The future is uncertain; however, at times one gets a glimpse of possibilities that may soon have enormous impact on underwriting. In addition, there are several developments affecting underwriting and the life industry today, especially concerning the regulatory environment and the use of data and algorithmic solutions in life underwriting. All of this and more was presented and discussed at the recent Association of Home Office Underwriters (AHOU) meeting.

This year's AHOU was the largest on record with nearly 1000 attendees. Participants included underwriting leadership from all major life carriers, reinsurance companies, exam/lab vendors, industry data/information providers and technology companies. The opportunity to network with industry pros and peers was terrific, along with important insights into the current and future state of the underwriting profession.

Below are highlights from a few of the sessions. Some of these are important for both distributors and sellers of life insurance products to be aware of today, while some are developments to keep an eye on as technology and other changes continue to drive the industry forward. 

Possible Future States of Underwriting

Several underwriting advancements have occurred in the past three years and the pace of change is gathering speed as medical information from various sources is digitized and accessible on a nearly instantaneous basis. For example, one notable life carrier recently discarded their age/amount requirement grid, instead determining requirements on a case-by-case basis after quickly assessing client medical information from various data providers. Because this is a single carrier solution, its usefulness is limited in today's brokerage market where cases are shopped among several carriers. However, it certainly sets in motion a model for others to consider.

What else may be in store for underwriting? Several interesting ideas were presented.

The first was to further innovate by leveraging the power of data. This may be done by using ChatGPT or some other artificial intelligence (AI) model that would be integrated with carrier underwriting rules and standards. In this model, all that's needed is an up-front client authorization that allows carriers to ping various medical and non-medical data sources. This information will feed carrier requirements and rules, which in turn will populate an application. Only necessary items absent from the information gathered will be requested from an applicant.

Another future underwriting model involves developing a "mortality score" similar to a credit score. The score would be fed by digital data sources, with updates done on a continuous basis. Applicants will have their own individual score along with suggestions on how they may improve their score, such as completing a routine clinical test. The mortality score would be mapped to an underwriting risk class and associated premium.

Artificial Intelligence in Insurance

A startling quote from McKinsey and Company was shared to kick off this session indicating that by 2030 underwriting as we know it will cease to exist for all types of insurance. Underwriting will be completed in a few seconds with the use of data and other tools.

Artificial Intelligence (AI) is the development of computer systems to learn, understand and deal with new situations in a seemingly intelligent way. AI can deal with issues outside of the problem at hand, whereas a traditional software program can only handle problems it was built to solve.

AI can address and solve problems in place of a human, and acts as an automation enabler by using Computer Vision and Language Modeling. Computer Vision is where the program looks at images, objects, terms in APS records, and other data repeatedly to recognize and organize information. Language Modeling in turn creates understanding for APS summarization and information extraction necessary for decision making.

In addition to underwriting, other potential applications of AI are in customer service, claims processing, fraud prevention and even sales.

As a profession, underwriting needs to maintain a voice around AI and other technology to understand and articulate both its usefulness and its limitations — which at this point are many. In addition, a challenge will be to develop governance models to make sure the output provided is complete and accurate. 

Data for Good in Underwriting

There are a number of state-led initiatives concerning the use of big data and algorithmic solutions in underwriting. Much of this involves accelerated underwriting but has expanded to include the use of data in any form and the use of underwriting models.

The issues concerning the use of data and algorithmic models revolve around fairness pertaining to protected classes (race, color, religion, gender, marital status, etc.), privacy and transparency. In addition, AI will come under scrutiny as new underwriting tools develop and are put into use.

The underwriting industry is responding to several concerns in these state-led initiatives:

• Unlawful discrimination--using protected class information
• Proxy discrimination--substitution of a factor for a factor based upon protected classes
• Technical bias--models work differently for different groups and may unfairly discriminate
• Disparate outcome—models render a disproportionate underlying correlation impacting one class more than another

In view of above, there are key questions that have yet to be answered by the states:

• Is any difference problematic?
• Is there a reasonable correlation the model cannot exceed?In other words, is it discriminatory only when it exceeds some amount/number?

To protect the industry's ability to underwrite there needs to be well thought-out industry and carrier risk management plans to deal with these issues. Also, a set of common principles should be developed to provide to states concerning the use of data and underwriting models.

