Sweet dreams – Using machine learning to eliminate pricing nightmares

Sweet dreams – Using machine learning to eliminate pricing nightmares

Stuart Rose

Blogpost Image

How do you price a product when you do not know how much it costs to produce? This is a problem the insurance industry faces every day. While most industries know the cost of materials, labor, and profit margin to calculate the price of their products, insurance does not know the cost of the product when it is sold. The true cost of the product may not be known for many years once all the claims have been paid. Hence insurance companies, and specifically actuaries, rely heavily on using historical data to predict future behavior to create premium rates to price its products.

Competition is forcing insurers to adjust rates more frequently in order to retain existing customers and attract new business. Most insurance companies use multivariate statistical techniques like generalized linear modeling (GLM) for pricing to develop accurate pricing structures. Yet many insurers take weeks, if not months, to implement a new rating structure and the effective performance of these models rapidly deteriorate over time.

Today, in order to gain competitive advantage, insurance companies are beginning to use machine learning technology to mine data to find surrogates that reveal information about the insured risk. The results from the machine learning algorithms can be used as an adjustment to existing rating structures. Thus changes to pricing models can be deployed a lot quicker than traditional methods, thereby eliminating many of the implementation complications associated with traditional GLM projects.

Besides speed-to-market, machine learning algorithms have other advantages over traditional analytical methods. The predictive models from machine learning can be built considerably quicker, up to 10 times faster, and the models are more accurate and stable. Machine learning is not limited to human hypotheses and intuition. For example, not all 17-year-olds are bad drivers. Machine learning is able find non-linear relationships in the data that overcome these limitations.

To learn more about how machine learning can end your pricing nightmares, watch this short video by Chris Cooksey, Head Actuary, Data & Analytics at Guidewire.