Aureus Insights

A Data Science Approach to Improving Insurance Retention and Persistency

Written by Yogesh Dhavale | May 26, 2022

In the Insurance industry, persistency is an essential term to understand. It's all about an insurance company’s ability to retain its customers over a period of time. High persistency means policyholders continue to pay their premiums with little to no lapse. Satisfied customers will continue to pay their premiums. The lack of payment is a good indication of an unhappy customer.

The term Persistency Ratio is also important in the Insurance industry. It compares the number of premiums paid against the total number of payable premiums. A company can sell 100 policies, but if only 30 are renewed, the persistency ratio will be 30%, which is not very good. The average 13th-month persistency ratio is about 80% - 90% in the life insurance industry.

Persistency Ratio is measured on three metrics:

  • Number of policies
  • Premium
  • Sum assured (fixed amount, pre-determined by the insurer, which is paid in case of death during the tenure or maturity of the policy)

Insurers use metrics to measure the persistency ratio. This is done in the 1st month after the 1st year, up to the 5th year. The 13th month persistency tracks the percent of policies renewed after the first year and have not lapsed. The metrics are tracked for the following years:

  • 2nd year – 25th-month persistency
  • 3rd year – 37th-month persistency
  • 4th and 5th year – 49th and 61st-month persistency

A measurement of persistency is a block of insurance policies determined by the percentage of business that remains in force at the end of a specified period starting at policy issuance. In other words, persistency is the percentage of business retained without lapsing or being surrendered. Low lapsation means high persistency and vice versa.

Persistency is considered to be one of the important differentiators used in understanding the performance of any insurance company. Here are 3 key areas in which persistency helps to show the strength of an insurance company:

  • Indicates the health and quality of the business sourced by the insurer
  • Assists in the computation of the embedded value of the insurer
  • Impacts the cost structure of the operations of the insurer

A sure sign that a policyholder is satisfied with the product’s benefits and features is when they consistently renew their policy. This results in a higher persistency ratio.

According to the article, Why persistency ratio matters in insurance, “The persistency ratio paves the way for building a long-term relationship between a policy buyer and the insurer. When the insurance sales are value-based and are backed up by proper service and all queries of the buyers are handled promptly, in that case, the buyer will not only be satisfied but also can suggest the transparency and loyalty of the insurer to others.

How Machine Learning and Data Science are Helping Insurance Companies to Improve Persistency

Data Science and Machine Learning (ML) algorithms can come to the rescue for life insurers as they deal with huge amounts of data from both internal and external data sources.

ML algorithms learn past persistency patterns of policyholders and predict the future persistency behavior for each policy. These predictions help the insurer optimize their follow-up efforts with policies and improve overall persistency.

The following predictive models are used to improve persistency:

Payment Propensity (or Premium Collection) Models

These models predict the propensity of a policy to pay their premium when it falls due and helps the insurer to improve persistency by targeting customers that are less likely to pay their renewal premium. Given that policies paying their first renewal premium typically behave very differently from the policies which have been with the insurer for some time, these models are divided into two groups:

1. 13th Month Payment Propensity Prediction


This prediction is about the payment propensity of the 13th-month (first year) renewal premium. It helps insurance companies improve persistency by targeting customers less likely to renew the 13th month (first year) renewal premium.

The 13th month (first year) persistency is based on renewal premium payments by policyholders at the commencement of the second year.

For instance, a policy issued in January 2021 is considered 13th-month persistent if the policyholder pays the premium due in January 2022 by February 2022. This reflects how satisfied a policyholder is with the policy they purchased.

The significant predictors for 13th-month payment prediction typically include Auto Pay registration as an essential indicator, along with factors like:

  • Customer demographics (location, state, occupation, education, gender)
  • Agent and channel behavior
  • Product characteristics
  • Premium amount
  • Payment frequency (monthly, bi-monthly, quarterly)
  • Period for which customer will get policy benefit
  • Payment mode (automated, offline)
  • Sum assured
  • Misselling complaints about the product

2. Non-13th month (25th Month through 61st Month) Payment Propensity Prediction

This prediction is about payment propensity after the 13th-month renewal. It helps the insurance companies improve persistency by targeting customers less likely to renew their policies after the second, third, or later years.

Persistency in the 25th, 37th,48th, and 61st months indicates the percentage of policyholders that pay their premiums and choose to continue with their policy plans.

