01 / Decision Test
Start with the decision that needs support.
A good research question names the action, the timing, the person responsible and the consequence of being wrong. That prevents the project from becoming a detached prediction exercise.
Research profile · 2026A / 001
Nathan Azoulay
Applied Mathematics · Data · Decision Systems
Frommathematicalmodelstoreliabledecisions.
Applied mathematics, actuarial science and machine learning for decisions that remain inspectable under uncertainty.
01 · Research workbench
Not every accurate model is useful.
The real question is whether its evidence can support a decision for someone who has to act — with uncertainty, context and limits still visible. This is the working method behind the case files on this site.
02 · Featured case
Machine learning / reserving
Master’s dissertation · 17/20CASE 01
Machine Learning for Insurance Reserving
A master’s dissertation and industry internship project exploring whether supervised machine-learning models could help prioritise open non-life insurance claim files for review.
DecisionWhich open files should be reviewed first when capacity is limited?
MethodData preparation, feature engineering and comparison of supervised models.
ScopeA decision-support study — not an automated closure system.
03 · Trajectory in one line
Follow the trajectory A quantitative path that keeps moving closer to real decisions.
2017Mathematics
2021Applied modelling
2023Risk & actuarial science
2024Insurance ML
2026Applied research
04 · Open a dossier
Have a question worth investigating?
For doctoral research, applied modelling or systems where evidence needs to stay visible, let’s discuss the next problem.