The goal of Clarifi Forecasting is to marry manager field knowledge and experience with machine learning. By fusing the two together, the impossible task of capturing a high-quality forecast becomes not only possible, but entirely achievable.
Forecasting is broken into four different pieces.
- Weekly Forecasts
- Daily Forecasts
- Events & Promotions
- The Forecast Algorithms
To access Forecasts, users can select the Forecasting tab from the Clarifi side navigation panel.
Weekly Forecast View
This section will display a chart view that has actual sales and projected sales side by side. Actual sales will only appear for days in the past.
To change the date range, click on the back facing or forward facing arrows above the chart. Users have the option to view projected data for -
- 2 weeks into the past
- 1 week into the past
- The current week
- 1, 2, and 3 weeks into the future
Users can change the forecast metric they are viewing by clicking the drop-down menu just above the graph.
The following options will be available for customers depending on how they are set up:
- Guest Count (if provided by the POS)
- Sales Item
- Labor Driver (if provided by the POS)
Daily Forecast View
Beneath the weekly forecast section will be an area displaying the day to day forecast for the week you are viewing.
Events & Promotions
After daily forecasts will be an area displaying Events and Promotions that have been applied to the week. You can see the events and promotions for each individual day by clicking on it from the left column.
You can create new events by clicking on Add an Event. Clicking the question mark modal will display a window that defines what events are.
The forecasting module will use three different algorithms whenever it is run. The primary one being a Generalized Additive Model, Clarifi will also use two different variations of the moving average calculation:
- Exponential Moving Average - Applies a filter that decreases the weight of the recent dataset (12 weeks) exponentially over time.
- Weighted Moving Average - Assigns a higher value to recent data.
The algorithm used and, by consequence, projection chosen will be based on historical accuracy, by comparing prior 2 weeks forecast to their actual amounts using three different error rates. They are as follows:
- MSE: Mean Squared Error between forecast actual values
- MAD: the Mean Absolute Deviation between forecast and actual values
- MAPE: the Mean Absolute Percent Error between forecast and actual values.