Analytics

to view this content, you must download the flash player.

How you look at a problem can change the solution.

The use of predictive analytics has proven that a one-size-fits-all model does not deliver optimum results, and black box (nonexplainable) models are typically ignored by underwriters. By partnering with Valen, our customers can quickly develop custom, explainable models that enhance their business processes. These models can be built to reflect customer corporate goals, experience and current market conditions. And, as these factors change, Valen's automation makes it quick and easy to update the models.

Valen uses pattern recognition, evolutionary computation and machine learning algorithms in our predictive models. This technology has already been tested, proven and profitably applied in property and casualty insurance companies.

Because of the quality and rigor of Valen's Data solution, Valen customers have a solid foundation from which to apply Valen's analytics technology. They can develop nonlinear, predictive models that segment insurance carrier risks at a more granular level than previously possible.

Below are some of the activities encompassed in Modeling & Analytics:

Documenting model requirements
Examining and analyzing data to determine optimal modeling approach
Identifying candidate risk factors
Building models using training, test and cross-validation data sets
Sharing model results
Finalizing models
Testing candidate models with a blind validation data set
Preparing models for production
Delivering conclusions to management

Specific deliverables from Modeling & Analytics include:

Visualizor analysis in univariate model
A Valen report that ranks the predictive risk factors and details the model results
Production-ready predictive model
An "explain" function that gives the primary reasons why a policy was given a particular score

Model Training & Testing

Valen separates the analytic data set into four groups:

Training data set
Testing data set
Cross-validation data set
Blind validation data set

Valen processes and analyzes the training data set and develops an initial model. We then optimize the model and measure the model against the "testing data set," the second of the four data sets mentioned. Valen continues training and testing (iteratively) until the most accurate and best performing model is achieved.

Valen's customer retains the "cross-validation" data set at the beginning of the Data phase and delivers it to Valen when a final blind validation is ready to occur. An example of this blind validation process would be if Valen were to develop a model with 2001-2004 data and never receive or leverage 2005 data. Upon completion of model building, Valen would obtain the cross-validation data set and execute the model using the third data set.

The results of Valen's blind validation test would then be compared to the actual performance of the policies (actual claims). Valen would not have had access to the claims data from the blind validation prior to this final analysis. It is through this blind validation process that Valen "proves" the validity of the model.

Visualizor

The Visualizor provides a visual representation of data, allowing carriers to better understand their data patterns and unearth meaningful insights in the early stages of their predictive analytics initiative. Valen customers use the Visualizor at the beginning of the Modeling & Analytics phase to better understand the results of the Data Enrichment & Preprocessing phase. The Visualizor helps Valen customers:

Understand, from a univariate perspective, which data elements are predictive of future loss
Identify segments where claim frequency, claim severity, pure premium or loss ratio are above or below the norm
Highlight potential target industries, market segments and geographies that should be actively pursued
Recognize difficult segments that necessitate additional premiums or warrant declination
Confirm or debunk underwriting guidelines and loss prevention tactics