Currently the health insurance underwriters usually apply a “universal” wear-off factor to determine the renewal premium for small groups. Recently predictive modeling approaches begin to attract more attention. Some underwriters average the members’ predictions into group prediction. We propose new methods to more accurately assess renewal group’s future health risk based on predictive models with aggregated (at group level) member measures. We compare three models: 1) Member model (predictors: individual member measures); 2) Group model (predictors: aggregated member measures) and 3) Combined model (predictors: aggregated member measures and member predictions). Our experiments with 1.5 million Non-HMO commercial members show that by carefully creating and selecting predictors, both group and combined models work well, but combined model yields better results in most cases.