Create a forecast manually

Overview

The application provides a flexible way to create forecasts for the products. The following are key features

• Ability to create a forecast for specific items with different forecast parameters.

• Ability to create a forecast as a scenario before publishing it to demand planning.

• Ability to adjust the forecast parameters based on the type of the product family

Forecast creation

Click the Create Forecast button on the forecast list screen to create a forecast. This will navigate you to the page where you can create a new forecast.

Steps to create a forecast

Attribute selection

• Select the filter attributes.

1. Add the items from the filters section to create a forecast for the selected criteria. 

2. You can also select existing data filters to create a forecast.

3. Click the Apply filters button after selecting the items from the selected attributes.

• Hierarchy level

By default, the selected hierarchy level is Product: All: All. If only the product attribute is added, this hierarchy will remain unchanged. 

However, if you select any other attribute, such as Customer or Location, the hierarchy will reset accordingly, as shown below.

To create a forecast, you must select the hierarchy level. Click the select hierarchy edit button to select the level. Once clicked, the pop-up below will appear on the screen.

Select the level in the pop-up to see the results. If the results appeared blank, the selected input measure does not have data for the selected hierarchy level.

In this case, you can change the input measure to see the search results for the selected hierarchy level.

Forecast settings:

Time range and Input measure are the default settings on the create forecast screen.

Time range: The default time selected is the Next 12 months. Clicking the time range opens a pop-up with available time ranges.

What to forecast: The application picks one measure as a default input measure from the available measures in the forecast. To select other input measures, click the input measure field, and a dropdown will appear below.

Advanced settings

Enable the Advanced Settings toggle to access additional configuration options for forecast creation. This will allow you to customize parameters such as outlier removal, confidence levels, forecast methods, and other advanced forecasting preferences.

Once the Advanced Settings toggle is enabled, all forecast configuration options will be displayed on the Create Forecast page. 

After enabling advanced settings, all forecast settings will be displayed. These settings are similar to those created for the template.

1. Time horizon: The Historical Period and Forecast Period settings allow planners to define the time frames for analysis and forecasting:

   1. Historical Period: Planners can choose a specific date range (year and month) or select the "Entire History" option, which includes all available historical data for the selected items.

   2. Forecast Period: Planners can specify the year and month to generate forecasted values.

2. What to forecast: By default, Statistical forecast units and statistical forecast revenue are pre-configured output measures for the forecast creation.

   1. Output measure: To generate the forecast results, select units or revenue as an output measure.

   2. Input measure: The input measures will appear in the dropdown based on the selected output measure type.

3. Influencing factors include changes in market positions, such as the Consumer Price Index, Oil/gas prices, Raw material Costs, and Retail sales. Adding influence factors to the forecast is optional, but you can do so to explain the reasons behind changes in sales and demand. Selected factors will appear on the screen. You can remove them by clicking on the remove button.

4. Forecast methods: Firstshift.AI leverages Machine Learning, Deep Neural Networks, and advanced statistical models to generate the most accurate and reliable technical forecasts. Our system intelligently selects the best forecasting technique based on your data patterns, ensuring optimal results.

   1. The best model (let the forecast engine choose): If this option is selected, the system automatically determines the best forecasting model based on historical data patterns. AI-driven algorithms analyze trends, seasonality, and demand fluctuations to pick the most suitable model.

   2. Let me choose: If this option is selected, the planner can manually select a forecasting method from a dropdown menu. This allows planners to apply specific models, such as ARIMA, Holt-Winters, or Moving Averages, based on their preferences or business needs.

5. Remove outliers: Removing outliers from the data helps create more accurate forecasts by eliminating extreme values that could distort trends. However, it can also undermine the accuracy of statistical patterns, particularly if those outliers reflect recurring business trends.

   1. Selecting ‘Yes’ will remove the outliers caused by temporary events that are unlikely to recur.

   2. Selecting ‘No’ will keep outliers that represent important business events (e.g., seasonal spikes or significant product launches).

6. Confidence level: You should carefully select the confidence level for your forecast, as it affects the reliability of your predictions.

   • Higher confidence levels (e.g., 95–97.5%)
     These intervals give you greater certainty that the true value lies within the range. However, to achieve this higher certainty, the prediction interval must be wider—covering more possible outcomes. Wider intervals are more conservative and safer when you have high stakes or limited tolerance for missing actual values.

   • Lower confidence levels (e.g., 80%)
     These intervals are narrower, which makes them look more precise. But because they are tighter, there is a greater risk that the actual values will fall outside the range. Lower levels are useful when you’re willing to accept more risk of error in exchange for a sharper, more specific estimate.

   • Practical guidance

   • Use higher confidence levels when the costs of under- or over-estimating are significant, or when you want a conservative forecast.

   • Use lower confidence levels when you need narrower ranges for planning or decision-making and can tolerate more misses.

   • Remember that uncertainty grows as you forecast further into the future; choose a balance between confidence and interval width appropriate to your planning horizon.

7. Disaggregation rules: 

• Consider the historical data of individual lower-level items.

This option allows the forecast engine to consider the historical data of each lower-level item while disaggregating the forecast from higher to lower levels. This way, the disaggregated forecast at the lower-level items is based on the weighted average of the historical data.

• Consider the historical pattern of individual lower-level items.

This option allows the forecast engine to consider the historical trend of individual lower-level items while disaggregating the forecast from higher to lower levels. For example, the forecast for items in the family is on an upward trend. However, the sales of one item in that family have fallen in the last few months. In this case, the new forecast disaggregated at the item level should consider the falling trend and predict accordingly.

Once all the configuration is made, click the Run forecast button to create a forecast job.

Clicking Run Forecast opens a pop-up where you must decide the forecast's run options.

The application will automatically fill in the name of the forecast job. The name can be edited based on the planner's preferences.  

Run as a scenario: 

This option allows users to run forecasts, verify their accuracy, and adjust the forecast algorithm by modifying parameters. However, the forecast results will not be copied to the downstream time series. 

For example, if a tenant configuration stores forecast results in Statistical Forecast Units, these results are typically copied to the Consensus Demand Plan, Sales Forecast Units, and Marketing Forecast Units. When this option is selected, the application will not copy the statistical forecast units to these time series, allowing users to test and refine forecasts without affecting other business functions.

Run and publish forecast: 

This option runs the forecast and copies the results to the downstream time series. Consider a workflow where Statistical forecast units (generated by the forecast engine) will be copied to Sales manager forecast units, Marketing manager forecast units, Consensus plan units

Statistical forecast units →  Sales manager forecast units →  Marketing manager forecast units →  Consensus plan units

• Copying forecast results to the downstream time series

The application copies the statistical forecast results to downstream measures only for periods without overrides.

If the planners override the downstream measures for a few periods, the new forecast results will not be copied to those periods. However, they will be copied to the period in which the planner made no overrides.

For example, consider an item “Etab 2000 v1: Walmart: All: All.”

• The planner has submitted overrides for the Sales manager's forecast from Jan 2024 to April 2024

• A new statistical forecast is generated from January 2024 to December 2024.

• The new forecast will not be copied to the Sales manager forecast from January 2024 to April 2024, as the user has already submitted overrides. But the forecast will be copied from May 2024 to December 2024

To check the status of the created forecast, click on the "Refresh" button as shown below.