by Charles Sybert –
When you mention AI, many people jump to IBM’s Watson or a massive computer with super complex algorithms built by a group of scientists. In reality, AI is already everywhere, with many people not realizing it is there. For example, this very blog was built using Office 365 built-in AI such as Editor for correcting grammatical errors, and we used MS Teams video conferencing to perform the final review where the background was blurred out allowing participants to focus on the discussion and not what is going on in the background.
In your personal life, AI is present in places such as Gmail auto-suggesting responses, or in the Pinterest LENS tool where you can take a picture of that amazing wood furniture and find similar tables and even help find a custom woodworker to build the amazing table. As you can see, AI is everywhere, and a study by PwC estimated that global GDP would increase 14 percent by 2030 as businesses adopt AI, contributing an additional $15.7T to the worldwide economy Source.
Before discussing the power of AI, we will provide a brief description of AI and the associated technologies:
Machine Learning: training the computer system by a human to identify patterns in data and perform an expected action or decision. For example, when reading a document, the computer will understand there could be one or two spaces after a colon.
Deep Learning: Similar to machine learning, but the model performs all of the learning without being trained or given explicit instructions or frameworks.
Neural Networks: algorithms designed to mimic the human brain and recognize patterns in data. They can identify, classify, and analyze diverse data and can find patterns that are too complex for human programmers to write code for. A fun example of deep learning and neural network is Google’s QuickDraw, a sketching game that uses a massive database of user sketches to guess what you’re drawing accurately.
Natural Language Processing (NLP): a program that helps computers understand, interpret, and communicate in natural language that is understandable by humans. For example, the Cosmopolitan Vegas Hotel built Rose an NLP chat bot that has a personality that responds to customer requests such as housekeeping, making reservations, or finding a great place to go out. Using Rose’s playful personality has wooed hotel guests who booked directly into spending 37% more than guests who do not engage with her1.
Too often, we see executives assume AI is going to reduce costs when the more significant opportunity lies in harnessing its true power and meeting customers on their terms which should increase sales. Through the power of AI, there is the ability to create a unique experience tailored to each individual’s needs as predicted by prior experience, external data, and experimental factors. Below we walk through a few simple Insurance use cases to demonstrate how AI can empower business outcomes.
With the advent of the Internet of Things (IoT), there has been a shift from using proxy data to source data. As an example, auto insurance has, based on age, sex, driving history, and cost to repair a car, group people into rough pools that could be refined by zip codes and a few other factors. Using IoT, the insurance company can gather actual driving experience (i.e., speeding, hard brakes, excessive acceleration, the density of cars, etc.) and combine it with individualized characteristics of the driver to price the exact risk being covered minute by minute. Many insurance companies are offering this type of policy and will continue to expand as on board telematics becomes more available with new technology such as 5G becoming more prevalent.
To prevent the insurance company from becoming just another name in the insurance market, the use of AI must be applied to identify consumers where your unique offerings meet the customer-specific needs.
This will require sifting through mounds of data, countless sales models, and continually experimenting to identify new consumers. One grouping of ideal consumers could be those that buy Starbucks coffee in the morning hours, has an active social media account, drives a car that is 2 -5 years old and enjoys boating. Once the customer has been identified, AI would then build a pricing model to provide the most attractive pricing at the customer’s most opportunistic time during the day to buy. It is only through the use of AI and deep learning that the ideal customer profile can be built and continually tuned to attract new customers.
If the hardest part of growing a business is getting the customer, the second hardest part will be getting the customer to pay for the services. Offering different payment methods such as credit card, Venmo, Apple Pay, etc. with standard billing plans such as monthly, quarterly, one-time payment are expected by consumers.
What does a company do when they have a customer non-pay? Most often they apply a set of cancellation rules with a limited amount of override. What would happen if AI was applied? There could be automated outbound calls/emails using NLP to work with the consumer to confirm when the late payment will be made. Then follow-ups could be done with a second call by a human if the agreed-to terms were not met.
One question that is very difficult for executives to answer is, “Is it okay not to cancel for non-payment?” In some cases, it is better to keep the habitually late consumer as they will be a loyal customer even though their payment is always late. The only way to identify these customers is through AI building a complex model and taking into account many different factors that are continually being acquired and refined such as payment history, insurance policies, driving history, personal items, etc. Based on the model, the company would be able to determine which consumers will pay consistently late versus the ones that will never pay.
We find the two most significant concerns in claims management are providing the customer with a fair adjustment and reducing fraud, which according to the FBI for non-health insurance, is estimated to be more than $40 billion per year. That means insurance fraud costs the average U.S. family between $400 and $700 per year in the form of increased premiums2.
