Consequential Predictions

A prediction is consequential when it influences a decision that materially affects people, risk, safety, money, or access to public services.

In lead service line replacement, a risk estimate can influence whose pipes are replaced first. In gas operations, a forecast can affect staffing and response. In sewer inspection, a model can decide which videos receive scarce expert attention.

Consequential predictions raise the bar for modeling practice. Accuracy is still valuable, but it is not enough. The model also needs:

• interpretability for experts and reviewers
• uncertainty estimates that communicate what is not known
• transparency about inputs and assumptions
• fairness checks across affected communities
• accountability for decisions made with the output

This is one reason to investigate interpretable statistical models and probabilistic programming alongside conventional machine learning.