Prepare Product and Site FAQ Topics

Learning Objectives

After completing this unit, you’ll be able to:

• Explain the difference between structured and unstructured data.
  
• Describe how to upload unstructured data to the Agentforce data library.
  
• Summarize how to create a data cloud search index for FAQ data.
  
• Summarize how to create search index retrievers for Data Cloud data model objects.
  
• Explain how to create a Data Cloud ensemble retriever for product FAQs.
  

FAQs and Shopping Decisions

Answers to frequently asked questions (FAQs) play a critical role in shaping online shopping decisions. Clear and concise responses to FAQs help shoppers quickly find the information they need, such as product specifications, return policies, or shipping details. By providing accessible and accurate answers to FAQs, your business reduces shopper hesitation and uncertainty, minimizes cart abandonment rates, encourages repeat purchases, and ultimately drives customer satisfaction and loyalty.

Your Guided Shopping Agent can quickly answer common customer questions thanks to the Product and Site FAQs topic, and Salesforce Data Cloud. The Product and Site FAQs topic is automatically added when you create your agent using the Guided Shopping for B2C Storefronts template. The topic directs your agent to knowledge articles and other FAQ-related data resources.

Structured and Unstructured Data

To respond to FAQs, the Product and Site FAQs topic uses structured and unstructured data.

• Structured data: This is information that’s organized in a predefined format, such as in a database. Examples include customer profiles, purchase histories, and product catalogs. Structured data from your B2C instance flows to Data Cloud, where your Guided Shopping Agent accesses it.
  

To enable your agent to answer product FAQs within your B2C Commerce store, make sure your team regularly ingests structured product data into Data Cloud.

• Unstructured data: This type of data lacks a consistent format and includes sources such as chat transcripts, emails, and product reviews. You upload unstructured FAQ data to the Agentforce data library.
  

To provide useful answers to FAQs, Guided Shopping Agents rely on structured and unstructured data. Salesforce's Data Cloud facilitates this integration, enabling agents to deliver personalized, efficient, and insightful shopping experiences. By analyzing patterns in both structured and unstructured data, your Guided Shopping Agent can anticipate customer needs and suggest products or services before they are explicitly requested.

Let’s take a look at how you set up your storefront data to work with the Product and Site FAQs topic.

Upload Unstructured Data for Site FAQs

The Guided Shopping Agent responds to product inquiries based on unstructured content, such as specification sheets, usage guides, return policies, and store information. To make this content accessible to the agent, upload the relevant unstructured data to the Agentforce data library (ADL). Be sure to retrieve the corresponding retriever ID, you’ll need it when you create a search index retriever.

The unstructured data must have the product name so the agent can look up the correct document for a given product.

1. From Setup, in the Quick Find box, search for and select Agentforce Data Library.
   
2. Click New Library.
   
3. Enter a name for your library. For example, Site FAQs. Keep the Retriever ID or API name handy for future set up.
   
4. Select the data type, which is typically Files.
   
5. Upload the files that your site-wide FAQ relies on.
   
6. Save your work.
   
7. Wait for Data Cloud to index your data. Indexing makes your data searchable and usable by your agent.
   

For consistent performance, do not enable Enhance Data in Advanced Settings.

Example:

Create a Data Cloud Search Index for Product FAQs

To make your Guided Shopping agent a more effective tool for answering FAQs, create a Data Cloud search index for the Goods Product Data Model Object (DMO), and the Product DMO. A Data Cloud search index enhances search capabilities with semantic understanding, search precision, AI responses, and efficient data management.

Goods Product DMO

The Goods Product DMO represents a specific product that a company offers, such as a handbag or a set of towels. The DMO is designed to help businesses organize and manage their product-related data effectively.

Why Is the Goods Product DMO Important to the Guided Shopping Agent?

The Guided Shopping Agent relies on accurate and well-structured product data to provide personalized and efficient shopping experiences for customers. Creating a search index based on the Goods Product DMO makes perfect sense.

