AI and the Insurance Industry — Winners and Losers of the Coming Revolution
By Alex Kainz on The Capital
Machine learning and Artificial Intelligence (AI) sound futuristic and lead to associations of programs that develop consciousness. In reality, machine learning and AI are firmly rooted in statistical mathematics. Insurances have employed these methods to calculate risk longer than computers existed.
But Machine learning combines statistical techniques these with innovations such as neural networks, natural language processing, and big data. That opens up new and fascinating use-cases. In this article, I’ll examine 5 ways how to use machine learning and AI in a way that profits insurance and customer and one that doesn’t.
Read below to see how to use machine learning and AI for the profit of the customer and how to use it to maximize profit and face regulatory and customer backlash.
Health Insurance Pools
AAlibaba the Chinese e-commerce giant turned its payment juggernaut AliPay into ANT Financial to enable financial services beyond payments.
Only 5 years later, ANT Financial has a valuation of $150 billion and is the most valuable startup in the world.
Among payments and investments, ANT Financial offers an insurance product that uses machine learning to offer health insurance at a fraction of traditional insurances.
Launched in 2018, by the end of 2019 the product, called Xiang Hu Bao, already had 100 million users. Xiang Hu Bao is not an actual health insurance but a claims pool that offers protection against 100 types of critical illnesses.
Members of the pool get their medical expenses paid, up to 300,000 RMB ($45.000), when falling critically ill. The amount is shared equally by members of the pool. This means when an incident occurs, members of the pool pay after the fact. Disputes are managed by volunteers that vote on the validity of the claim.
This product is not a traditional insurance, but it fills a similar niche, financial protection in the case of serious illnesses. With low overhead costs and transparent pricing, as fees are set an 8% management fee.
To increase the quality of the pool members, the pool requires a credit score of 650 or more to join. The score, called Zhima or Sesame Credit, is calculated using machine learning from “hundreds of sources — ranging from purchases on Alibaba’s Taobao marketplace to subway fares.”
The whole operation had only 50 employees in 2018, as receipts are also analyzed and handled by machine learning software.
For end-users, the end result is health insurance that is more affordable than other options. Capped at RMB 188 ($27) per month, this opens up health insurance for poorer people. More than two-thirds of the pool participants make less than 100,000 RMB ($14.000).
Reducing Customer Service Costs with Chatbots
Facing rising demands for cost-cutting, insurances are turning to AI and machine learning for customer-facing interfaces.
The major touchpoints with the customer are branches, call centers, web sites, and mobile apps.
Branches and call centers are costly as they require personal and real estate.
Web sites and mobile apps are impersonal and require some technology affinity when used.
To solve this conundrum, insurances are building what is called conversational interfaces.
Well-known conversational interfaces are Apple Siri, Amazon Alexa, and Google Assistant.
While some insurances rely on textual chatbots, others use voice-activated bots similar to the well-known examples above.
But current chatbots are facing criticism, they “have no memory” they offer “Cold Experiences” and “Low error tolerance.”
Chatbot looks great in demos if the natural text follows the playbook. When typing in “I want health insurance for my trip to Italy” leads to an answer, “A trip to Italy is always a good idea.”
But misspell Italy as Italie and the chatbot is stumped.
Also, the storyboard like process of most chatbots may lead to a loop, if the question doesn’t fit into the current storyboard. This leads to the frustrating experience that a lot of people associate with automated menu systems.
Often chatbots are used as a complicated way to search the Frequently Asked Questions (FAQ) or a set of prepared responses.
But with new AI-driven platforms like Google Dialogflow and Amazons Lex, chatbots may be able to solve much more complicated requests. The experience these companies have with Google Assistant and Amazons Alexa will allow a new generation of chatbots. With sophisticated machine learning algorithms and deep pockets for research, this new generation will likely allow more natural conversations.
For example, KLM is offering a chatbot to book flights based on Google Dialogflow. It is whimsical and doesn’t take itself to serious which makes it fun and less frustrating to interact with. This encourages people to explore more than a strict automated phone menu in chatbot form. It seems to be also better at handling misspellings and deviations from the storybook, while still keeping a memory of the conversation.
Optimize Risk prediction by Analyzing Telematics
Fairness in insurance means charging every customer the exact premium related to the risk (and cost) of a claim. The higher the potential for a payment the higher the premium. In a fair insurance, the uncertainty would be zero and everyone would pay the exact premium to match the risk. In reality, however, there is always a level of uncertainty.
Car insurances use any number of factors to find the correct risk level for an individual. Factor such as age, driving history, marital status, and so on. But while each factor reduces the level of error, there still remains a lot of uncertainty. Maybe a 25-year-old single man is a careful and safe driver, but due to his age and marital status, he will be seen as a high risk.
Some insurances have offered reducing premiums if the customer is willing to share his behavior on the road. For example, State Farm, the largest property and casualty insurance provider in the United States, offers a program called Drive Safe and Save. The program tracks and analyses driver behavior on the road. This way, the company does not have only have to rely on statistical data, but can directly relate the driver’s behavior to his risk. That way, the careful 25-year-old can get a quote that is related to his personal risk level and not the risk level of his peers.
State Farm says that the customer receives up to 5% discount on his insurance just for signing up and “some customers could see a discount of up to 50%.”.
One reason why insurances can offer 5% is the process of self-selection. Just like the people how respond to a Nigerian spam mail are more likely to fall for it, the people who sign up for telematics are more likely to be safe drivers.
