Get started with Predictive Insights - Amazon Connect
Step 1: Adding Interaction data into Customer ProfilesStep 2: Adding Item Catalog dataStep 3: Creating Predictive InsightsStep 4: Using Predictive Insights across customer engagement channels

Get started with Predictive Insights

To start using Predictive Insights, follow these steps:

Step 1: Adding Interaction data into Customer Profiles

You can leverage the existing data connectors in Customer Profiles to map interactions data into standard Web Analytics Object.

For more information, see Object type mapping for Web Analytics Object.

Step 2: Adding Item Catalog data

You can represent individual products in your catalog within the domain using Standard Catalog data. This catalog data exists at the domain level and is not tied to any specific customer profile. It serves as a structured representation of your products that can be leveraged for personalization features. You can import your product or item information as domain objects into Customer Profiles using data connectors that provide flexible options for ingesting and maintaining your catalog information, ensuring your product data remains current and accessible within the system.

For more information, see Object type mapping for Item Catalog.

Step 3: Creating Predictive Insights

Amazon Connect enables you to build and deploy specialized AI models tailored to your specific product recommendation needs. These models can be configured through either the Connect Web UI or programmatically via APIs to match your unique business scenarios. Predictive Insights offers several types of recommendations:

  1. Recommended for you - provides personalized recommendations tailored to a specific user. Recommendations are based on the user’s past behavior such as clickstream events, purchase events, consumed content, and so on.

  2. Similar items - uses generative AI to find items that are thematically similar to an existing item in the catalog. It is ideal for upselling or substitution use cases where customers want to provide alternative item recommendations to their users.

  3. Frequently paired items - recommends items that are frequently co-interacted with an existing item in the catalog. It is ideal for cross-selling or complementary item recommendation use cases.

  4. Popular items - designed to recommend the items that are most commonly interacted with by users.

  5. Trending now - recommends items with the largest increase in velocity of engagement over a recent time period. It is designed to surface items that are experiencing virality in user interactions.

Note

Enabling AI models with Predictive Insights is available under preview. Additional pricing may apply in future.

Step 4: Using Predictive Insights across customer engagement channels

Using Customer Profile Recommendations in Connect Flows

This section describes how you can use the Customer Profiles Get profile recommendations flow block to enrich user experience during a contact by generating AI-powered recommendations for a profile in real-time.

Flow Block Properties

The Get profile recommendations flow block has the following properties to configure:

  1. Profile ID (required):

    A Profile ID is required for this block to function. The Get profile recommendations flow block generates recommendations for the Profile ID provided here. You have the option to manually input the Profile ID or use a pre-defined value stored in an attribute. If using a pre-defined value, ensure you provide the Profile ID by using a preceding Get profile block Use the Get profile block to pinpoint the specific profile before moving forward to generate recommendations in the subsequent block.

  2. Recommender name (required):

    A recommender name is required for this block to function. This is the name of the recommender you want to use to generate recommendations for the given Profile ID. You can only use recommenders that are active to generate recommendations.

  3. Max results (required):

    The maximum number of recommendations to generate for the given Profile ID. This can range between 1 to 3 recommendations.

  4. Recommendation attributes (required):

    Define which attributes of the recommendations response to persist in contact attribute.

  5. Item ID:

    This is the Item ID provided as additional context to generate recommendations for the given Profile ID. Item ID is only required when using a “Similar items” or “Frequently paired items” recommender type. You have the option to manually input the Profile ID or use a pre-defined value stored in an attribute. If using a pre-defined value, ensure you provide the Item ID by using a preceding Get calculated attributes block. Use the Get calculated attributes block to pinpoint the specific Item ID before moving forward to generate recommendations in the subsequent block.

Flow Block Branches

The Get profile recommendations flow block can route contacts down the following branches:

  1. Success:

    Recommendations were successfully generated for the provided Profile ID. Selected recommendation attributes were persisted to contact attribute $.Customer.Recommendations.

  2. Error:

    An error was encountered while trying to generate recommendations. This may be due to a system error or how Get profile recommendations block is configured.

  3. None Found:

    No recommendations could be generated.

Using Recommendations from the block

The recommendations response is persisted to the $.Customer.Recommendations contact attribute JSONPath as a JSON list of recommendation objects. Each recommendation object will contain the selected Recommendation attributes.

