- A deployed Interactive Price Analytics solution
- Excel of a recent vintage
In this walk-through, you will simulate the effect of a promotion. A "promotion" means simply changing the price of an item for a specified amount of time.
If you haven't downloaded the interactive Excel Worksheet yet, please download it now. It contains an Instructions tab that will help you with most common setup issues.
Open the AzureML plugin pane on the right and delete the example web services (Titanic Survival and Text Sentiment Analysis).
Click Add and paste the URL and API key of the PromotionSimulation web service from the Instructions page shown when the solution deployment completed. If you closed that page, you can return to it from the Deployments section of Cortana Intelligence Quick Starts.
In the workbook, find the Simulate Promotion spreadsheet. It has a yellow box with instructions. You have already completed the first step of those instructions (set up web service).
Follow the instructions in the spreadsheet, but choose A18 as the destination (output) cell. As datasetName parameter of the service, you can use 'latestDemoBuild' if the solution has run for a while and a model has been generated.
If the table at A18 fills out successfully, then your model has already been built by Azure Data Factory. If you get an error, please build a model according to the instructions in the next section.
Experiment with the price of Minute Maid on week 2. Note that both the sales of Tropicana and Dominicks are affected the same week (cannibalization) and sales of the same product are affected in the next week (pull-forward).
If your solution is newly deployed, the modeling pipeline may not have run and the model names 'latestDemoBuild' does not exist.
In this case, build a model from your data ad-hoc.
To build a model, open the Excel workbook and navigate to the Data and Model tab. Follow the instructions on the spreadsheet, choosing some name other than 'latestDemoBuild' for the datasetName. For forecastPeriod, choose 3.
Creating the model will take about 3 minutes for the OJ dataset and produces as output the lists of valid values for Products, Locations, etc. that the model recognizes.
Now repeat the instructions in the Promotion Simulation exercise with the datasetName being your chosen one. You may need to change the values in the CurrentDate columnn to correspond with the date range represented in the SalesDates column in the "Data and Model" worksheet.
If you have other pricing data you would like to try your model on, please format it to the same schema as the Orange Juice data and repeat the Build Model step.
The dashboards will not update (they are pointed to the 'latestDemoBuild' model and run weekly), but you can explore the elasticities, product relations and forecasts from Excel. We will introduce dashboarding of results from bring-your-own data in a future release.