Data science in insurance is often associated simply with the use of machine learning models, but the role of a data science team is generally to use data to create insight or models that can then be used to facilitate business decisions or a product. At a high level, pricing teams and data science teams have a very similar function. However, both the typical team for each will look and operate very differently.
Skill sets
One key difference comes from the skill sets, strong programming skill is a pre-requisite to be a data scientist. Pricing teams use tabular data with rows and columns, whereas data scientists deal with a much wider variety of data that can only be manipulated well with code, examples include, text, image, graph, and geospatial data.
On the pricing side, the main skill set is pricing domain knowledge, and there is very little programming or engineering knowledge in the industry compared to data science. Because of this, it is common to use software to carry out modelling or rating instead of building from scratch and experience in these is usually required for pricing analyst roles.
Data Pipelines
The largest datasets used in pricing are typically quote datasets, and these are still quite small compared to what a data science team may use, and so working with large volumes data that involve more complex data types will require more sophisticated pipelines where it is not enough for code to work, it needs to be optimal and maintainable.
Data scientists will build pipelines with an engineering approach, focusing on building automated, robust, scalable pipelines and storage solutions, with version control, testing, a defined release process and documentation. Even if the pipeline is not particularly complex or dealing with large volumes of data, utilising these practices means the pipelines are much less likely to fail.
Whereas in pricing teams it is still very common to see data pipelines involving multiple manual steps, often with very complicated Excel workbooks, that are prone to breaking, having errors and are difficult to make amendments, peer review or test. They are also difficult to re-use in other processes, and so there will be many pipelines that have duplicated logic, and if the logic requires changing in one pipelines, it needs changing in multiple.
Systemised Analytics
Pricing teams often work on ad-hoc analysis projects, resulting in sporadic analysis projects that are built from scratch and once completed are not revisited. A data science team will build a system that produces analysis, and if a question arises that the system does not not produce an answer for, that functionality is added. This means that any new analysis that is similar to an existing piece, it can reuse a lot of the existing functionality.
This means that when starting an analytics task, rather than working towards answering that single question, the objective should be to build a process that can be put into production that can answer this question, can be easily adapted to answer similar questions, and makes use of components from other analytical projects.
Conclusion
Given the overall similarity in functions, data science teams and pricing teams generally take very different approaches. At the core, a pricing team is a data science team, and so has plenty of opportunity to look to the data science profession on tools and approaches to make workflows and processes more efficient.
