Quantitative conclusions should withstand technical scrutiny.
The central question is whether the data, method, assumptions, validation, and interpretation actually support the conclusion being claimed.
Statistical evidence & quantitative analysis
Statistical methods are useful only to the extent that they fit the question, data-generating process, and inferential claim.
Representative areas
- Regression and multivariable analysis
- Statistical inference and uncertainty
- Interpretation of uncertainty and quantitative evidence
- Forecasting and time-series analysis
- Outlier and anomaly identification
- Validation and reproducibility
- Design and interpretation of quantitative analyses
Predictive model evaluation & validation
A technically sophisticated model can still be unreliable, misapplied, poorly validated, or incapable of supporting the proposition being asserted.
Representative areas
- Model specification, performance, and validation
- Stability analysis and supporting quantitative evidence
- Model interpretability and variable-importance evidence
- Variable relationships and association measures
- Subgroup performance and fairness-related statistical analysis
- Machine-learning model evaluation
- Transparency, auditability, and implementation limitations
- Fit between model outputs and the decision process in which they are used
Data quality & reproducibility
Many analytical disputes begin before the model is ever run: with the source data, inclusion rules, transformations, joins, exclusions, missingness, and analytical pipeline.
Representative areas
- Data provenance and integrity
- Analytical transformations and cohort definitions
- Missing-data issues
- Reproduction of reported analyses
- Large-scale administrative and operational datasets
- Pipeline and workflow review
- Consistency between data, code, outputs, and reported conclusions
Algorithmic decision systems
When quantitative systems influence consequential decisions, the technical system and the surrounding institution need to be understood together.
Representative areas
- Scoring, classification, and risk-assessment systems
- Model use in operational settings
- Outcome definitions and decision thresholds
- Reliability, verification, and auditability
- Human oversight and model reliance
- Translation of technical outputs into institutional decisions
Insurance & regulated model review
Applied experience includes technical review of predictive and machine-learning models in insurance regulation and analysis of complex regulatory and financial data.
Representative areas
- Insurer-submitted AI and machine-learning models used in pricing, ratemaking, and underwriting
- Model specification, stability, interpretability, and variable relationships
- Subgroup and fairness-related statistical analysis
- Regulatory, financial, and administrative data quality
- Exploratory financial-distress modeling and anomaly detection
The focus is statistical and methodological. Actuarial opinions, legal conclusions, and regulatory-compliance determinations remain outside the scope of this practice unless independently supported by appropriate qualifications.
Engagements are accepted only where the requested work falls within an appropriate scope of expertise.
Need to evaluate a quantitative claim?
Start with a brief, non-confidential description of the issue.