ElectrifAi
July 20, 2021

Long-Term Disability (LTD) Claims Optimized with Machine Learning

Long-term disability (LTD) insurance provides the most comprehensive and cost-effective income protection. Insurance providers offer LTD insurance to help maintain customers’ standards of living if they are unable to work due to non-work-related illnesses or injuries.

Who qualifies for long-term disability? Those who cannot work due to health-related reasons. Depending on the employee’s policy, coverage begins several weeks after filing a claim and can payout for a year or two up to 10 years. Some policies can also cover employees up until age 65 (retirement age).

That is a long time for an insurance provider to payout. How do providers weigh the risk of a customer becoming disabled versus how much to charge premiums? Keeping premiums for this insurance low increases customer satisfaction and retention rates.

So, how do insurance providers keep the rates low? By predicting the likelihood of an LTD claim. Machine learning can help the provider with accurate plan pricing that creates a better customer experience and decreases unforeseen costs.

How does machine learning accomplish this predictive capability?

ElectrifAi’s Long-Term Disability Claims machine learning model is very effective at:

  • Risk prediction
  • Improving LTD incidents/cost predictions to better match rate to morbidity risk.
  • Enhance pricing
  • Optimizing LTD claim rates and reduce high-rate variability by addressing overlapping feature contributions that drive up cost predictions.
  • Expand internal and external data use
  • Helping companies in a highly competitive market where peers are beginning to draw insights from full spectrum of historical and external data.

This model has also been used in the real world with real business impact:

  • $15 million improved margin opportunity to align premiums with predicted risk and enhanced reserve calculations.
  • Improved LTD claim cost prediction to better match rate to morbidity risk.
  • Optimized pricing and reduced high-rate variability by addressing overlapping feature contributions that drive up cost predictions.
  • Enhanced reserve calculations.

What technology does this machine learning model use that makes it so accurate and effective?

  • Predicted LTD incidence at member level.
  • Integrated data from 7 different systems including mix of transactional and snapshot tables requiring complex joining and inheritance logic across multi-tiered products and members.
  • Approached the problem both as a classification problem to predict the likelihood of filing a long-term disability claim and a prediction problem to estimate the cost of each claim.
  • LTD Incidence Probability Prediction
  • Conducted stepwise model build to achieve variable reduction, ranked signal contribution and best model performance for predicting probability of a member filing LTD claim.
  • Predict LTD Severity
  • Predicted claim amount if a member filed LTD claim to derive expected LTD cost prediction.
  • LTD Cost Prediction
  • Layered on a claim costs regression model that also exhibits high performance, with an Actual to Expected ratio of close to 1 on average across business defined buckets and with a low variance.
  • Built Actual to Expected comparison for different plan size and member levels.

What data sources and features are used to run the model?

  • Greater predictive power from linear and non-linear interactions to capture complexity going beyond conventional hypotheses. Predictive tree-based model is less sensitive to correlation and collinearity.
  • Leveraged use of non-traditional data sources and third-party data sources to improve predictions such as dental claims experience, unemployment rate, etc.
  • Useful feedback into actuarial/business planning processes.
  • Different cuts and profiling of A/E is a mechanism to highlight areas for further review for rate changes and potential reserve changes.
  • Identification of high-risk claims/claim drivers and flexibility on levels granularity
  • Member level prediction to facilitate granularity and rank ordering members/plans to establish high-risk areas.

The model uses the following data to output a likelihood to file an LTD claim and to estimate the cost of an LTD claim:

  • Group plan level data (e.g., coverages and provisions)
  • Tenure
  • Plan size
  • SIC
  • Employee-level data (e.g., gender, salary, dependents, dental claims)
  • Plan level attributes
  • Dental claims
  • LTD claims cost/claims incidence
  • Member data
  • Premiums/pricing
  • Billing data
  • Bureau of Labor statistics data

How can ElectrifAi help your business?

ElectrifAi has extensive domain knowledge and experience helping well-known insurance companies increase revenue, decrease costs, and reduce risk. Would you like to enhance your customer satisfaction and create a better business?

To find out more, reach out to us today!