Originally shared by The Actuary Magazine, written by Montoux's Stephen Carlin
Decision science is an interdisciplinary domain that brings together data science, machine learning and artificial intelligence, business context and subject matter expertise. It offers actuaries a framework with which they can leverage their unique and valuable skills, and an opportunity to take on new, influential roles to deliver better results for insurers and their customers.
Crucially, decision science leads with the decision to be made, not the data or technology available. It also treats decision-making as a scientific process. Whether informed by data or intuition, decisions are built around hypotheses that can be tested, refined and improved.
While data science and programming languages are key enablers, decision science goes further. It recognises that current approaches create siloes and leave gaps between data and the people who make decisions. For example:
- Starting with the data, rather than the decision, can lead to insights that are not actionable, or that have limited value
- A lack of domain expertise leaves room for misinterpretation of data, or the ‘discovery’ of insights that are already well understood
- Data may be imperfect or even biased, meaning expert judgment plays an important role
- Data, and the insights it delivers, are not static; taking an agile and iterative approach can refine analysis over time
- Key business decisions are often made on a recurring basis, but results are rarely monitored in a way that ties those decisions back into day-to-day actions.
Increasingly, big tech firms such as Google and many fintechs are appointing ‘head of decision science’ roles to plug these gaps. The ability to gather data, experiment and model scenarios, monitor results, and feed that information back into the decision-making process is how insurers can transform their decision-making so that it meets modern demands and opportunities. Actuaries can and will play a vital role here, due to their industry expertise and technical skillset.
A use case
For products such as long-term care or disability income, wellness and rehabilitation interventions could produce a win-win outcome, lowering claims costs for insurers while improving customer outcomes. The challenge is to design intervention packages that do the most good for the best return on investment. Decision science plays three important roles here.
The first is developing a granular propensity-to-claim model to identify the policyholders who are most likely to claim. This is a fairly standard machine learning application. However, constructing a consistent data universe for the analysis requires deep expertise and understanding of the datasets, covering policyholder information, claims, medical data and social determinants of health. It also requires the ability to slice and dice the dataset and explore different customer segmentations.
The second role is the ability to model the efficacy of interventions, and how this translates to changes in claims rates, claims costs, and impacts on reserves and capital. Again, this sits at the intersection of data science and actuarial modelling.
The third and perhaps most distinguishing aspect is treating this process as a dynamic learning system. Efficacy data for interventions may be limited, especially for new intervention types, so new interventions often start out as pilot programmes. Decision science maximises the data and lessons from these programmes and feeds this back through the claims propensity and intervention modelling to refine and improve the models.
Decision science and actuaries
It’s often said that actuaries are the original data scientists. Similarly, actuaries have a unique set of skills that lend themselves directly to the application of decision science:
- Domain knowledge, including a deep understanding of the decisions that insurers must make
- Experience and comfort working with a wide variety of data sources
- Increasing comfort with data science
- Experience and expertise with actuarial processes, including their advantages and inefficiencies, that decision science could leverage or improve on
- Recognition that data may be imperfect, and promoting a test-learn-monitor-iterate approach – familiar to actuaries from the actuarial control cycle.
There are an incredible number of use cases for decision science within insurance, including pricing, go-to-market decisions, sales and distribution analysis, and claims interventions.
Combining actuarial skillsets with the granular and effective insights provided by decision science, actuaries can deliver value beyond traditional compliance-oriented roles. Instead, they can apply their skills and time in strategic, commercial-oriented ways to improve business outcomes, process efficiency and results.