The Goods Product DMO is used across many Salesforce channels including sales, service, and commerce. This means that the Guided Shopping Agent can provide consistent product information, whether the customer is shopping online, in-store, or through a mobile app.

Master Product DMO

The Master Product DMO represents various aspects of products or services that a company sells or tracks. The Product DMO is essential for managing and categorizing product information in a structured way, enabling seamless integration and data mapping across Salesforce systems.

Why Is the Goods Product DMO Important to the Guided Shopping Agent?

After importing asset and customer data into Data Cloud, the Product DMO can be used to map data for segmentation, activation, analytics, or other operations. The Product DMO facilitates the organization and harmonization of product data across Salesforce systems, ensuring consistency and accessibility.

Data Chunking

As part of the indexing process, you configure a data chunking strategy for indexing the DMOs’ fields. Chunking for a Data Cloud search index involves breaking down structured and unstructured data into smaller, coherent units that preserve the original text's meaning and context. Chunking optimizes data processing, indexing, and retrieval. Don’t worry, we won’t be sending you into the data for this step. You just make a few process selections and Data Cloud does the rest.

Search index chunks ‌contain text fields that help with semantic search. A field with one or a few words doesn’t add much value to the search index chunks, but the words can be prepended to each chunk to improve search results.

Create a Search Index for the Goods Product DMO

1. From the App Launcher, select Data Cloud.
   
2. On the Search Index tab, click New.
   
3. Click Advanced Setup, and then click Next.
   

4. Keep the data space as the default.
   
5. Select Vector Search.
   
6. Search for and select the Goods Product data model object.
   
7. Select Chunking.
   
8. Click Manage Fields and add these fields.
   

• Product Description
  
• Product Long Description
  
• Product Name
  

9. Click Save.
   
10. Select a chunking strategy for each field based on the content that you’re working with.
    For example, think about the fields that provide the most important context and meaning, how to split those fields up, and how you expect the agent to use the results.
    
11. Select Prepend fields to each chunk, and prepend ProductName and Product SKU to each chunk for Product Description and Product Long Description.
    For example, if a chunk contains a detailed description of a product, pre-pending the ProductName and Product SKU ensures that the chunk is associated with the correct product. This is particularly useful when multiple products have similar descriptions.
    
12. For Vectorization, select an embedding model, and click Next.
    Vector embeddings measure the semantic closeness of different pieces of text to create accurate and relevant results in your prompts and searches.
    
13. (Optional) Select related fields for search filtering.
    By default, Data Cloud preselects text fields from your unstructured DMOs to index.
    To enhance a search index configuration, add fields from an object or related objects. Including additional fields gives users more ways to filter their results.
    
14. Click Next.
    
15. Review your search index, and save your work.
    

Create a Search Index for the Master Product DMO

1. From the App Launcher, select Data Cloud.
   
2. On the Search Index tab, click New.
   
3. Click Advanced Setup and then click Next.
   
4. Keep the data space as the default.
   
5. Select Vector Search.
   
6. Search for and select the Master Product data model object.
   
7. Select Chunking.
   
8. Click Manage Fields and add these fields.
   

• Product Description
  
• Product Long Description
  
• Product Name
  

9. Click Save.
   
10. Select a chunking strategy for each field based on the content you're working with.
    For example, think about the fields that provide the most important context and meaning, how to split those fields up, and how you expect the agent to use the results.
    
11. Select Prepend fields to each chunk. Prepend ProductName and Product SKU to each chunk for Product Description and Product Long Description.
    
12. For Vectorization, select an embedding model, and click Next.
    Vector embeddings measure the semantic closeness of different pieces of text to create accurate and relevant results in your prompts and searches.
    
13. (Optional) Select related fields for search filtering.
    By default, Data Cloud preselects text fields from your unstructured data model objects to index because those are the primary data sources for embeddings.
    Add fields from an object or related objects to enhance a search index configuration. Including additional fields gives users more ways to filter their results.
    
