Reading your forecast
How to interpret point forecasts, prediction ranges, selected models, quality, and cold starts.
The forecast chart
The Forecast view places historical demand, the future point forecast, and optional backtest values on the same monthly timeline. Each forecast value is y_hat for one DFU and one future month (ds in exports).
The chart also draws 80% and 95% prediction ranges. Treat the point forecast as the model estimate, and the bands as the range the engine associated with that estimate — not as a guaranteed upper or lower bound.
When several series are visible at once, the displayed range is the summed range of those series. That total is not a statistical interval for the aggregated demand.
You can switch the measure between units and growth %, and compare against the previous period or the previous year.
Composition and breakdown
The Composition view shows how the selected filters break down future demand (stacked or heatmap), with a drill-down table by hierarchy level. Use it to find which families, products, or locations drive the total.
Open a series for the selected model, score, dimensions, and a per-series export.
Selected model and score
Each DFU result contains:
| Field | Meaning |
|---|---|
| best_model | The model selected for that DFU from the configured candidates |
| metric | The metric configured for model selection |
| score | The selected model's value for that metric; it may be unavailable for a cold-start DFU |
| cold_start | Whether the DFU was forecast without normal historical model selection |
| cold_start_method | The fallback method used when cold_start is true |
The available configured metrics are:
| Metric | What it measures |
|---|---|
| WMAPE | Volume-weighted percentage error — larger series weigh more |
| MASE | Error on the series’ own scale — comparable across items of different volume |
| MAE | Mean absolute error, in the same unit as demand |
| RMSE | Penalizes large misses; useful when peaks matter more |
| MAPE | Simple percentage error — small series can distort the result |
Lower scores generally indicate less error when DFUs use the same metric. Do not compare raw MAE or RMSE values across DFUs with very different volumes without considering scale.
best_model is the selected result, not a complete model leaderboard. Model details on the page show the distribution of winning models and the score histogram for the run metric.
Quality
Quality is a volume-weighted reading of the visible series. Lower is better. For WMAPE and MAPE the bands are:
| Label | Score |
|---|---|
| Reliable | below 0.15 |
| Moderate | 0.15 up to 0.30 |
| Low confidence | 0.30 or above |
New product is the cold-start state: historical forecast accuracy is unavailable, so those series are not scored on the same bands.
Filter by quality, by method (Normal vs Cold start), by hierarchy values, and by period. Search by name or code. Exports respect the filters on the page.
Cold starts
A cold start is a DFU that cannot use the normal historical evaluation. The result is labeled New product and uses one of these methods:
| Method | Meaning in the product |
|---|---|
client_relation | Explicit analogous product — the relation you configured |
similarity_analog | Similar product |
hierarchy_share | Parent-category distribution |
mean | Historical group average |
When the method is client_relation, confirm that the source DFU is still a meaningful analogue. A cold-start score can be null because there is not enough actual history for the selected metric.
Reviewing the portfolio
Start with DFUs that combine material forecast volume with a high score (worse error). An aggregate portfolio value can hide a small number of operationally important DFUs.
Then check:
- whether the historical series matches the cleaned monthly source data;
- whether the selected metric is appropriate for the DFU's scale and zero frequency;
- whether the DFU is a cold start;
- whether known-future regressors were populated in the uploaded future rows.
When the forecast disagrees with the commercial view
Check in this order:
- History — does monthly
demandreflect the business event correctly? - Dimensions — does the selected hierarchy define the intended DFU?
- Regressors — was a planned event present in the input, including future months when known?
- Cold start — did the engine use a fallback or client relation?
- Configuration — were the intended models, horizon, metric, and reconciliation selected?
If the input itself is wrong, correct it in a new planning scenario. If only the run parameters need to change, use Adjust and reprocess.