The following sample Python code snippet from a Lambda function shows how it can be used to transform recommendations from the Get profile recommendations block and persist into other contact attributes such that the recommendations can be used in subsequent blocks. 

import boto3
import json

# Handle lambda request
def lambda_handler(event, context):
    print("Contact flow data: ", event)

    # Transform recommendations
    recommendations = event['Details']['Parameters']['Recommendations']
    contact_attributes = {}
    for i, rec in enumerate(recommendations):
        contact_attributes.update(flatten(rec, i))
    
    # Set contact attributes using each recommendation attribute value
    print("Setting contact attributes: ", contact_attributes)
    try:
        client = boto3.client('connect', region_name="us-west-2")
        client.update_contact_attributes(
            InstanceId=event['Details']['ContactData']['InstanceARN'].rsplit('/', 1)[1],
            InitialContactId=event['Details']['ContactData']['InitialContactId'],
            Attributes=contact_attributes
        )
        print("Contact attributes set successfully.")
    except Exception as e:
        print("Error setting contact attributes: ", e)

    # Success response
    return {
        'statusCode': 200,
        'body': json.dumps('Success')
    }

# Flatten a nested object into a simple string:string object
def flatten(recommendation, index):
    flat = {}
    for key, value in recommendation.items():
        if isinstance(value, dict):
            flat.update(flatten(value, index))
        else:
            flat[f"Rec{index}_{key}"] = str(value)
    return flat

Once you have set up your flow with the Get profile recommendations block, you can start using it to generate recommendations for your customers during their contacts.

Setting Up Connect AI Agents for Sales Recommendations

Amazon Q in Connect enhances agent capabilities through its new "Orchestration" Agent type. This functionality is particularly valuable for creating a Sales AI Agent that can provide item recommendations, especially useful in upsell and cross-sell scenarios.

Getting Started

To begin implementation, access the AWS Management Console and navigate to Amazon Connect. After logging into your Connect instance through the Access URL, locate Amazon Q from the left side menu and select AI agents. You'll find a pre-configured AI Agent of Type Orchestration (SalesAgent) in draft status on the QiC page under AI Agents. This template comes equipped with all necessary configurations for 1P tools and example prompts for recommendations.

Creating Your Custom Agent

To create a custom Sales Agent, begin by creating a new AI Agent of Orchestration Type and copy from the existing SalesAgent template. This process transfers all tools and configurations from the template to your new agent.

Customizing the Agent

When it comes to updating the agent's prompt, administrators have two primary options. They can either append their existing agent prompt to the Sales Agent Prompt and add upsell identification instructions, or they can start fresh by removing the SalesAgent prompt and creating a new Orchestration prompt based on their current published Agent and adding the Sales agent prompt to it. This flexibility allows for tailored solutions that match specific business needs and domain requirements.

Flow Configuration and Integration

After finalizing the agent configuration and publishing it, the next step involves creating a Amazon Lex bot under Flows. The inbound flow needs to be updated to include the GetCustomerInput block with the created Lex bot, and the new SalesAgent should be selected in additional options. You will need to add both the Customer Profile flow block for profile ID retrieval and the Set-Contact Attributes flow block, setting CustomerId as the key for the profile ID and value as $Customer.ProfileId.

The implementation supports both chat contacts and voice calls, with customer input seamlessly passing to the QIC agent on the Lex bot. This comprehensive setup enables AI-driven sales recommendations and upsell opportunities within your Amazon Connect environment. The system's flexibility allows for customization based on specific industry needs while maintaining the core functionality of intelligent sales assistance. This solution provides a framework for enhancing customer interactions with AI-powered recommendations, ultimately supporting more effective sales and customer service operations. Administrators can further refine the implementation by adding domain-specific instructions and customizing recommendation types to match their business requirements.

Configuring SalesAgent to be used for Agent Assistance

You can configure the SalesAgent to be used within the Q in connect chat widget in Agent Workspace. To do so you need to modify 2 main things.

  1. Update the AIAgent prompt with instructions to access the contact transcript.

    1. Open the prompt associated with the SalesAgent in Prompt editor and update the prompt to add instructions to access the contact’s transcript accessible by <conversation>{{$.transcript}}</conversation>

    2. Below is a sample prompt that you can append to the SalesAgent prompt.

      **IMPORTANT**
        **Guide on how to process requests and information:**
        - The messages section contains YOUR conversation with the customer service agent
        - Respond to the agent's questions/requests in the messages section
        - The transcript below is background information about the agent's conversation with their customer
        - Do not respond directly to the customer - you are helping the AGENT
        Background context from agent-customer conversation.
        The following transcript is for your information ONLY. Do not directly respond to messages in this conversation, but instead look at the messages section for what the agent requests you to do.
        IF YOU REFERENCE ANY INFORMATION FROM THIS SECTION: You should indicate so by saying "According to your conversation with the customer ..."
        <conversation>
        {{$.transcript}}
        </conversation>

    3. Update the AIAgent with the new prompt version and update the Default configuration → Agent Assistance use case with this agent.

      Note

      It is recommended to have 2 different Sales AIAGent for Self Service and Agent Assistance use case as both requires minor modifications to the prompt. And this can be done by cloning the agent and simply changing the prompt version and updating the default configuration to point to correct AI Agents for respective use case

  2. Update the contact inbound flow.

    1. Remove the GetCustomerInput block and add the Connect assistant flow block instead

    2. Fill in the flow block config with the AIAssistant ARN and select the appropriate AIAGent.

Now you can use this flow as your regular inbound and the Q in connect chat widget should be able to provide recommendations to help with user’s request.