In general, a driver would consider itself a safe-driver to sign up. That’s the 5% signup bonus. Only 5% because of the everyone-else-on-the-road-is-an-idiot- syndrome(76% of Americans think they are good drivers while 93% commit bad driving behaviors). For the rest of the insurance discount, machine learning analyzes the driving behavior and produces a risk profile for the customer. Currently, that is mostly based on the usage of the car, which is why this model is also called Usage Based Insurance or (UBI). But a better model can produce risk profiles that are exactly tailored to the driver in the future.
Not everyone is excited about having every move on the road being tracked by the insurance. A hack on the size of Equifax could expose the driving data of millions of people. Even the most secure data may be used for law enforcement or state security purposes.
Data privacy concerns aside, telematics results in fairer insurance pricing for all the safe and unsafe drivers.
Predict Estimated Car Damages with “The Box”
Munich Re is the 2nd largest reinsurance company in the world. Munich Re specializes in insuring insurances. By taking risk out of other insurances hands, the risk to the individual insurance company is reduced. To accurately estimate the risk, Munich Re models risk for floods, earthquakes but also pandemics and cybercrime.
These models are a perfect fit for data science and machine learning.
“In the past we had statisticians and mathematicians, that build our models. Now we need data scientists and data engineers.”
Doris Höpke, Member of the Board of Management Munich Re
One of the areas that will change dramatically is motor insurance. Not only because of telematics or usage based solution but also because of assisted driving technologies. “94% of vehicle crashes are due to human error. 80% of these crashes may be reduced or eliminated by automation.”
Munich Re provides machine learning tools to the customer-facing insurers for motor insurance. “The Box” can be used by these insurance companies to accurately estimate, taking into account the changing motor insurance market. It can take weather information, accident statistics, or socio-economic information into account.
“[Data analytics and Artificial intelligence mean] faster claims estimates and handling, and better pricing as a result of improved accuracy in risk assessment. Not to mention better loss prevention in the first place.”
Doris Höpke, Member of the Board of Management Munich Re
Preventing Fraudulent Claims
According to the Coalition against Insurance Frau “Conservatively, fraud steals $80 billion a year across all lines of insurance”.
Machine Learning and Artificial intelligence are ideally suited to find hidden patterns in data and help with Fraud prevention.
A human analyst or a human reviewer can only look at a handful of signals at a time and make a determination. But there is enough data out there and that’s really when machine learning comes into play. Because it’s literally able to crunch thousands of signals and look at probabilities of abuse or probabilities or fraud. That’s really where the industry is going from a machine learning viewpoint.
Kevin Lee, Trust and Safety Architect at Sift Science
The Insurtech 100 2019 list lists 6 vendors that are tacking fraud with machine learning and artificial intelligence. Among them the Shift Technology at number 4 and the only fraud prevention vendor in the top 10.
FWD is the leading Pan-Asian insurance provider with offices in Hong Kong, Malaysia, Singapore, Indonesia, Japan, Vietnam, the Philippines, and Thailand.
FWD Group has an artificial intelligence and machine learning center in Singapore. Since the end of 2019, FWD Singapore is using Shift Force for fraud detection and is the first user in Southeast Asia. According to Shift, FORCEachieves a 75% fraud detection rate versus the industry standard of 35%.
Overcharge The Customers Least Likely to Leave
Earlier examples in this article show how insurance companies use machine learning to model the risk as exactly as possible. Both insurance and customer benefit, the better the risk model, the fairer the pricing.
Allstate, one of the largest insurance providers in the US, however, has used machine learning for a not so customer friendly purpose.
According to an investigation by the markup, Allstate insurance used machine learning to find the maximum premium they could charge. The insurer claimed to examine each of the individual policies to make sure that the rates are still accurate.
Allstate, one of the largest insurance providers in the US, however, has used machine learning for a not so customer friendly purpose.
According to an investigation by the markup, Allstate insurance used machine learning to find the maximum premium they could charge. The insurer claimed to examine each of the individual policies to make sure that the rates are still accurate.
[We] found that, despite the purported complexity of Allstate’s price-adjustment algorithm, it was actually simple: It resulted in a suckers list of Maryland customers who were big spenders and would squeeze more money out of them than others.
To measure the likeliness of a customer leaving, it seems the algorithm looked at how much the customers were already being charged and calculated the risk of leaving to be less if the customer was already paying a huge premium.
And according to the markup the algorithm did not seem to be equally distributed
In Maryland, seniors were overrepresented among those customers who were owed discounts but would not have gotten them. Allstate proposed giving those Maryland customers over the age of 62 a median discount of $1.64, far less than many deserved, according to its new risk calculations.
This illustrates a danger of using machine learning for the customers. They can be opaque and hard to understand. Some algorithms require knowledge and understanding of statistics. Other algorithms, such as neural networks, are black boxes that make ithard to explain the results even for experts.
Conclusion
Used in the customer’s interest, machine learning, and AI can benefit both the customer and the insurance company. By opening health insurance up to lower-income populations through health insurance pools, by improving customer service through chatbots, by pricing car insurance fairer through telematics and by preventing fraud. But used in the interest solely of the insurance company the use of these techniques can lead to maximal extraction of profit from the customer. The danger of using machine learning in that way is a backlash from customers and regulators that is in no one’s interest. Customers already are wary about privacy issues and feeling reluctant towards artificial intelligence. Insurances will be well advised to make sure that machine learning and AI are used transparently and create value, but not to extract maximum profit.