Mitigating Misrepresentation

Fraud is a $75 billion per year cost to the life industry. Tobacco use misrepresentation alone is a $4 billion cost. The evolution of accelerated underwriting has increased the risk of misrepresentation in areas that include tobacco, build, blood pressure and other impairments. It's important to note that mortality costs for dishonesty are often modest in the early years but grow over time.

"Material Misrepresentation" refers to a lie or omission that concerns an important or substantive matter that would have changed a risk classification. It's more than being one pound off on a weight disclosure. The industry is constantly looking for ways to get applicants to be more truthful in answers to underwriting questions.

Ways to reduce misrepresentation include:

• Direct to Consumer sales: behavioral economics, digital data, and modeling concepts
• Intermediated (agent) sales: behavioral economics, distribution monitoring, digital data, and modeling concepts

Behavioral Economics (BE) is using economics and psychology to understand how and why people behave as they do. BE tools include:

• An honesty declaration that the proposed insured signs indicating they will be truthful when completing the application.
• Choice architecture on how questions are framed. For example, "when was the last time you smoked" versus "have you smoked in the past 1 year, 3 years" and so on.
• Anchoring techniques. For example, give the applicant an average weight for a similar person at client's age and height. This may help applicants believe they are not overweight and answer the build question more honestly.
• Supervision. This requires having someone ask questions of the applicant. People disclose differently when someone is listening. However, non-supervision may be better for certain types of history such as psych, alcohol and drugs.
• Hawthorne Effect vs Sentinel Effect. The Hawthorne effect is the alteration of behavior of a subject due to their awareness of being observed. The Sentinel effect maintains that the perception of increased oversight is associated with improved behavior. Using a sentinel approach may provide more truthful answers if someone, for example, believes they may have to step on a scale to confirm their weight, or digital data may be pulled to verify information that was provided.

Distribution monitoring: Monitoring different distribution channels to determine where and how disclosures are made and whether they appear to be accurate. One way to improve accuracy is to develop partnerships to get perspective from distribution on how information is gathered during the field underwriting process.

Digital data: Looking for data sources that provide information similar to what was obtained in the past by a paramedical examiner or another third party source. It may help to move applicant authorization to the front of process, before client disclosures are made. Doing this will provide client health history up front for things such as pharmacy prescriptions, doctor visits, and medical diagnoses. These items can be verified with the applicant while an application is being completed.

Modeling concepts: Creating models targeted at encouraging honest disclosure. Behavioral Economics is heavily leveraged in these models. Also this involves using models to look for clients who change answers or pause for an unusually long time at a question on a digitally completed application.

There may be carrier regulatory and ethical concerns for focusing too much on dishonesty, which can lead to unfair bias. In view of this, it's important to understand potential sources of unfair bias and when/how it exists. 

Underwriting Potpourri

Below are some tidbits of information discussed in various sessions which may be of interest.

• Mortality decreases with more exercise and increases with less exercise. The effect is greater at older ages. All of this is not a big surprise. However, there is a U shaped mortality curve with exercise—both too little and too much exercise show higher mortality. However, mortality is only modestly higher with too much exercise.
• Wearables (Apple Watch, Fitbit) create improved persistency in insurance. People do take additional steps and stay engaged. Frictional issues in the use of these include privacy, data capture, and logging information.
• Alternate evidence sources such as claims and pharmacy data provide substantial mortality lift to carriers--not a big surprise.
• Data Sources. There is a fair amount of redundancy across digital health data sources. Challenges are in understanding what is relevant or noisy, and how to deal with conflicting data points. Also, how does one explain decisions to producers based upon digital health data algorithms? This last point is a huge challenge for everyone involved in the underwriting process and needs to be more transparent.
• Future of insurance labs. Labs have substantial protective value. There is regulatory safe harbor in the use of labs in insurance, which is a plus for the industry. Effort should be made by carriers to use labs completed by an applicant's personal doctor since in many cases they are similar in nature to insurance labs.
• Complete Blood Count (CBC) blood test. To understand these, one needs context for the results. For example, children and those who are pregnant can produce unusual results. A single CBC result, like many lab results, is just a snapshot in time so no one can provide a diagnosis based upon this information. The three major groups of a CBC that should be focused on are (1) White Blood Cells, (2) Hemoglobin/Hematocrit, and (3) Platelets. When reviewing CBC results, it's important to look at trends, associated clues (age of applicant, family history, habits, medical history, other lab results, and medical provider comments) and red flags for results that are abnormal.