A Recipe for Success: Integrating Predictive Models into Your Claim Handling Process

What are the most important things to consider when implementing predictive models into your claim handling process?

Make sure you have the right people on the project team. Integrating predictive models into a business process is really an exercise in change management. Implementing change is a team sport. You should consider a multi-disciplinary team that involves IT, the predictive model team, and the business unit (say the Claims department). IT can help with system integration and deploying the final model.The predictive analytics team will need the help of the Claim department to develop a model that solves the identified business need.

Ensure that you have leadership support for the project. Integrating a predictive model into the business process takes time and resources. Discuss the resource commitment with leadership upfront. Make sure there is an understanding and support for seeing the project through. Communication is key make sure leadership, the project team, and other key stakeholders are aware of the project’s importance. Change requires constant reinforcement.

Figure out the exact business need you are attempting to automate with the use of a predictive model. One size does not fit all. Suppose you would like to identify claims that would benefit from nurse case management. Avoid using a model that identifies claim severity to make the decision about nurse case management. Consider training a separate model that is specifically tuned to identify opportunities for involving a nurse case manager.

A predictive model will return a score or probability for a given claim. Sometimes that information is categorized into buckets – say low, medium and high. For an adjuster, the immediate question is, “What has the model identified that I have not considered?” This is particularly true when there is a difference between a model’s evaluation of a claim and that of an adjuster. Give consideration to what additional information would be helpful for the adjuster to better understand a prediction. For example, maybe a model is trained to be sensitive to long-term opioid use. If the predictive model identifies this situation on a claim, the score moves to a higher value. In this example, provide both the score and information about the presence of extended opioid use on the claim. Not only does it help validate the model in the eyes of the adjuster, but it also may speed up the time towards determining an appropriate intervention.

Ensure close collaboration with the modeling team. Provide feedback on a regular basis. Look for outliers and missed predictions. Work with the modeling team to understand the predictions that do not make sense. If an adjuster does not agree with a prediction, document the reasons. This information can be used to improve future generations of the model.

Business expertise is crucial to building accurate and reliable predictive models. In most instances, a business subject matter expert is far more in-tune with the dynamics of the business need solved by a model. You will find that the best results occur when there is a free-flow of information between the business and the modeling team.

Building predictive models is an iterative process. Do not be discouraged if your first model does not perform as well as you expected. A predictive model is never perfect. Learn from the model – what it gets right and what it gets wrong. Provide this feedback to the modeling team. As I’vepointed out, success is dependent on team collaboration.

The business subject matter experts should ask questions about the data. The modeling team can then explore the data and provide answers to the questions. For example, an adjuster may want to know how many hospitalizations result because of slips and falls.Your findings can lead to new variables that will make your models more accurate.

Once you have a production model, set up challenger models with new variables that may enhance performance of the prediction. Make sure the team continues to explore the problem and look for ways to improve.

Leveraging predictive models to speed business decisions is not a once and done endeavor. It takes patience and commitment to the process.