OneClick Augmented Forecast Tutorial
A step by step guide to master forecasting within 20mins
Getting Started Tutorial
Step 1 - Data Preparation
Download the data set here.
The basic data for demand forecasting is historical sales records. As long as you provide the three fields of date, product, and sales, you can forecast the future sales. In practice, we recommend collecting all information that affects sales such as transactions, product attributes, store attribute, promotion, sales engagement from CRM and etc.
You can use above sample data to predict the sales of next 10 weeks (from 2012/10/26 to 2012/12/28)
Step 2 - Connect to Data Sources
You can choose to upload CSV/Excel directly or setup other connections.
Step 3 - Setup Data Hierarchy
In order to facilitate data visualization later, it is recommended to define the hierarchical relationship within the data. In this example, the store type is the highest level, which is followed by store and store department.
Step 4 - Submit Forecast Task
- Prediction Column: Choose the column in the data set that you would like to predict. In this example, we choose 'Sales'.
- Date column: Choose the column which contains the date.
- Start Date: The starting date of the forecast.
- End Date: The ending date of the forecast.
- Forecast Frequency: Select if this is monthly, weekly or daily forecast during the time-span defined above.
- Aggregation Level: This multi-selector defines the forecast granular level e.g if it is on SKU-Store, please select both.
- Metric: Metric is a mathematical formula that measures the accuracy of the forecasts. Mean absolute percentage error is the most common one.
- Back-test Starting and Ending Date: You can specifically choose a time range in the past to test the consistency of the forecasting model. This time-span should be no shorter than the future forecast time-span.
- Dynamic Columns: It defines the dynamic factors that users can perceive ahead into the future - e.g., promotions, marketing campaign, holidays, weather and etc.
Step 5 - Select a Model for Future Continuous Forecast
We select the best model automatically for you given the metric defined in last step. You can also do additional analytics by slicing and dicing the data and making changes on the chart setting.
Step 6 - Schedule the Time for Continuous Forecast (Optional)
Schedule when to run the forecast automatically in the future. Click 'View Dashboard' to proceed.
Step 7 - Upload Latest Data If Continuous Forecast is Scheduled, Skip This Step Otherwise
Once the continuous forecast is setup, let us say daily forecasting, the system will fetch the data automatically from remote SQL database or FTP server at scheduled time and generate the latest forecast. If you are uploading the file manually, please make sure you do so before that scheduled time. There is no need to upload the entire history, daily delta change is enough.
Step 8 - Dashboard
You can view both KPIs and reporting on the dashboard. The charts are configurable to support slice and dice analysis.
Step 9 - Forecast Adjustments
You can see all the forecasting results here and make any adjustment needed. To override on the higher aggregation level, it is more convenient to drag the forecast to the desired value directly on the chart and the adjustments will be applied automatically to all the products within that aggregation.
Step 10 - Automatic Alerting
There are two ways users can be alerted by forecast outliers.
- OneClick.ai can proactively push the alerts to users if it observes the forecast outliers within the data.
- The users can also setup their own alerts on this interface. For example, once the accuracy of the store 29 is lower than 80%, an alert is generated and an email is sent to the recipients . Users can define the name of the alert, the triggering condition, and the severity level of the alert. If you want to monitor a specific product or store, you can do so by defining the filtering conditions accordingly.
The users can click on any displayed alert to view more details.