Forecasting legacy income with confidence for a leading UK charity

Machine learning | Regression | Bayesian forecasting

Legacy gifts are the largest single source of fundraising income for one of the UK's leading charities. We built an interpretable forecasting model that the charity's own team can maintain. It gives leadership a reliable, evidence-based view of medium-term income to support strategic investment decisions.

The challenge

The charity was refreshing its strategy and needed confidence in how much it could afford to invest over the coming years. Legacy income underpins those plans, but many internal and external factors make it volatile:

• Distortions in data caused by Covid
• Long-term demographic shifts
• Changes in the charity's share of the legacy market
• Irregular patterns caused by probate backlogs
• Large one-off legacies that skew historical trends

Existing approaches struggled to separate the underlying signal from this noise. Leadership needed a forecast they could trust and explain to others.

Our Approach

We worked closely with the charity's data team, Chief Financial Officer and Chief Customer Office to understand their data; their forecasting pain points and how the outputs would be used in decision making.

Exploratory data analysis and correlation studies identified what actually moves legacy income. Total donation values tracked closely with house prices and equity market performance. Different legacy types responded to different economic drivers, so we built type-specific forecasts. Residuary legacies, which drive most of the portfolio, proved the most sensitive to economic conditions.

Developing the Model and Quantifying Uncertainty

We developed an empirically driven machine learning regression model that combines the charity's internal data with verified third-party inputs, including national population projections and sector legacy forecasts. Features were selected and engineered from the correlation analysis, so each input has a clear, evidenced link to income.

A Bayesian forecasting layer turns single-point predictions into probabilistic ranges. Leadership can see both the most likely outcome and the realistic upside and downside, which is essential when deciding how much investment is affordable. We deliberately avoided black-box complexity. The model is transparent and interpretable, so finance teams can understand, challenge and explain every forecast.

The Impact

• Stronger strategic decisions: a reliable, evidence-based foundation for assessing the affordability of long-term investments as part of the strategy refresh.
• Confidence in the numbers: forecast ranges that reflect real uncertainty, which supports more robust planning and risk appetite discussions.
• Built to last: a transparent model that the charity's internal teams can maintain and update independently, with no long-term reliance on external support.