Small Commercial Data Pool Project - Leveraging Partnership
Issue: November 2016 | General Liability, Property | Download PDF | English By Lori Walters
Gen Re’s recently completed Small Commercial Data Pool Project leveraged the power of partnership to provide our clients access to the advantages of predictive modeling. This project—which included Building, Business Property, and Liability coverages for Small Commercial Package policies—enabled the participating companies to enhance underwriting results and achieve their unique market strategies.
Over the past two decades, the benefits of predictive modeling have headlined insurance articles on a daily basis. For many insurers, predictive modeling has delivered significant improvements in underwriting results through pricing accuracy, target marketing and underwriting process efficiency.
Despite the high rate of adoption in Personal Lines, relatively few companies have successfully deployed predictive modeling in the area of small commercial insurance. Most companies find themselves barred from the advantages provided by predictive modeling due to common barriers, such as sufficient quantity and quality of data, specialized actuarial and modeling expertise, the budget required to fund the analysis, and infrastructure to implement and maintain the models.
What if you could leverage a partnership to access modeling expertise and sufficient data, and reduce project costs as well? Gen Re’s recently completed Small Commercial Data Pool Project demonstrated the power of such leverage. In this initiative, Gen Re partnered with existing clients and Pinnacle Actuarial Resources to develop refined loss cost models based on a voluntary pool of detailed exposure and loss data for Small Commercial Package business. The project accomplished the following:
- Pooling data created the scale needed to use advanced predictive modeling.
- Predictive modeling expertise was accessed to develop the models and validate results.
- Results feedback was structured to implement near and longer-term action plans for greatest utility.
The resulting data pool consisted of nearly $2.5 billion in earned premium over five policy years, and represented a broad set of risks across 48 states. The breadth of pooled loss experience was leveraged to develop loss cost models by coverage, and a wide variety of traditional and non-traditional rating characteristics were considered as potential predictors of loss. The resulting models were found to be highly predictive and demonstrated a 25-time difference in loss costs, comparing the best-performing to worst-performing segments.
Results were structured to allow flexibility in how the pool participants implement what was learned to achieve their unique needs and market strategies. This flexibility allow the participants to:
- Refine product pricing
- Identify which policies appear strongly or weakly priced under their existing rating plans
- Inform underwriting rules and decisions
- Identify opportunities to improve data capture and governance
- Gain valuable experience with predictive modeling
Gen Re’s Small Commercial Data Pool results also presented immediate value to the participants through data quality and data capture feedback, and awareness of the lift available in underwriting results through predictive modeling.
Data quality and standardization is both at the core and the most challenging aspect for this type of project.
It is critical to have high-quality, consistent data, in sufficient volume and detail. Many in the industry are going through systems replacement initiatives and understand the difficulty and cost of consistently capturing information. Pooling data across companies and systems presented additional challenges of understanding, communicating and verifying data element differences so that standardization could be done to maximize the data availability.
We believe this portion of the project was key to the success of the Small Commercial Data Pool’s model development. Based on the extensive insight developed during this process, customized feedback was structured for each participant to improve data quality and governance to enable more accurate reporting and analytics.
Data Capture Feedback
Improvement in data capture and exposure characteristics presents significant opportunity to improve underwriting results through more accurate and granular pricing.
All companies face the challenge of balancing process optimization with maximizing the return on investment, and it is difficult to monetize the gain from improving data. As demonstrated in the Small Commercial Data Pool results, there is significant lift available in pricing results by incorporating new, non-traditional rating characteristics.
Although no sole participant provided every rating characteristic, by leveraging the pool, the participants are now able to quantify the amount of differentiation and return on investment provided by these new, statistically significant rating attributes.
Results interpretation and validation are critical to assess the model’s effectiveness at predicting outcomes for a new or future set of policies.
Model validation assesses the accuracy of the model’s predicted loss costs and the amount of lift the model provides in differentiating loss costs. Best practice is to perform model validation on a “Holdout” set of data, which was not used to develop the model.
In Gen Re’s Small Commercial Data Pool, 40% of the pool data was randomly “held out” to assess the model’s performance. In addition to validation results on the pool’s “Holdout” data set, each participant received validation results specific to their own unique data, as well as output sufficient to assess the amount of lift and return on investment the pool model provides over their current rating plans.
Predictive models can provide detailed insight on a book not otherwise accessible through traditional pricing methods. However, accessing the data and expertise needed to do the work is not easy for companies of all sizes. Leveraging partnerships can provide another way to achieve your goals.