Forecast events and demand adjustments
How forecast events, promotions and past demand adjustments change the forecast.
Two things change the demand signal on top of the AI forecast:
Forecast events add a known impact to the forecast over a period: a promotion, a catalog, a planned disruption.
Past demand adjustments correct the history the model learns from: a stock-out, an outlier, a past promotion.
You create forecast events. Past demand adjustments come mostly from the Demand anomaly agent, and you can add your own.
Kinds of forecast events
Every forecast event lives in the Promotions module, in a list and a calendar. Each event has a type:
Forecast adjustment: an impact you enter yourself, in % or in units, with no promotion computation behind it. Events you create from Demand are of this type.
Price discount, Catalog, Featured placement and Other: promotions. Their impact can be entered by hand or, when the AI pricing features are activated, computed by Flowlity.
You can create events from the Demand product detail, for many products at once from the Demand list, from Product strategy, from Promotions, by importing a CSV file, or by asking the Co-planner. See Create and apply forecast events.
Which events change the final forecast
Forecast adjustments count as soon as you create them.
Promotions count only once they are Confirmed. Planned and In review promotions are shown on the Demand chart for information only.
Future and past events
A future event changes the forecast of its products during its period. In Demand, the impact appears in the Forecast events row and in the Final forecast.
The impact applies to the Flowlity forecast. On periods where you entered My forecast, or on products whose Forecast source is set to External forecast, the event shows in the Forecast events row but does not change the Final forecast.
Once the period is over, the event keeps working: Flowlity adds a past demand adjustment that removes the event's impact from the actual demand (Forecast event smoothing), so the model does not learn the uplift as normal demand. To find the products concerned, use More... > Events > Forecast event smoothing in the Demand list.
Example. You plan a +30% catalog for March on a product that usually sells 100 units. Once the catalog is confirmed, the March forecast becomes 130. After March, Flowlity takes the catalog's impact out of the March sales it learns from, so next year's forecast does not start from a promotional peak.
Past demand adjustments
A past demand adjustment replaces the demand of a past period with a corrected value. It changes Cleaned past demand, never Raw past demand. The Anomaly adjustments impact row of the Demand table shows the corrected values. A day can hold only one adjustment.
The Demand anomaly agent looks for four kinds of anomalies:
Outliers: unusual peaks that do not reflect normal customer demand.
Shortage: periods where sales were held back by stock, either 0 stock and 0 demand or stock < lot size.
Smooth past promotions: promotional spikes replaced by normal demand.
Launch detection: the end of a product's launch phase, so launch peaks do not count as normal demand.
Adjustments created by Flowlity are marked Created by Flowlity. On each one, you can change the Past demand value, delete it, or Mark as read once you have checked it. Products with unread adjustments show the Possible outlier or Possible shortage demand alerts. See Investigate and fix demand anomalies.
How the agent is set up
For each anomaly type, the agent is either active or not, and its adjustments are accepted automatically or manually (the Choose how events are accepted setting). The Configuration tab of the agent, in Agents, shows the current setup. Your Customer Success Manager can change it for you.
How events and adjustments work together
Adjustments clean the past, so every future forecast learns from a truer history.
Future events add what the model cannot know, such as a planned promotion.
Past events stop a past promotion from being learned as normal demand.
Related pages
Last updated