Over the past few years, vehicle telematics has generated a lot of excitement, especially regarding its deployment to support usage-based insurance (UBI) programs. I believe that UBI allows insurers the ability to analyze and price risks far more accurately than traditional pricing models, provided the right predictive models are in place. Like many other potential game-changers across a broad spectrum of industries, UBI is underpinned by ‘big data’ and ‘data analytics’, or ‘predictive analytics’, which are collectively seen as the secret sauce for business innovation.
Data and analytics have been the focus of business attention for decades so it might be tempting to dismiss all the recent chatter and nomenclature as just marketing fluff from consulting firms and technology companies seeking to increase revenues. The reality is that a new breed of analysts, often wearing the title ‘Data Scientist’, using sophisticated data management and analytical tools, are teasing new, important, and actionable insights out of massive and rapidly growing pools of data.
Not long ago, businesses with large volumes of transactional data started building warehouses to store it all. They also built and deployed business intelligence (BI) systems with the intent of providing ready access to the aggregated data. These tools were intended to allow users to ‘dice and slice’ data on their desktop. Such efforts were often costly and only marginally successful because the outputs did not align with user needs.
A 2012 Gartner Inc. study suggests that “fewer than 30% of business intelligence projects meet the objectives of the business.” As well, it is not uncommon to find that the principal benefit of BI systems is ease of access; they don’t yield any more insights than conventional data extraction and reporting methods.
In the claims analytics space, most BI ‘products’ are descriptive statistics, e.g., the number of estimates written, the number of claims made, the number of total loss claims, and the value of settled claims. Typically, these systems also provide derived summary statistics, e.g., percentages, or measures of central tendency, i.e., the average or midpoint values of a dataset. They may be presented in tabular form or as graphs and often accessed through user-friendly ‘dashboards’. A key assumption is that future results will be similar to those of the past. If, for example, average revenue per transaction has generally trended upward then it seems reasonable to believe it will continue to do so.
While descriptive and summary statistics are useful when depicting business results in aggregate, they are not necessarily insightful with respect to individual transactions and what drives results overall. Where big data and predictive analytics can add value is in capturing and managing information that may not be obviously important, and uncovering relationships between multiple data elements that can be used to reliably predict future outcomes. In turn, this type of information can be used to modify existing practices in order to reduce costs or build better, competitive, and more equitable pricing models.
The process of settling collision repair claims generates massive amounts of data. Accident and vehicle details, parts and other material costs, labour hours required, rental vehicle costs and decisions made by various participants at key points in the process all affect the final outcome of a claim. By analyzing millions of records and many millions more data points, data scientists at Audatex use certain key details to accurately predict how a particular claim can be settled.
They do this by constructing complex mathematical models that identify critical relationships in the data and then test these models to assess their predictive capability. The next step is to incorporate these models into next-generation estimating or claims management tools that can help reduce cycle time and costs, and improve customer satisfaction. With more than four million claims records added to the database annually, and over four hundred explanatory variables to comb through, this is no simple undertaking for the data science team at Audatex.
Possible applications include making total loss decisions more rapidly, identifying ‘sets’ of parts that will be required for particular repairs, and making cost effective repair or replacement decisions that are compliant with insurers’ rules. These analytical models will not replace human judgement acquired through years of experience, but supplement it, and allow collision repair professionals and insurance company claim specialists to make more productive use of their time.
While this particular use of big data and predictive analytics will likely never draw the media or consumer attention that UBI has attracted, it is no less important to the evolution of claims settlement practice than is UBI to auto insurance rate-making. As the volume, velocity and variety of the data increases – some of it, perhaps, coming from vehicle telematics – it will be interesting to see the new kinds of relationships data scientists will uncover, and how they will be used to reshape and reinvent, this corner of the automotive industry.