Forecasting
Applies to Write-Back Server and Write-Back Cloud
Question
How to create a simple forecast with Write-Back?
Forecast is an important management process and every main process the company engages in, tends to be based on some prediction to support them. Forecasts can be quantitative or qualitative.
Quantitative forecasting is about hard numbers and historical data: these methods try to remove the human element from the analysis, looking just at the data and typically leveraging machine learning algorithms.
Qualitative forecasting has less to do with numerical analysis and more with experience and expertise. This type of forecasting relies upon your people’s knowledge to provide insights into future outcomes.
Write-Back solves the challenges imposed by doing qualitative forecast on Tableau. With Write-Back, the user(s) responsible for the forecast can easily input the predicted data straight from the Tableau visualizations.
Answer
The attached workbook has the full solution implemented and is using Write-Back public demo environment. For more information on using it check our Public Demo instruction page.
Attachments & Related Links
Forecast.twbx Write-Back Configuration Connecting to the Dataset
Step 1: Create an Anchor Worksheet
All Write-Back instances have a common starting point a source worksheet. You cannot have a Write-Back instance per se, it must always be related to a visualization.
A forecast is intrinsically related to time periods. Whether be months, quarters, or years, this kind of prediction always has a time dimension associated. This, combined with the premise above, gives you the starting point for the construction of a forecast - a worksheet with a time dimension.
Step 2: Set Up Write-Back for Forecast Submission
To achieve a forecast dataset configuration, you need to have a table typically with the time dimension on columns and the KPIs to be populated by users as rows. Thus, users are able to submit forecast values for each period. You typically also want to give some context to the forecast submitted and this can be a comment, version, or a status that the users submit and also a dimension associated with the forecast.
Add Write-Back to the dashboard and your configuration should look like this:
Time Dimension Field on Columns.
Sales Forecast (Input Field) on Rows.
One Input Field on the Main Form - e.g. Comments or other to provide context.
One Source Field on Configure Fields - corresponding to the dimension associated with the forecast.
Step 3: Submit Forecast
With the data set created and the submission form set up, it's time to use Write-Back to insert the forecasting data.
To do this, you need to select one mark from the worksheet. In this case, the user must select only one, and here is why: Write-back uses a combination of columns/marks to insert data into the database. So, for each mark, Write-Back inserts twelve rows.
Step 4: Connect to Forecast Data
It is crucial for the organization to achieve the proposed objectives to maintain regular control over the actual data vs forecast. With Write-Back, it is possible to create visualizations that help in this control. All you need to do is create an integrated view with both data sets.
The forecasting data is now in the database ready to be used as a data source. However, if you look closely at the structure, you will notice that the date field that identifies the month is a numeric value and not a date in the standard format. This is because when you set up Write-Back, the date field is set to discrete, which causes the database field to take the value of the month number. To get a valid date field, you need to create a calculated field. Also don't forget to filter Write-Back table by IsActive set to True to ensure you only get the latest snapshot provided by the users for each record.
Step 5: Create a Combined View with Real and Forecast Data
Finally, you can create a view comparing actual sales to forecast sales. The ideal is to have both measures on the same view. To do this, you first need to combine the two data sources using a time dimension as a blend relationship, and then create a combined view.