Predictive Modeling for Premium Audit

By Phil Hatfield, CPCU
Vice President of Product Development
ISO Innovative Analytics

Early adopters of predictive modeling for ratemaking created a competitive advantage for themselves by being the first movers in this area. Today, using this powerful technology to generate more accurate loss costs for individual risks is quickly becoming a competitive imperative. Now, seeking a new edge in the marketplace, forward-thinking carriers are looking to use predictive modeling in all facets of their operations, including the area of premium audit.

Are carriers’ current premium audit procedures as efficient as they should be? How do carriers prioritize audits? How do they select the type of audit to conduct? Traditionally, premium audit decisions have been based on simple rules or intuition. Predictive modeling can change the paradigm of audit selection by providing a statistically superior foundation for making premium audit decisions.

By applying sophisticated analysis techniques to statistical experience data, predictive modeling can generate accurate predictions of future results. Predictive modeling can more efficiently examine a greater number of potential predictors, take into account more interactions among predictors, give predictors different relative importance, and efficiently examine all the available historical data. The end goal is to predict future results more accurately than by using simple rule-based heuristics or human intuition.

This technology prioritizes which policies carriers should audit first and what type of audit they should conduct. In jurisdictions where auditing is at the discretion of the carriers, predictive modeling can help carriers articulate a fact-based reason for choosing one type of audit over another. Predictive modeling also helps optimize recognition of additional premium in the terms of existing contracts, which should be a business priority regardless of jurisdictions’ auditing regulations.

By combining operational data with additional third-party data and advanced modeling techniques, predictive models can provide an accurate prediction of net additional premium/return premium for each workers compensation account. For example, the model can analyze factors including comparison of class codes and business NAIC code; comparison of payroll size and length of time as a business entity; comparison of payroll to sales, NAIC code, and geography; history of large premium discrepancies in a distribution network; consistency of accident descriptions and class codes; and even medical utilization rates per capita in a geographic zone.

Carriers can use a premium audit model to prioritize audits most likely to require a large premium adjustment and concentrate audit resources on those accounts first. Prioritizing audits will result in addressing those discrepancies faster. Even moving average receivables forward 30 days can be worth millions of dollars to the bottom line of a carrier with a sizable audit base. Additionally, using these premium discrepancy predictions proactively at the underwriting stage of the process can result in better underwriting decisions and increased customer retention, since there will be fewer end-of-policy cash-flow surprises for the customer and agent.

Challenges and opportunities
While predictive modeling offers many benefits and opportunities, carriers face four primary barriers in creating a premium audit predictive model. First, individual workers compensation carriers don’t have a large enough portion of the market on a national basis to really “see the market.” Second, extracting, formatting, cleaning, and understanding the required data from internal policy, audit, and claims systems require specialized knowledge and can be very time consuming. Third, assembling a team of analysts with the required combination of predictive modeling skills and workers compensation knowledge is very challenging. And fourth, identifying, purchasing, and managing the required data from third-party data sources can be very expensive.

However, the exercise of building a predictive model can create a virtuous cycle of success. For example, as the modeling process identifies data items that are important predictors of future policy performance, underwriters will start to use that information in the underwriting process, which can eventually lead to refined data requirements for the underwriting systems. Carriers can then feed this enhanced data back into the model-building process to create better models. The more data collected for the relevant predictors, the more accurate the prediction and the more sound the business decisions. Carriers have nothing to lose — except opportunity.

Add to del.icio.us  |  Sign Up