Risk, measured before it costs you.
Cerdata builds actuarial and machine learning models for credit scoring and anomaly detection — for SMEs and financial institutions that do not always have an in-house data team.
83.4% Model accuracy |
< 3 weeks Time to production |
0.84+ Discrimination ROC-AUC |
−28% Delinquency rate |
Results measured on out-of-sample validation sets and B2B portfolios monitored in production. |
Three disciplines, rarely brought together.
Cerdata helps SMEs and financial institutions assess and manage financial risk. We build reliable models for safer decisions, even without an in-house data team.
Our approach brings together three rarely combined disciplines: actuarial statistical rigour, financial risk analysis, and technical data science implementation — validated on a real industrial case presented below.
Actuarial Science
The statistical rigour of risk: default probabilities, prudent provisioning, and calibration on heavy-tailed distributions.
- Extreme probabilities & copulas
- IFRS 9 / Basel III provisioning
- Survival distribution calibration
Finance & Risk
Why a model matters, not just how to build it. Analysis of balance-sheet ratios, working capital, and actual cash flow.
- SME financial statement analysis
- Liquidity stress testing
- Risk-adjusted pricing (RAROC)
Data Science & MLOps
Supervised and unsupervised pipelines, real-time anomaly detection, and direct integration into your systems.
- Gradient Boosting & Isolation Forests
- Decision explainability (SHAP)
- Real-time APIs & microservices
Empirically validated on a real industrial case.
Calibrated and tested on real transaction data: payment delay reconciliation, proactive default detection, and false-positive elimination.
What we deliver.
Tools that integrate directly into your operations — not static PDF reports, but living, actionable models.
Credit scoring & counterparty risk
Financial anomaly & fraud detection
Profitability modelling & risk pricing
Integration options: sovereign cloud API, on-premise Docker container, or standalone spreadsheets / BI.
Our methodology, applied to a complete industrial case.
Every portfolio project follows the same standard as our client deliverables: real data, a benchmark against machine learning, rigorous statistical validation, and effective deployment — not a purely academic exercise.
Actuarial pricing — Motor insurance
Choosing the Negative Binomial GLM over Poisson was statistically validated (overdispersion detected, Pearson χ²/df ≈ 2.55). The XGBoost benchmark confirms that traditional actuarial risk factors (bonus-malus, vehicle age) already capture most of the signal — a conclusion directly applicable to our client scoring engines.
The convergence between the GLM and machine learning on the same dominant risk factors increases confidence in the diagnosis — a principle we apply to every model delivered to our clients.
Credit pricing — Basel II/III scoring & pricing
A critical calibration bias (4.9x overestimation) was detected and corrected through isotonic calibration — an essential validation whenever a model feeds a monetary calculation, not just a binary decision.
Detecting and explicitly correcting calibration bias reflects the validation rigour expected in a regulated credit decision.
Portfolio management — VaR & Markowitz optimisation
The PDF report is generated automatically from calculated results (no manually entered figures), ensuring full traceability from analysis to presentation.
Comparing three VaR methods on the same data reveals where normality assumptions underestimate actual risk — a reflex we apply to every risk model.
From raw data to informed decisions.
A structured four-step process to deliver a robust model without overburdening your operational teams.
Data audit & mapping
Analysis of existing data (ERP, CRM, accounting, delinquency history). Quality assessment, bias detection, and definition of business objectives.
Modelling & rigorous backtesting
Iterative development of actuarial and machine learning models. Strict out-of-sample (out-of-time) evaluation to ensure robustness in real-world conditions.
Explainability & compliance audit
Generation of individual explanations (SHAP values) and technical documentation for auditors, statutory accountants, or risk committees.
Integration & continuous monitoring
Smooth production deployment (REST API, SQL/Excel connectors, or interactive dashboard) and setup of data drift alerts.
A question, a project?
Do you have exposure, invoice, or transaction data and want to assess its predictive potential? We reply within one business day.
NDA / prior confidentiality agreement available on request.
Frequently asked questions
Define your needs in a few lines