I find this discussion particularly relevant. In my work, I often use time series analysis and predictive modeling to estimate future trends based on historical data.
I would add that while foresight approaches that do not rely on past performance are essential, especially in contexts characterized by high uncertainty or limited data, predictive methods grounded in historical data remain among the most robust tools we have when sufficient and reliable data is available. These methods allow us to identify patterns, quantify trends, and generate evidence-based projections that can effectively complement more qualitative foresight approaches.
With the advancement of AI and machine learning, we now have the capacity to go further by integrating large volumes of data across multiple projects, regions, and even donors. This creates important opportunities to build more accurate and context-sensitive predictive models, particularly when working with similar interventions within a country or sector.
However, a major constraint remains data availability and fragmentation. Data is often siloed within individual projects or organizations, making it difficult to build sufficiently large and diverse datasets for robust modeling. In many cases, data from a single project is not sufficient to support reliable predictions.
One potential way forward would be to strengthen national ownership of project data. Governments could play a key role in consolidating data generated across projects into centralized and accessible databases. If well designed, such systems could support research, inform project design, and enable more rigorous ex-ante analysis of potential success or failure.
In that sense, I see strong complementarities between foresight methods and predictive analytics. Foresight helps us explore uncertainty and alternative futures, while predictive models help us quantify likely trends where data allows. Bringing both together could significantly strengthen evaluation practice and decision-making.
RE: From Hindsight to Foresight: How Evaluation Can Become Future-Informed
Canada
Rhode Early Charles
Posted on 25/03/2026
I find this discussion particularly relevant. In my work, I often use time series analysis and predictive modeling to estimate future trends based on historical data.
I would add that while foresight approaches that do not rely on past performance are essential, especially in contexts characterized by high uncertainty or limited data, predictive methods grounded in historical data remain among the most robust tools we have when sufficient and reliable data is available. These methods allow us to identify patterns, quantify trends, and generate evidence-based projections that can effectively complement more qualitative foresight approaches.
With the advancement of AI and machine learning, we now have the capacity to go further by integrating large volumes of data across multiple projects, regions, and even donors. This creates important opportunities to build more accurate and context-sensitive predictive models, particularly when working with similar interventions within a country or sector.
However, a major constraint remains data availability and fragmentation. Data is often siloed within individual projects or organizations, making it difficult to build sufficiently large and diverse datasets for robust modeling. In many cases, data from a single project is not sufficient to support reliable predictions.
One potential way forward would be to strengthen national ownership of project data. Governments could play a key role in consolidating data generated across projects into centralized and accessible databases. If well designed, such systems could support research, inform project design, and enable more rigorous ex-ante analysis of potential success or failure.
In that sense, I see strong complementarities between foresight methods and predictive analytics. Foresight helps us explore uncertainty and alternative futures, while predictive models help us quantify likely trends where data allows. Bringing both together could significantly strengthen evaluation practice and decision-making.