Machine learning (ML) is all the rage, oftendescribed as the new frontier in predictive analytics. While itshistory dates back to the 1950s with pioneering research on simplealgorithms, more recent developments in the 1990s paved the way forthe new wave of applications in artificial intelligence. Most agreethis technology will have a notable impact on the industries andthe jobs of the future. From a predictive modeling perspective,incorporating new data into evolving models allows the predictivepower to continually improve.

Related: The power of analytics for insurance: You ain'tseen nothin' yet

Industry implications

Insurance presents interesting and nuanced use cases whenapplying predictive models to assessing risk andpricing policies. Being a heavily regulated industry bringschallenges and the need to explain precisely how risk selection andpricing decisions are made. Also, consumers still want trustedadvisors (agents, direct insurers, etc.) to help them navigate theright insurance coverage for their needs. That means continuedhuman interaction in explaining coverage and pricing details.Particularly with the larger, more complex risks in commercialinsurance, it's clear that insurance analytics must go well beyonda 'one-size-fits-all' modeling technique.

Veteran data scientists in the insurance industry use a variety of techniques to develop predictivevariables and final model output. They understand that insurersare required to go beyond a simple understanding of model scores;regulators often require an explanation for the decisions andrecommendation made by the model, as well as a deeper understandingof why it makes the recommendations it does. Underwritersand claims adjusters also need the same context and transparency tohave confidence incorporating model recommendations into theirdecision making.

As such, it's imperative that the model doesn't become a blackbox.

Related: Using catastrophe modeling to predict riskscenarios

Best practices

Since machine learning automatically changes the output as themodel consumes more data, underwriters or claims adjusters can be left in a positionwhere an explanation is impossible. The same policy may be treateddifferently in February than it was in May, and without asatisfactory explanation into the change, drawing the ire ofregulators. In other words, the biggest benefit of ML is also itsbiggest problem. 

Using multiple techniques is also important for variableselection. Different methods grab different triggers from the data,which become key variables for a model. There are multiple forms ofboth univariate and multivariate modeling techniques to choosefrom, including tree based methods, stepwise regression, and Lorenzcurves. Stepwise regression seeks variables that are linear,whereas tree-based methods look for both linear and nonlinear.

The best way to approach analytics is to use multiple modelingtechniques that support an insurer's unique goals and position inthe market.

Related: 21 emerging risks for the insurance industry andthe global economy

Getting the right mix of modelingtechniques

Linear Models or Generalized Linear Models (GLM) can be thoughtof as global models. That is a prediction equation that isestimated and applied to the entire data set. While thesetechniques are the gold standard in insurance, and will likelycontinue to be, other nonlinear methods can be useful forunderstanding and predicting a target value of interest.Specifically, these methods can help when a data set containsheterogeneous groups that should be treated differently, such assplitting a data set by premium size or geographic region.Tree-based methods are often used in these cases to deal withinherent differences in the data set. By partitioning the dataspace, we can fit highly accurate models to these subpopulations(called leaves in tree-based modeling), which can often overwhelmthe predictive power of global solutions. Beyond making accuratepredictions, the very structure of the tree can be informative forthe business as well. With this machine-learned intuition, one canalso use these learnings to help with other statistical techniquessuch as GLM. 

Keeping the human connection

A recent Wall Street Journal article pointedout that AIG, which has invested heavily in predictive modeling,still proudly uses underwriters to make judgment calls based ondata. Valen's data also supports this, understanding that humanjudgment and analytics need each other to deliver the bestresults. 

The graph above illustrates the lift of a predictive model. Thegreater the lift, the more effective an insurer's predictive modelis. The blue line represents the loss ratio improvement based on acombination of the underwriter and the model when making decisionson pricing policies. There is a more significant lift here whencompared to both the underwriter (green) and predictive model alone(red).

Organizational culture comes is an important consideration inany analytics implementation. Many underwriters are still wary oftechnology taking over their jobs. GLM provides stable answers anda high degree of transparency into the final model output. It armsunderwriters with the appropriate context and places them asoverseers to the model, with the ability to make adjustments asneeded.

Successful analytics initiatives require that the tech beingrained in the strategic principles of the business, and apredictive analytics approach must converge with the overallcorporate strategy. That can't happen without executives who areconfidently able to make important decisions and incrementalimprovements based on the insights from predictive modeling. Thereare many modeling techniques available and each has theiradvantages and disadvantages when incorporating within a completeanalytics strategy. The key is finding a way for them to worktogether to support the proper use case.

Dax Craig is CEO of Valen Analytics. To find outmore, message Valen Analytics® at [email protected].

Read other articles by thisauthor:

A proactive approach to working withregulators

Barriers to the innovative mindset

Data analytics key to solving the puzzle ofcommercial auto insurance