14. Click Next.
    
15. Review your search index, and save your work.
    

Create Search Index Retrievers for Each DMO

Retrievers are tools used within Salesforce's Data Cloud to enhance the relevance and accuracy of responses generated by large language models (LLMs). They work by retrieving data from Data Cloud. This retrieved data is used to augment agent responses, ensuring that responses are grounded in accurate and pertinent information. The retriever improves results and gives you more control over your data and how it’s exposed to the agent.

Imagine a scenario where a Guided Shopping Agent needs to provide answers based on unstructured data such as a set of knowledge articles. A search index retriever can:

• Pull the most recent and relevant articles from the knowledge base.
  
• Filter results to include only articles related to a specific product or issue.
  
• Return the filtered data to the agent, enabling it to provide precise and helpful responses.
  

Shoppers do not see these backend filters directly. Instead, they interact with a user-friendly interface where they can select from predefined filter options–for example, size, color, or category. The Guided Shopping Agent uses the refined search results, which have been filtered based on the backend configuration, to present relevant products to the shopper.

Here are the steps to create a retriever and select which fields are returned to the agent.

1. From the App Launcher, open Einstein Studio.
   
2. Select Retrievers | New Retriever.
   
3. Select Individual Retriever, and then click Next.
   

4. Select the Data Cloud retriever type.
   
5. Select the default data space.
   
6. For the Goods Product DMO:
   

• From the Select a data model object dropdown, select Goods Product, and click Next.
  
• Select how you want to define the filters for the data that you want to retrieve, and click Next.
  
• Select these fields to return, and click Next.
  

• Do not select Review your selections, but save your work.
  
• Enter a name and optional description.
  

For consistent performance, don’t select Advanced Settings or Citation Settings.

7. For the Master Product DMO:

• Select Master Product, and click Next.
  
• Select how you want to define the filters for the data you want to retrieve, and click Next.
  
• Select these fields to return, and click Next.
  

• Do not select Advanced Settings or Citation Settings.
  
• Review your selections, and then save your work.
  
• Enter a Name and Description (Optional).
  

For consistent performance, don’t select Advanced Settings or Citation Settings.

When creating a retriever, consider how you structure data spaces that the retriever can access. If you have multiple sites for your business, assign each agent to the retrievers for that site’s data space. If your business has multiple channels with a site for each channel, you might want to assign the common data space’s retrievers to each store agent, to keep context across their channels. A single site can have a single data space that the retriever accesses.

Create a Data Cloud Ensemble Retriever for Product FAQs

Ensemble retrievers add data that a large language model can use to generate more relevant responses.

Here’s how to create an ensemble retriever that includes the data library retriever for unstructured product data, the Goods Product retriever, and the Master Product retriever.

Use the retrievers that you created or default retrievers to create an ensemble retriever.

1. From the App Launcher, open Einstein Studio.
   
2. Select Retrievers | New Retriever, then click Next.
   
3. Select Ensemble Retriever.
   
4. Select the GoodsProduct, MasterProduct, and your data library retrievers.
   
5. Click Next.
   
6. Select these fields for the retriever to return:
   
   • Brand
     
   • Product Description
     
   • Product ID
     
   • Product Long Description
     
   • Product Name
     
   • Product SKU
     
7. Click Next, then save your work.
   
8. Enter a name and an optional description for the retriever.
   
9. Activate the retriever.
   

Sum It Up

In this unit, you learned how to set up your agent to use structured and unstructured data to help your Guided Shopping Agent answer FAQs. In the next unit, learn how to connect your Guided Shopping Agent to a messaging channel and deploy it to your site.

Resources

• Salesforce Help: Manage Search Indexes
• Salesforce Help: B2C Commerce Connector
• Salesforce Help: Chunk Data
• Salesforce Help: Retrieve Data