For instance, a policy issued in January 2020 is considered 25th-month persistent if the policyholder pays their premium due in January 2022 before the end of the 25th month.

A policy is considered non-persistent if the policyholder doesn’t pay the premium due for the 37th month within the 38th month of the policy’s life. Following this train of thought for the 49th and 61st months, it would be considered non-persistent as well.

If a policyholder renews at the 6th year, this 61st-month persistency indicates the policyholder’s satisfaction and loyalty to the insurer—the odds of a policy being canceled after the 61st-month decrease with time. Policyholders are unlikely to cancel after six years.

The significant predictors in this case typically give higher importance to the individual customer’s past payment behaviour but also include the factors mentioned for 13th-month prediction like:

  • Past payment behavior of the customer (e.g., has the customer been regularly paying on or before the due date)
  • Premium amount
  • Product characteristics
  • Payment mode (automated, offline)
  • Customer demographics
  • Payment frequency (Monthly, bi-monthly, quarterly)
  • Period for which customer will get policy benefit
  • Historical payments
  • Recent payment date
  • Time since the most recent payment is past due

Payment propensity models described above may be further subdivided by channel, payment frequency, etc., to be able to achieve the desired accuracy on each policy portfolio.

Lapse revival models

Lapse revival prediction indicates policies that are most likely to revive, allowing insurers to get more out of their revival efforts.

The insurance industry is very dynamic. Because of this, it is essential to understand the customer’s behaviour.

Insurance companies face a common problem when customers stop paying their renewal premium. If this lapse is a considerable amount of time, it becomes crucial for the customer to understand whether or not the lapsed policy can be revived.

Lapse revival models identify which policies are highly likely to revive in the near future. Insurers can run revival campaigns for these policies and customers. Doing so will help the insurer reclaim a considerable amount of premium.

Since recently lapsed policies are typically much more likely to revive than older lapses, this model is divided into two variants:

1. Revival From Recent Lapse

The significant predictors for recent lapse typically include factors like:

  • Policy vintage – how old the policy is
  • Product characteristics
  • Sum assured
  • Time left for revival of policy
  • Premium amount
  • Customer demographics (location, state, occupation, education, gender)
  • Agent and channel behavior
  • Time left for payment term to end
  • Frequency of policy paying premium on time in the past

2. Revival From Older Lapse

The significant predictors for older lapses typically include factors like:

  • Sum assured
  • Premium paid till now by the policy                
  • Premium left to be paid by the policy
  • Policy vintage – how old the policy is
  • Product characteristics
  • Time left for revival of policy
  • Premium amount
  • Time since the payment i is past due
Surrender Risk Models

Customer retention is vital in the insurance industry.

Surrender risk models identify which policies are highly likely to surrender. Customers surrender a policy for various reasons, such as dissatisfaction with the insurance company’s service or inability to afford the premiums.

The prediction acts as an “alarm” for insurance companies to target such customers and policies to understand their concerns. Typically surrender risk predictions are executed over an extended period of time (typically 3-6 months in the future). This allows the insurer to reach out to policies at risk well in advance and try to prevent surrenders. Preventing surrenders will help insurance companies have vast amounts of premium.

Significant predictors for surrender risk models typically include factors like:

  • Time left for benefit term to end
  • Customer demographics (location, state, occupation, education, gender, age)
  • Agent and channel behavior
  • Agent termination rate
  • Surrender rate by agent for last two years
  • Lock-in period was completed in the last six months or next six months
  • Time left for payment term to end
  • Period for which customer will get policy benefit
  • Policy status (in force, lapse, etc.)
  • Payment frequency (monthly, bi-monthly, quarterly)
  • Time left for policy’s maturity
  • Premium amount
  • Sum assured

These predictors are typically divided based on the product characteristics, e.g., ULIP products exhibit a different behaviour than others. Typically, a separate model is required to predict the surrender risk of ULIP products.

Conclusion

Payment Propensity Models, Lapse Revival Models, and Surrender Risk Models are three algorithms that learn from past persistency patterns to predict future persistency. All three of these predictive models assist the insurance companies in building a better relationship with their policyholders.

In the next blog, we will be looking into the topics of risk of early claims and fraud in claims.

Aureus' machine learning engineers continually help insurance companies improve their persistency. Check out how the leading insurer leveraged Aureus expertise to generate a 10% increase in persistency.