Current portfolio analysis

Evaluating profitability, trends, areas of strengths and weaknesses. We look at the transactional data of the insurer and derive insights regarding the insurer’s processes, profitability, reserving accuracy, market position, products, overall performance, or any other area of interest.
During the exercise we clean and process the client’s data (sometimes algorithmically) and get into a standardized form that is convenient for further use.
In past years we have seem a lot of portfolios around the region and built up a wide range of experience.

Decision-making, planning and implementation support

We provide expertise to help the insurer during their innovative phases where they venture into the unknown. Studium’s accumulated knowledge is on offer.

  • Traditional methods (multiple linear regressions for smaller portfolios, GLMs if enough data is available). These are often required by market regulators.
  • Price optimizations. We collect competitors prices and learn their pricing models, then optimize our client’s prices to perform bests in the now known competitive environment.
    This can be a simple price level optimization where we determine the optimal price level of a GLM with the maximal margin, or a per-factor optimization with a genetic algorithm.
  • AI backed methods. Studium has its own pricing AI that learns the market from success rates of given quotes, and adjusts accordingly, while keeping risk based models and profitability in mind.
Building of predictive models

We help insurers to find the best model predicting targeted outcomes. (Finding good f: X->Y multivariate functions, both for classification and regression problems)

Our recommend tool is artificial neural networks for their outstanding performance and flexibility in most situations, but we cover other mainstream methods as well.
Recently built examples are support vector machines (for smaller-scale data sets), random forests, various bagged and boosted models, several linear model variants.

This service is typically used for:

  • pricing
  • pricing adjustments based on quotation success feedback
  • continuous renewal success optimization
  • cross- and up-selling recommendations (e.g for intermediary facing systems)
  • fraud detection:
    • marking suspect claims for review, helping with optimal use of claims team resources
    • marking medical providers, motor workshops, agents for review
  • ‘unknown’ problems, models for utilizing unusual data sets (e.g. Strava, Fitbit data for health or life insurance, telematics data for usage-based motor, URA data for property)

After the models are built, we provide support for implementing them in production systems.

We offer long term tracking of model performance with updates where beneficial.

Insurance Maths & IT