Public Preview: Create and manage models to make product recommendations

Article
08/16/2017

Applies To: Dynamics 365 (online), Dynamics CRM Online

Imagine being able to make product recommendations to your customers when they select an item for purchase. When you connect Dynamics 365 (online) to the Microsoft Cognitive Services recommendation service, you can. Use the Cognitive Services recommendation service to build an advanced recommendation model for automatic cross-sell product recommendations that are based on historical transaction data.

Important

The Cognitive Services Recommendation API required for this feature will be removed as of February 15, 2018, which means this feature will no longer work after that date and cross-sell product recommendations based on an active recommendation model will return an error. We recommend that you deactivate existing active product recommendation models and use Product Relationships. More information: Recommendations API and Define related products to increase chances of sales

The Dynamics 365 product catalog includes basic modeling ability to link various products for cross-sell/upsell/accessory recommendations. The limitation with these recommendations is that they require a customizer or business analyst to maintain hard links, and often require making assumptions about which products will sell together. While these assumptions may be valid, they may not reflect the real world, with users often buying other products together. Even if customizers or analysts have specific data points for maintaining these links, they will keep evolving over time with new or retired products and services. Maintaining these links requires the constant overhead and complexity of analyzing how to rank the recommendations and imagining all possible combinations of products that can be sold together. A better solution would be recommendations that use real-world transactions as a basis, and can evolve over time without requiring any maintenance overhead..

Recommendation modeling techniques eliminate this complexity by using transaction or behavioral data as the basis to make recommendations without manual intervention. Using real-world transactions or interactions to find products that are sold or viewed together is the best way to make the right recommendations.

Once you add the product recommendation feature in Dynamics 365 by using the Cognitive Services recommendation service, and configuring connectivity to the Cognitive Services service, a capability is added to the product catalog to generate automatic recommendations. You'll set up the product catalog and synchronization to build a “machine learning based recommendation model” that makes recommendations in a ranked list at various places in Dynamics 365, such as at the account, opportunity, or order level. These additional cross-sell product suggestions will help improve the total value of the deal.

The Microsoft Dynamics 365 product recommendations feature supports existing line item entities (OpportunityProduct, QuoteDetail, SalesOrderDetail, and InvoiceDetail) and custom line item entities, as well as standard and custom product relationships.

This topic walks you through the process of connecting Dynamics 365 (online) to the Cognitive Services recommendation service and how to build a product recommendation model.

Important

This feature is currently only available for instances in the United States (US) region.
A preview feature is a feature that is not complete, but is made available before it’s officially in a release so customers can get early access and provide feedback. Preview features aren’t meant for production use and may have limited or restricted functionality.
We expect changes to this feature, so you shouldn’t use it in production. Use it only in test and development environments.
Microsoft doesn't provide support for this preview feature. Microsoft Dynamics 365 Technical Support won’t be able to help you with issues or questions. Preview features aren't meant for production use and are subject to a separate supplemental terms of use.We expect changes to this feature, so you shouldn’t use it in production. Use it only in test and development environments.
We are making this preview available so that you can try it and let us know what you think. Your feedback will help us prioritize work to include the capabilities you need most. We ask that you give us your suggestions and report problems by using our publicly available feedback site: Microsoft Connect

In this topic

Requirements
Enable Dynamics 365 cross-sell product recommendations
Connect Dynamics 365 (online) to Cognitive Services
The Recommendation Model page
Test your model
Activate your model
See the recommendations in action
See product recommendations in Dynamics 365 for phones and tablets

Requirements

This feature requires December 2016 update for Microsoft Dynamics 365 (online) or later version.
A Cognitive Services recommendation service connection configured in Microsoft Dynamics 365. More information: Set the recommendations connection

Enable Dynamics 365 cross-sell product recommendations

To enable cross-sell product recommendations, do the following:

Go to Settings > Administration.
Click System Settings, and then open the Previews tab.
Under Cross-sell Product Recommendations Preview, set the Enable Dynamics 365 Cross-sell Product Recommendations Preview to Yes.
Click OK to give your consent.
Click OK to close the System Settings dialog.

Connect Dynamics 365 (online) to Cognitive Services

To use Cognitive Services recommendation service with product recommendations, a Cognitive Services recommendations connection must be configured. More information: Set the recommendations connection

The Recommendation Model page

Once you have a connection to the Cognitive Services recommendation service, you can create a model for automatic cross-sell product recommendations based on historical transaction data.

Click Settings > Product Catalog > Product Recommendations.
A recommendation model with default values appears.

Recommendation Model page

Product recommendations are built based on two factors: 1) items that are are sold together (Basket Data entity), and 2) what users typically buy with that item (Recommendation entity). Basket Data entities have product line items with historic data that can be analyzed by the Cognitive Services algorithm. You can slice the data you want analyzed by updating the query for each of these entities. Recommendation entities are entities that you want to show these automatic suggestions for.

By default, product recommendations are based on opportunity, quote, order, invoice line items and the associated account. You can also configure custom line item entities to include with product recommendations.

Let's do a walkthrough of the various items on this page.

Command bar

Prodouct Recommendation Model command bar

Item	Description
Activate	Once you've built and tested your recommendation model, click Activate to make it available for cross-sell analysis.
Build Model Version	You can build multiple recommendation model versions to adjust your recommendations as your historical data set changes over time or if you want to try different slices of data.
Test Recommendations	Test your recommendation model to see if you're getting sensible and desired cross-sell recommendations. You can do a side-by-side comparison for up to three models.

