parallax background

Proxy modeling

Request a demo


steps to create your models


options of cloud deployment


plots and tables to analyse results


increased predictive force on case study

Life Insurance Proxy Modelling

Our product is dedicated for actuaries working on capital modelling in Internal Model. It allows to reduce risk and produce more accurate SCR figures as well as improve cost-effectiveness of proxy modelling process by:

  • Data preprocessing – automatically handles missing values in risk scenario data.
  • Model Selection – automatically recognizes important interactions between risk drivers, models traditional actuarial estimators as well as novel machine learning algorithms such as boosting machines or neural networks together with Explainable AI to meet regulators interpretability requirements.
  • Risk Monitoring – compares models and visually inspect results to select the best strategy for your proxy modelling problem. Produces comprehensible visualizations to understand effects on SCR of your modelling.
  • Model Deployment & Monitoring – save your project or deploy the best models for production purposes.

Want to see it in action? Request a demo

SCR calculations boosted with AI

Traditionally, the loss function problem aka proxy modelling is solved by manual selection of scenarios, time-consuming search for meaningful interactions between risk drivers and linear estimator fit (or linear stepwise selection algorithm). The linear models do not catch the non-linearity and complexity of cash-flow models in products such as financial guarantees which in turn can lead to inaccurate SCR figures, questions from regulators and reruns of the Solvency II reporting.

ActuAI platform comes with a set of interpretable AI techniques, which reduce this tedious and error-prone process from weeks to hours.

(a) Linear Backward Stepwise Selection BEL residuals

(b) Gradient Boosing BEL residuals

Figure shows the distribution of normalized residuals between true BEL (Best Estimate Liabilities) and predictions from (a) linear backward stepwise selection algorithm and (b) gradient boosting. Both estimators serve as interpretable proxy models. Pattern of linear model residuals shows that although we have added most important interactions to the training data, the algorithm still does not catch all non-linearity and complexity of a cash-flow model. In turn, residuals of gradient boosting estimator seems to be distributed randomly, which increases the accuracy of SCR. Gradient boosting is interpretable thanks to Break Down and Partial Dependency Plots coming from our Explainable AI module.

Automation for managers + Flexibility for actuaries

Our automated machine learning algorithm and comprehensible UI allow building, deploying and analysing results of a proxy model without any programming knowledge in a few minutes.

On the other hand, for most advanced users we offer a possibility to load their own models or data transformations and still benefit from model serving infrastructure as well as other parts of the platform such as XAI. Moreover, your IT department can use our API to plug selected parts of our technology into existing solutions & products.

Simplified Capital Modelling
with ActuAI Platform

Use ActuAI cloud-based platform to automate SCR calculations from weeks to hours thanks to simplified workflow, direct deployment and intuitive reports.