Scouting through the data jungle: Predictive analytics
Big data and complex analyses have become an integral part of Munich Re’s daily business. Predictive analytics allow risks in the life and health fields to be forecast with increasing accuracy. Fabian Winter (Health Analytics, Munich Re, Munich) and Patrick Sullivan (Integrated Analytics, Munich Re, U.S. Life, Atlanta) explain how.
Topics: What does big data mean for the insurance industry?
Fabian Winter: Big data is important for us in two ways. Firstly, it enables us to optimize and automate existing processes all over the value chain: in sales, underwriting, pricing and claims. And second, big data also means that we have sufficient information available to make completely new risks insurable.
Patrick Sullivan: As our business evolves and transforms, it’s essential that we take advantage of the increasing availability of new data sources to create new products and markets. For example, life insurers have the specific challenge of making it easier for people to buy insurance. Wherever possible, we’re employing big data to replace intrusive underwriting requirements.
What is the current data situation in the life and health sectors?
Fabian Winter: The health sector in particular is extremely data-intensive as health claims are quite frequent. This automatically leads to the creation of large tables. This essential internal information is enriched by new data sources, such as wearable devices, which add further information about physical activities, lifestyle or eating behaviour.
Patrick Sullivan: In the USA, there are many third-party data sources that can support accelerated underwriting and intelligent claims management. In Life and Health, we have keen interests in medical data for underwriting. But the data situation varies considerably by current jurisdiction.
Big data and business analytics are being called the perfect match. What types of modern analytical processes do you use to model and predict risks as accurately as possible?
Patrick Sullivan: Increased computing power has made machine-learning techniques like random forests and boosted-tree methods popular. But we have to ensure that clients, distribution partners and regulators can understand the factors driving the models. That means we need less opaque techniques, but can always employ more complex ones to give us an upper bound on predictive accuracy.
Fabian Winter: We are able to run much more complex algorithms than some years ago. But model complexity can quickly lead to wrong conclusions. Therefore we follow an old rule in statistics, which means to systematically use the simplest appropriate model for the data given. In a nutshell: We try to find the best, not the most complex model for our task.
Can analysis help to map future trends?
Fabian Winter: Generally speaking, you can only predict things that have already occurred in some form, or partially, at least once before. But unlike traditional methods, analytics can predict new combinations of known effects – in other words, new situations.
Patrick Sullivan: Data-intensive analytic techniques naturally look backwards to some extent, modelling or finding patterns in what has already transpired. Combining these techniques with improvements in data management to get us more data faster can provide earlier insights into the changes that are driving risk. For example, if we underwrite life policies without requiring fluids, we must quickly assess the applicant’s disclosure of medical conditions or habits, that will drive risk in the portfolio.
In what parts of the life and health segments have you made the most progress with predictive analytics thus far?
Fabian Winter: We have used predictive modelling for example in claims and network management in recent years to predict fraud and abuse, but also to measure the success of our health programmes. In close cooperation with our clients also other applications have been successfully developed in the last years, such as optimization of sales, segmentation of customers and improvement of data driven medical underwriting.
Patrick Sullivan: The focus in the USA has been on predictive underwriting – using models to optimise the use of underwriting factors for risk selection or assignment to preferred classes. Cross-selling is also a perennial topic of interest with primary insurers. But it only works when the underwriting process matches the expectations of the insured and intermediary. Analytics have done some great work in simplifying that process.
What is your personal vision for the future of data analytics?
Patrick Sullivan: Every decision about accepting or rating a risk will be supported by analytics. But I don’t see us turning decision-making over to machines wholesale. Sure, predictive models will be employed in many situations, but the value will not be derived simply from prediction. Isolating the reasons for the prediction, for further review or action, is where the real differential and competitive advantage will be delivered.
Fabian Winter: I hope that big data and analytics will be able to significantly slow or even reverse the cost spiral in the health insurance sector, so medical insurance can remain affordable for as many people as possible. Analysing big data will hopefully also lead to sustained improvements in medical care for patients with chronic conditions.