to search and discover many underlying risks. For example, the death certificate of my parents can be pulled up to verify their causes of death before selling me an expensive life insurance policy. My health can be monitored through my FitBit device making sure I am truly at the good risk that I claim to be for buying health insurance. Smart watches are already capable of detecting how often and how much I consume alcohol. Thus, a wearable can easily monitor the risk for a particular individual.
A large area of an insurance company is dedicated to verification of claim authenticity. Insurance fraud is an ever present worry of an insurance company. Most fraud detection today are done manually. With AI/ML, we can connect different data systems very quickly to detect fraud. For example, a payment for damage to a car requires several quotes from several garages. If one garage is found to be always producing higher quotes, we can blacklist it from future business.
Similarly, for medical insurance, if one clinic is consistently charging more from the same treatment, it can be identified quickly and disbarred.
In the case of life insurance, there have been many instances of faking own death. Recently, there was one case of one Prabhakar Bhimaji Waghchaure who tried to collect five million US dollars by killing one lookalike and pretending he was the victim of a cobra bite. The case unraveled only after the insurance company sent an investigator from the US to India. Such verifications are expensive. Access to phone records can easily unearth such frauds. In fact, this was precisely the method used in that particular case. But AI/ML methods can easily automate the process thereby drastically reducing the cost of fraud detection.