Demand Forecasting with Azure ML Service - Process flow
1. Gather historical data – You may wonder what should we do if do not have past transaction history in D365FO as we just went live – no worries MSFT got your back here. If we do not have enough initially data initially we can use DMF to import external historical demand (there is an OTB entity herefore: Historical external demand for V2 entity. Historical transactional data from the Supply Chain Management transactional database is gathered and populated in a staging table in D365FO. Via the entity we can hit that staging table. This staging table is later fed to a Machine Learning service. Alternatively, we can simply use past transactions in D365FO.
2. Generate baseline forecast
When generating the statistical baseline forecast we can select the forecast buckets (month, day, weeks) The number of buckets to generate a forecast for is set in the Forecast horizon field. Please note that if you are using the forecast strategy Copy over historical demand, the end of the historical horizon is ignored. In our case we use azure ML service so we use that field. To prevent confusion in master plans, a certain number of forecast buckets can be frozen. This number is set in the Freeze time fence field. On the Adjusted demand forecast page, the cells for the frozen buckets are disabled. (can’t be adjusted) Manual adjustments made in previous demand forecasting iterations can be automatically applied to the new baseline forecast if the Transfer manual adjustments to the demand forecast check box is selected. I can run it in batch and with reoccurrence if I want.
3. View generated forecast in D365FO – update values if needed manually (manual adjustments)
On the adjusted demand forecast page, You can do manual adjustment on the here. (see bold)
On the Statistical baseline forecast generation history page, you can review the past run. (select Statistical baseline forecast on the left hand side under session log time) – you can see what steps/tasks the system went through – it filled e.g. 934 records to the staging table.
(select Authorize on the left hand side under session log time) you can see that e.g. 800 records were authorized and are now picked up by master planning (if included).
You can also see the forecast horizon and current forecast dimensions as well as adjust such.
Under Forecast line detail you can see the actual historical demand vs. the historical forecast (means trying out the equation with historical data points – we specified 20% here e.g.) as well as the MAPE (9.24) as well as which forecasting model was chosen based on lowest MAPE. (I let it run with "ALL" before so D365FO determined which forecasting model is the most accurate - disadvantage is a longer run execution - short, performance impacts)
4. Authorize forecast for use in planning and other activities..
Not all forecast data must be authorized immediately. You can specify the start and end dates of the period that the forecast is authorized for. This functionality lets you freeze specific buckets for use in master planning e.g.
5. Master plan run
Run master plan and create item requirements based on demand forecasts. (if you included such in the master plan configurations)
Master plan configuration to include demand forecast: