AI-driven Reference Class Forecasting
Turning data, context, and reference-class evidence into better-informed capital decisions.
Forecasts of cost, value, and downside risk, using context-specific reference classes, to reveal the key drivers of risk and value, especially where estimation errors have material consequences.
Our models combine Predictive Reference Class Retrieval, machine learning and explainable AI with project, location and macroeconomic context to identify the historical evidence most relevant to the decision at hand — and are rigorously validated out of sample.
Applications: infrastructure, real estate and urban development, energy transition, public investments, and macro-financial risks.
Empirically validated forecasts across three domains
Dike reinforcements
50% lower prediction error
Early-stage cost forecasts validated out of sample against initial project estimates.
Land prices
38% improvement
Interest rates
68% improvement
We measure forecasting performance on data that was not used to develop the models.
Traditional risk assessments often miss context
Many estimates and risk assessments still rely too heavily on internal assumptions, static benchmarks or limited historical comparison. That makes it harder to see downside risk clearly, especially when project context and macroeconomic conditions matter.
Our approach
We strengthen Reference Class Forecasting with Predictive Reference Class Retrieval (Predictive RCR), quantitative modelling and rigorous out-of-sample validation.
Predictive RCR dynamically constructs a context-specific reference class by identifying and weighting the historical observations most informative for the decision at hand.
This allows the relevant evidence to depend on context — whether it consists of comparable infrastructure projects, locations or historical economic regimes.
Using explainable AI, we show which reference observations, contextual factors and underlying drivers shape the forecast.
- Predictive Reference Class Retrieval
- Context-specific reference classes
- AI/ML-enhanced forecasting
- Out-of-sample validation
- Explainable AI and visible drivers
- Risk assessments grounded in context
From assumptions to defensible risk decisions
This helps organisations move beyond subjective assumptions and produce more transparent, more robust and more defensible risk assessments that reflect project and macroeconomic context.
More transparent
See which comparable cases, macro conditions and drivers shape the risk outcome.
Enabled by explainable AI and reference-class evidence.
More robust
Ground estimates and risk assessments in quantitative modelling and out-of-sample validation.
Built to perform beyond in-sample fit.
More defensible
Support high-stakes decisions with explainable outputs, documented assumptions and observed performance.
Designed for governance, challenge and scrutiny.
Where this approach creates value
Large-project cost and risk estimation
Construct context-specific reference classes from comparable historical projects to provide a top-down empirical forecast for project cost and risk.
Land development risk
Retrieve historical market regimes relevant to current economic and local conditions to estimate context-dependent land-price and development risk.
Interest rate and inflation risk
Identify historical periods relevant to the current macroeconomic regime to estimate risk under changing economic conditions.
Context-driven forecasting for public decision making
Support public investment and policy decisions with more context-aware risk insight.
Built for transparency, validation and explainability
We publish model documentation and validation material to clarify assumptions, limitations and observed performance.
Out-of-sample validation
Explainable AI
Observed performance
Transparent documentation
Products, tools and decision support
Explore our products and services for context-dependent risk assessment, including AI/ML-enhanced Reference Class Forecasting, risk tools and decision support grounded in transparent validation and observed performance.
See if this improves your estimates and risk decisions
Explore whether AI/ML-enhanced Reference Class Forecasting can strengthen your decisions where uncertainty, estimate error and downside risk matter.
Water authorities
Improve dike cost estimates
See how AI/ML-enhanced Reference Class Forecasting reduces estimate error on large water infrastructure projects
Infrastructure projects
Add an independent outside view to project cost estimates
Use top-down, feature-based Reference Class Forecasting alongside detailed bottom-up engineering estimates to assess whether expected project costs are consistent with empirical outcomes from comparable projects.
Municipalities & developers
Better land development risk decisions
Assess development risk in context rather than through static assumptions alone.
Financial institutions
Strengthen interest rate risk models
Analyse risk under changing macroeconomic conditions with context-dependent modelling.
Use Cases in practice
See where AI/ML-enhanced Reference Class Forecasting creates value in practice, from large-project cost estimation to land development, interest rate risk and context-driven public decision making.
Dike reinforcement cost estimation
Read how we apply RCF-AI, or AI-enhanced Reference Class Forecasting, to dike reinforcement projects. This method helps to obtain better grounded and context dependent cost estimates.
Land Development Risk
The following use case explores how one municipality improves the process of risk measurement on a portfolio of land development projects by using context dependent risk estimates
Project Risk
See how to assess various aspects of investment and operating risk in public projects.
E.g. an outdoor swimming facility or an art museum.
Loan Portfolio Interest Risk
Learn how to obtain a risk estimate for a loan portfolio that is to be refinanced using the so called ‘Value at Risk’ method.
Insights
Reference Class Forecasting
Learn about the basics of reference class forecasting and its role in predicting expectation and risks. Discover how it improves risk forecasts and project success.
Risk Aversion
Discover the psychology of loss and risk aversion in decision-making. Explore how humans avoid negative impacts and uncertainty.
Trustworthy AI: why predictive power alone is not enough
Read why trustworthy AI is essential for decision support and how validation, transparency, and reproducibility contribute to it.
Land price risk in context
We analyse historical land prices in the Netherlands and use a context-driven regime-based model to forecast the risk.
Model Validation
In this series of articles we analyze the model performance of four interest rate models and show how context adds value.
- Solvency II: EU capital requirements for insurance sector
- Wtp: new pension funds regulation in The Netherlands
- BASE method: statistical method, no context dependence
- REGIME-BASED method: context dependent statistical method
Finally, let’s put these models together and select the optimal risk model.
Interest Rate Risk
Explore the main factors that need to be considered when calculating the refinancing costs.
Our team
If you wish to receive more information or schedule a (video) call please send us aRequest for Information
Our way of work & values
question assumptions
search for main risk drivers
take context into account
iterative goal oriented
focus on details
take ownership
integrity, humbleness
open communication
Latest news
Stay updated with thelatest newsfrom Asset Mechanics. Explore our articles and stay informed about current events and trends.