Forecast events and demand adjustments
Events lifecycle, anomaly detection, and demand adjustments.
Forecast events and demand anomaly corrections are the two main ways the demand forecast gets adjusted beyond the AI's baseline prediction. Events are created by you; anomaly corrections are proposed by the Demand Anomaly Agent. Both feed into the final forecast visible in the Demand module.
Forecast events lifecycle
A forecast event models a one-off or recurring occurrence that affects demand — promotions, product launches, market disruptions, or seasonal campaigns.
Create — Define the event in Forecast events settings, or directly from the Forecast events row in the Demand product detail data table. You can create a brand-new event or link to an existing one. The impact can be expressed as a percentage or in units.
Link to products — Select which products are affected, either individually or by tag. When creating from the Demand detail view, the current product (or aggregation of products) is pre-filled. You can optionally scope the event to specific customers.
Impact on forecast — During the event period, Flowlity adjusts the demand forecast for the linked products by the specified amount. The adjustment appears as the "Events" series in the Demand chart and "Forecast events" row in the data table.
Edit or delete — Click an existing event in the data table row or directly on the chart to open the edit dialog. Changes apply to all linked products.
Anomaly detection and cleaning
The Demand Anomaly Agent runs weekly and identifies three types of anomalies in your historical demand:
Outliers — Unusual demand peaks that don't reflect typical patterns.
Shortages — Zero-demand periods caused by stockouts (not genuine lack of demand) or periods where stock below lot size suppressed sales.
Smooth past promotions — Promotional spikes that are smoothed out to give a cleaner baseline for forecasting.
Each anomaly correction can be:
Accepted — The correction is applied to cleaned past demand.
Rejected — The original raw value is kept.
Not treated — No action taken yet.
Modified — You adjusted the agent's proposed value.
How events and anomalies interact
Forecast events affect future demand (forward-looking), while anomaly corrections clean historical demand (backward-looking). Both ultimately influence the final forecast:
Cleaned historical demand feeds into the AI model, so anomaly corrections improve future forecast accuracy.
Forecast events add explicit adjustments on top of the AI forecast for known upcoming changes.
Configuration
In the Agents module, you can configure the Demand Anomaly Agent to handle corrections automatically (accept all suggestions) or manually (review each one). You can also activate or deactivate specific anomaly types.
Related pages
Create and apply forecast events — Step-by-step guide.
Investigate and fix demand anomalies — How to review and act on agent suggestions.
Demand forecasting explained — How the probabilistic forecast is generated.
Forecast events — Where events are created and managed.
Demand — Where event impact is visible on the chart.
Agents — Monitor and configure the Demand Anomaly Agent.
Last updated