Description
At the core of a rigorous price analysis workflow is price elasticity modelling and optimal pricing recommendations. The state-of-the-art modelling approach mitigates the two worst pitfalls of modelling price sensitivity from historical data: confounding and data sparsity.
Confounding is the presence of factors other than price which affect demand. We use a “double ML” approach that removes the predictable components of price and demand variation before estimating the elasticity, immunising the estimates against most forms of confounding. The solution can also be customised by an implementation partner to use your data capturing potential external demand drivers other than price. Our blog post provides additional details about the data science of prices.
Data sparsity occurs because the optimal price varies at a fine grain: businesses can set prices by item, site, sales channel and even customer segment, but pricing solutions often only give estimates at product category level because the transaction history may only contain a few sales for each specific situation. Our pricing solution uses “hierarchical regularisation” to produce consistent estimates in such data-poor situations: in the absence of evidence, the model borrows information from other items in the same category, same items in other sites, and so on. As the amount of historical data on a given item-site-channel combination increases, its elasticity estimate will be fine-tuned more specifically.
This solution analyses your historical prices and
- shows you in just a glance at the dashboard how elastic your product demand is
- provides pricing recommendations for every product in your item catalogue
- discovers related products (substitutes and complements)
- lets you simulate promotional scenarios in Excel.
Estimated cost
The estimated cost for the solution is approximately $10/day ($300/month)
- $100 for S1 standard ML service plan
- $75 for an S2 SQL database
- $75 for app hosting plan
- $50 in miscellaneous ADF data activities and storage costs
If you are just exploring the solution, you can delete it in a few days or hours. The costs are prorated and will cease to be charged when you delete the Azure components.
Getting started
Deploy the solution with the button on the right. Instructions at the end of the deployment will have important configuration information. Please leave them open.
The solution deploys with the same example data set of orange juice prices that you find behind the Try-It-Now button on the right.
While the solution is being deployed, you can get a head start and
- see what is available in the Try-It-Now dashboard
- peruse the User Guide for usage instructions from the perspective of a pricing analyst (MSFT login required)
- review the Technical Deployment Guide for a technical implementation view (MSFT login required)
- download the interactive Excel Worksheet
After the solution has been deployed, complete the first walkthrough (MSFT login required).
Solution Dashboard
The solution dashboard’s most actionable part is the Pricing Suggestion tab. It tells you which of your items are underpriced or overpriced, and suggests an optimal price for each item, as well as the predicted impact of adopting the suggestion. The suggestions are prioritised by the greatest opportunity to earn incremental gross margin.
Other tabs provide supplementary information illuminating how the system arrived at the suggestions and are discussed in more detail in the User Guide. (You must be logged in to GitHub with a MSFT Azure account while the solution is in private preview.)
Solution Architecture
The solution uses an Azure SQL server to store your transactional data and the generated model predictions. There are a dozen elasticity modelling core services, which are authored in AzureML using Python core libraries. Azure Data Factory schedules weekly model refreshes. The results are displayed in a PowerBI dashboard. The provided Excel spreadsheet consumes the predictive Web Services.
Please read the Technical Deployment Guide for a more detailed discussion of the architecture, connecting your own data and customisation (GitHub login required).
Disclaimer
©2017 Microsoft Corporation. All rights reserved. This information is provided “as is” and may change without notice. Microsoft makes no warranties, express or implied, with respect to the information provided here. Third-party data was used to generate the solution. You are responsible for respecting the rights of others, including procuring and complying with relevant licences in order to create similar datasets.