AI can improve the speed and accuracy of claims adjudication through consistent and fair application of business expectations. For example, customers can report a claim through an app or a chatbot where basic information is taken. More advanced consumers can provide access to the phone or car to provide even more geographic or sensor-based data to aid in the adjudication in addition to providing pictures/video of the accident. Through AI, the claim can be scored for fraud, level of complexity, and provide a recommended settlement. If the claim qualifies, the consumer may receive an instant settlement from the AI or set up services from your company’s preferred vendors. If the claim requires a human due to complexity or lack of experience, consumers expect different payment methods such as credit card, Venmo, Apple Pay, etc. with standard billing plans such as monthly, quarterly, one-time payment the AI will provide recommendations of adjudication. If the adjuster offers a different settlement, the results would be marked for machine learning and improvement of future results.
Throughout the customer life cycle, insurance companies consume massive amounts of documents received from vendors, insureds, claimants, and other parties to the policy. Insurance companies have done a great job of moving from paper to paperless by increasing the use of interfaces and using Optical Character Recognition (OCR) software to make the document readable by the computer. The challenge becomes if the document does not have a consistent structure, the words are a bit jumbled, or there are slight variations in wording the OCR process will break. Through the use of machine learning, it is now possible to have the machine sort the various documents into natural groupings, read each document and for any that do not meet a known template, learn new rules.
For example, if a vendor moves the invoice date from the left to the right section of the page, the machine will automatically pick that change up and read the date. If the invoice changes invoice date to issued date, the machine will send that item to a work queue for a human to teach the machine the new term. The machine would then apply this learning to all future invoices reducing the need for human intervention and improving the straight-through processing percentages.
Allstate captive agents have a wide range of products in their portfolio, one of which is Allstate Business Insurance. While most agents are very proficient in personal lines, they do not have the same level of expertise in commercial insurance. To help overcome this deficiency, Allstate built a call center staffed with underwriters and sales support staff to respond to the agent’s questions that range from simple FAQ’s too complex coverage details. The call center quickly became overwhelmed with the demand, and agents had a difficult time navigating to some of the basic questions while in front of the customer. This quickly leads to upset consumers and loss of sales. To combat the long wait times, Allstate built an NLP bot called ABIe to provide agents with quick answers and direct them to exact locations of necessary documents. Now agents can quickly chat with ABIe who understands not only the question but also the context of the request and able to provide meaningful and quick responses.
The success of ABIe has just started with 25,000 inquiries a month and continues to increase as more agents discover the ease of working with the bot3.
The implementation of AI can be limitlessly creative, but it is only through focused leadership powered by a team of highly skilled resources that the vision of AI becomes a reality. To support these changes, insurance companies need to retrain the workforce and engage not previously used skill sets. Delivery teams will need to become more multi-disciplinary focused with skill sets such as data science, linguistics, computer science, pure mathematics, and even behavioral psychology, all of which will be focused on setting up, training and confirming the AI and machine learning are performing as expected.
The program and project management discipline will need to evolve to manage new skills such as psychology in combination with traditional computer science resources. This, while delivering a project that has a “softer” deliverable of judgment and response based on a unique set of variables instead of the traditional application delivery.
To accomplish the task, senior leadership in the insurance company delivery arena will need to commit to a change in management philosophy and approach. No longer will it suffice to adjust to the newest programming language. Now the delivery teams need to commit to cyclical learn, deploy, and evolve approach. As the delivery teams become comfortable implementing the current AI trends, new technologies and approaches will become commonplace, which will require new skills and techniques. It is only through the commitment of continual learning will the delivery of AI be successful.
McKinsey estimates a potential annual value of up to $1.1 trillion if AI tech is fully applied to the Insurance Industry4. Realizing the value and power of AI does not require building your own Watson or Neural Networks but rather smaller focused decisions such as building NLP chat bots, building machine learning models to determine the best pricing or implementing RPA to perform document management. Very few companies have the needed skills in house and even fewer have the focused leadership to manage the delivery. RCG has built a world-class organization focused on AI and achieving value-driven business outcomes.
1. Hospitality Technology (2018, Aug 23) "The Cosmopolitan of Las Vegas Updates Rose, Its SMS Chatbot" Retrieved from https://hospitalitytech.com/cosmopolitan-las-vegas-updates-rose-its-sms-chatbot
2.FB! "Insurance Fraud" Retrieved from https://www.fbi.gov/stats-services/publications/insurance-fraud
3. Allstate "Allstate’s Intelligent Agent Reduces Call Center Traffic and Provides Help During Quoting Process" Retrieved from https://ww1.prweb.com/prfiles/2016/01/07/13154986/EIS-Case-Study-ABI-Intelligent-Assistant.pdf
4. QuantumBlack AI by McKinsey https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-executives-ai-playbook?page=industries/insurance/