Details

Product recommendation model details

Item	Description
Model Name	This is the general name of your recommendation model. This name is used for all model versions.
Maximum Recommendations	Default: 10. Positive values only. Leave empty to show all above the recommendation rating. This is the maximum number of cross-sell recommendations you want the model to provide.
Minimum Recommendation Rating	Default: 0.25. Range: 0 to 1. Only recommendations with a rating higher than this value are listed. When product recommendations are retrieved from Cognitive Services, the rating for each recommended item is compared to this value. Only those recommendations with rating values greater than or equal to the value specified here are shown in Dynamics 365. The higher the rating value, the better is the match.
Product Catalog Cross-sell Link Rating	Default: 0.5. Range: 0 to 1. Shows the rating assigned to recommendations based on cross-sell links defined for products. In addition to recommendations from Cognitive Services, existing cross-sell links defined for products can also be surfaced as suggestions. Use this value to specify the rating of the recommendations coming from the product catalog as fixed links, and mix them with the product recommendations coming from the model. A value of 0 indicates product link based recommendations are not shown. A value of 1 indicates they are shown at the top.
Model Version	Select the model version to use for your recommendation model. Build a new model as basket or catalog data is added to Dynamics 365. Also, build new models to try out different ratings or entity filtering.

Model and synchronization information

Item	Description
Build Model Version Insights
Catalog Coverage (%)	The higher the value, the better the model is in terms of coverage of links between products. Refer to Cognitive Services recommendation API documentation to understand this metric.
Precision	Precision represents the frequency at which users select product suggestions. This helps determine the effectiveness of the model and can be used to evaluate and compare against other product recommendation models before you put the model in production. The precision is measured at build time using the top five recommendations from test data. The higher the number, the better the precision.
Synchronization Information
Catalog Last Synchronized On	The latest time when the product catalog was synchronized when a model version was built.
Catalog Last Synchronization Status	The status of last product catalog synchronization as either Success or Failure.
Basket Data Last Synchronized On	The latest time when the basket data was synchronized when a model version was built.
Basket Data Last Synchronization Status	The status of last basket data synchronization as either Success or Failure.

Basket Data entities

Basket Data entity recommendations are based on what products typically appear together in the customer's basket. For example, TVs and video cables are often bought together.

Basket data entities

Add Model Entity Mapping record. Add button

Click to add other entities configured for basket data.

Fill in the following fields to add a basket data entity.

Item	Description
Entity Name	Entity that's used for capturing sales transaction data with its associated product line item relationship. For example, Opportunity.
Primary Key Field	Automatically picked based on the entity selected.
Account Field	Lookup field that captures the Account.
Product Item Relationship	Relationship of entity to product line items. It can be an out-of-box or custom line item relationship.
Product Field	Lookup field in the product line item entity.
Data Filter	Filter for data to be used for sales transactions. For example, won opportunities created in last 12 months.

Click an entity to see and configure its mapping information. Click Edit Filter Filter button to edit the data filter.

Model entity mapping information and filter

Recommendation entities

Recommendation entities are used to define sales transactions where cross-sell product recommendations are shown to users. For example, a recommendation model based on historical data might find that customers who buy a TV often buy surge protectors and wall mount kits. So when a new opportunity is created for a TV, additional recommendations for surge protectors and mount kits are shown based on the recommendation model.

Recommendation Entities list in product recommendation model

Add Model Entity Mapping record. Add button

Click to add other entities configured to show product recommendations.

Fill in the following fields to add a recommendation entity.

Item	Description
Entity Mapping Type	Mapping type for the entity.
Entity	Entity that is used to show recommendations for its line items to suggest additional products that can be sold to an account.
Account Lookup Field	Lookup field that captures the account.
Product Line Item Relationship	Relationship from entity to product line items.
Product Lookup Field	Lookup field in the product line item entity.

Test your model

Once you've configured your recommendation model settings, you're ready to test it.

On the Recommendation Model page, click Test Recommendations.
Click Products, and then specify the products to use in your test.
Select one or more model versions to see the generated recommendations along with their rating for side-by-side comparison.

Activate your model

Once you've got the model producing the desired recommendations, activate it.

On the Recommendation Model page, click Activate.
Click Activate again on the confirmation page.

See the recommendations in action

Once the recommendation model has been activated, it will automatically start showing recommendations as part of suggestions in every entity configured to show recommendations.

See recommendations for out-of-box line items.

Open a record, such as an opportunity, and scroll down to Product Line Items.
Select a product, and then click Suggestions.

Observe the cross-sell recommendations for the selected product. Notice that there are also possibilities for accessory, upsell, and substitution items.
Click Pick for an item to include for cross-sell, and then click Add to List.
The new item appears in the record's product list.

See recommendations for custom line items.

Open a record, such as an order, and locate the area where the custom line items can be added.
On the Actions toolbar, click Suggest Products.

Observe the cross-sell recommendations for the selected product.
Click Pick for an item to include for cross-sell, and then click Add to List.
The new item appears in the record's custom line item product list.

Note

If, after clicking Suggestions, you don't see any cross-sell products listed, it could be that the price list doesn't contain related items for that product.

There are three conditions for product recommendations:

For out-of-box line items, the opportunity must have a product price list set for it.
There is at least one product in the product line item list.
The product suggestions identified by Dynamics 365 or Cognitive Services are present as price list items in the price list set on the opportunity, quote, order, or invoice record.

See product recommendations in Dynamics 365 for phones and tablets

Once you've configured product recommendations, you can see suggestions on your mobile devices.

When you first open a record, you'll see product suggestions available for this record.