Assembling the 'Justice League' for predictive analytics

Deciding what data to use and how to incorporate third-party data effectively is the most important piece of any analytics project. Following the bread crumbs of this decision typically leads insurers to weigh the pros and cons of building their own analytics model or working with an outside tech provider. The aptly named “Build vs. Buy” dilemma involves a variety of intersecting decisions and considerations, including the availability of talent and infrastructure, time-to-market, breadth and depth of in-house data, regulatory challenges and an insurer’s unique competitive position.

Over the last decade, predictive analytics have shifted from key differentiator to table stakes, as companies realize they cannot remain competitive without predictive models to guide risk assessment and pricing. Most companies assume a hybrid approach to the “build vs. buy” concept, developing their own analytics team to work with third party technology providers to build a sustainable predictive analytics strategy.

An analytics team should be comprised of members who have diverse experience with data as well as those with strong business acumen. For insurers engaging in their first analytics project, it’s easy to misconstrue it as a “set it and forget it” technical project, but an analytics platform should be the centerpiece of a data-driven insurance company. Success requires a cultural overhaul just as much as a technical one. These are the key human resources to lay the groundwork for this transformation.

Executive leadership support

The adage “change starts from the top” reigns true with data analytics initiatives. It’s crucial that the executive team creates specific goals for the analytics project and believes in its ability to deliver the desired results. Without this direction from an organization’s leaders, an analytics project can suffer from a lack of defined purpose, which can lead to poorly performing models or failed projects. Achieving leadership support requires business goals to merge with IT goals at the onset of an analytics program.

Using an Applied Analytics framework, the first step is to establish goals to guide the initiative. Whether it is loss ratio improvement or expansion into new geographies, the goal must be clear and measurable. Success metrics should be identified so executives can actively assess the project and and measure progress toward the desired outcome. Once the analytics are attached to measurable business milestones, leadership and the broader organization will view it as an important part of the path to success, as opposed to an isolated IT project.

Subject matter expert (SME)

The subject matter expert is responsible for guiding the purpose of the analytics project and should be intimately familiar with the line of business being modeled. They help determine the types of variables most likely to be predictive, and the outputs to be expected from the data. SMEs are privy to an insurer’s unique distribution channel and determine the most effective approach for collecting data points to feed the model without damaging the customer experience.

Finally, SMEs must also consider the human element’s impact on the data. Underwriters do not always follow the same/standard process or make consistent decisions when quoting and binding policies. This can lead to discrepancies in the data that require additional transformation or exclusion from the modeling data set. SMEs are the cornerstone of developing a targeted implementation plan to determine how the initiative will be governed, which is the second step of the applied analytics framework. This includes establishing underwriting and business rules that work best for each unique organization.

Project manager

Adhering to a strict schedule with checkpoints and oversight is crucial for any successful analytics project. It’s especially important for data-driven initiatives where each step may not be immediately obvious to all stakeholders. Project managers are responsible for sticking to the scope and timeline set forth by leadership and ensuring that the many stakeholders involved in the initiative are communicating diligently.  Without competent project managers, insurers are susceptible to prolonging time-to-market in an unforgiving and fiercely competitive environment.

Data engineers

Prior to building a predictive model, data engineers are responsible for scrubbing data sets to ensure they are clean, relevant and usable. They must ensure that in-house data is current enough to be pertinent and that the correct third-party data is being used to fit specific model needs based on business goals. Additionally, data engineers are charged with being well-versed in extraction of policy, claims and billing data from many potentially disparate legacy systems While technology is available to automate this migration process, it remains unreliable for this critical step in model building. Data engineers are often forced to make instinctive guesses about how best to merge files.  Nevertheless, whether by way of machine or human, an error will drastically reduce model effectiveness.

Data scientists

Traditional insurers looking to build their own analytics teams oftentimes believe veteran actuaries can port their skills into the role of data scientist or modeler. While it’s true that actuaries can become excellent predictive modelers, they need to develop competencies in the statistical techniques used by data scientists.

Modelers and data scientists are generally responsible for building the entire model with the data that was scrubbed and assembled by data engineers. Their aim is to generate the most lift from the model, which is achieved through careful iterative testing to determine the most predictive variables. This also includes knowing when to stop incorporating additional variables to avoid a common error of less experienced modelers: model overfit. This happens when a model is to fit an insurer’s risk appetite so closely that it overlooks good risks in areas they could grow, or is compromised in its ability to flag problematic accounts.

Solutions architect (SA)

If SMEs are considered the experts on the business, the solutions architect can be considered the SME of all the IT-related aspects of the model building process. SAs are responsible for the implementation of the model via software and front-end systems that the underwriters or claims handlers will use to pull the insights into their everyday workflow. Oftentimes, they take the overseer role of all aspects of the model building and serve as translators between the IT team that is developing the model and those who will be using it.

Internal stakeholders

Internal stakeholders include anyone else within the insurance organization connected to the development or use of the predictive model. This group includes underwriters, actuaries, sales, marketing, operations, IT, and leadership. The internal stakeholders are essential for identifying potential issues and ensuring that the model will work correctly in production. Input from underwriters is incredibly valuable for a model built to improve risk selection and pricing, because they are constantly using it in their workflows. Without the deep knowledge and support brought to bear by each underwriter in an organization, the predictive model will neither perform optimally nor remain sustainable.

This is why organizational buy-in is so crucial to applied analytics. If strict business and underwriting rules aren’t established to promote consistent usage and understanding of an analytics tool, the model can perform improperly because of human error instead of the model.

Although these are typically the members involved in an analytics team, it’s important to recognize that insurers should create a deep bench of talent. Replacing many of these roles can be incredibly challenging due to the industry-agnostic need for these skill sets. Making competent hires across the roles discussed is the best way for insurers to set themselves up for success today and in the long-run. Equally important is structuring the analytics team within the applied analytics concept to ensure the strategy, goals and success metrics are defined, the implementation plan is well-structured, and all stakeholders understand the benefits that analytics will provide.

Kirstin Marr (kirstin.marr@valen.com) is president of Valen Analytics. These opinions are the author’s own.

Also by Kirstin Marr: How data consortiums impact the accuracy of predictive analytics