Documentation Index

Fetch the complete documentation index at: https://docs.allyy.io/llms.txt

Use this file to discover all available pages before exploring further.

​

Regression models (DM Expected Amount as an example)

​

Prediction Accuracy Summary

The Prediction Accuracy Summary table provides a detailed view of how accurately the model predicts donation amounts for each campaign on the holdout dataset. Because the holdout consists of past data the model has never seen, this table reflects how well the model would have estimated revenue in real historical campaigns. Each row represents a campaign and compares predicted and actual values, along with multiple error metrics.

​

What Each Metric Means

avg_pred - The average predicted donation amount for the campaign. avg_actual - The average actual donation amount observed in the holdout period. MAE - The average absolute difference between predicted and actual amounts. Lower values indicate more accurate predictions. wMAPE - The average percentage error weighted by actual values. This reflects how large the prediction errors are relative to true donation amounts. sMAPE - A symmetric percentage error metric that balances over- and under-predictions.

​

What this table tells the user

Prediction accuracy by campaign Differences between avg_pred and avg_actual show whether the model tends to over- or under-estimate revenue for specific campaigns. Error variability Higher MAE or percentage errors indicate campaigns where predictions are less reliable. Campaign-level performance This view helps identify which campaigns are well modeled and which may require further tuning or segmentation.

​

Amount Share by Score Group

This chart compares individual predicted donation amounts with actual amounts from the holdout dataset. Because the holdout consists of past data the model has never seen, this view shows how accurately the model estimates donation values in a realistic setting. Each point represents one donor. The diagonal line represents perfect predictions (predicted = actual), while the dashed bands indicate acceptable error ranges of ±10% and ±20%.

​

What this chart tells the user

Prediction accuracy Points close to the diagonal line indicate accurate predictions, where the model correctly estimates donation amounts. Error distribution The spread of points around the line shows how much predictions deviate from actual values across different donation levels. Tolerance bands Points within the ±10% and ±20% bands represent acceptable levels of prediction error, helping users assess how often the model meets accuracy expectations. Outliers and extreme errors Points far from the diagonal highlight cases where the model significantly over- or under-predicts donation amounts. Model behavior at different values Larger deviations often appear at higher predicted amounts, indicating that high-value donations are harder to estimate precisely.

​

Calibration by Score Group (10 = highest predicted)

This chart compares the average predicted donation amount with the average actual amount across score groups in the holdout dataset. It shows how well the model is calibrated across different value segments.

​

What this chart tells the user

Calibration quality. When predicted and actual values are close, the model is well calibrated. Systematic bias. If predicted values are consistently below or above actual values, the model is under- or over-estimating donation amounts. Consistency across groups. A smooth and aligned pattern indicates stable model behavior across segments. Model reliability. Good calibration is critical for making accurate revenue forecasts and planning campaigns.

​

Absolute Error by Score Group

This chart shows how prediction error varies across score groups in the holdout dataset.

​

What this chart tells the user

Error increases with value Higher score groups tend to have larger errors because higher donation amounts are more variable and harder to predict. Stable low-value predictions Lower score groups typically show smaller errors. Risk in top segments While high-score groups drive most revenue, they also carry higher uncertainty.

​

Amount Share by Score Group

This chart shows how the total donation amount from the holdout dataset is distributed across the model’s score groups.

​

What this chart tells the user

Higher score groups generate most of the revenue. Lower score groups contribute little. Score-to-value alignment. Targeting implications.

​

Campaign Error Summary

The Campaign Error Summary table provides a detailed view of how accurately the model predicts donation amounts for each campaign on the holdout dataset.