Webex AI Agent Integration with CJDS

Webex AI Agents allow you to use the power of GenAI inside our Customer Experience flows. Our Autonomous agents assist customers in real time by leveraging a knowledge base with Retrieval-Augmented Generation (RAG) techniques and executing actions through Webex Connect flows. This lab focuses on creating an AI Agent that can retrieve information from and post events to CJDS while supporting customer interactions.

Lab 4.1 Create an AI Agent and a Knowledge Base

To begin this lab, navigate to Control Hub and select the AI Agents card from the left pane of the Contact Center menu. Click the Build your AI Agent button. The AI Agent Studio will open in a new browser tab, where you will configure your AI Agent and Knowledge Base.

Create a Knowledge Base for your Webex AI Agent

From the AI Agent Studio, select the notebook icon on the left navigation menu. This is where you will manage your Knowledge Bases.
Click Create Knowledge Base, enter the name as PODXX_AI_KB (replace the XX with your POD number), then click Create.
Download this generic KB: Webex_Sneakers_KB.
Go to the Files tab and select the option Add File.
Upload the Webex Sneakers KB and select the option Process Files.

Knowledge Base with Files Processed IMG

Create your AI Agent

Navigate to Dashboard from the right-hand side menu panel and click Create Agent
Select Start from Scratch and click Next
On the Create an AI agent page, select the type of agent: Autonomous
A new section called Add the essential details will appear. Enter the following information:

Agent Name: PODXX_JDS_AI (Replace the XX with the POD ID)

System ID is created automatically

AI engine: Webex AI Pro-US 1.0
In the Agent's goal section, enter the following:

Answer general questions about our online store, Webex Sneakers! Provide information about the customer orders and route to human agents when required.
Click Create on the bottom right corner.
Once the agent is created, update the Welcome Message to say"

Hey! Welcome to Webex Sneakers! How can I help you?
Switch to Knowledge tab and from Knowledge base drop-down list select the knowledge base you created.
Click Save Changes.

You need to add Instructions to orchestrate how the AI Agent will execute an action, but first the Webex Connect flows have to be built. Let's do that!

Lab 4.2 Build Connect AI Agent Fulfillment Flows.

Webex Connect has a powerful and intuitive flow builder tool, with its low-code/no-code approach it makes it easy for administrators to build fulfillment flows for our AI Agents. In order to start this lab, go back to Control Hub and select the Overview card on the left pane. In the Quick Links section you will see the Webex Connect hyperlink, that will open in another tab of your browser.

Warning

The variables used in Webex Connect contain the node ID at the beginning, for example in this variable $(n4.Customer_ANI) the n4 is the node ID. This node ID might be different in your flow, so always double check that the node id matches. This can be done by double clicking the node that is generating the variable and looking at the node id on the bottom left corner.

Create Fulfillment Flows in Webex Connect

Find the Service matching your POD ID and click it.
This will open the dashboard section of your service, click the flows tab to start building.
Let's start by building a Connect flow that allows the AI Agent to fetch the caller details:
- Select the option Create Flow and add the name JDS_Identity. Click the "Create" button.
- This opens the trigger category page, select the AI Agent integration trigger.
- Now you should see the Configure AI Agent Event window open up. In the sample JSON box, remove all of the default values and just leave an empty JSON, like this:
```
{}
```
- Click the Parse blue button and hit save.
JDS_Identity Trigger IMG
- From the Node Palette on the left, drag and drop an Evaluate node into the canvas. Connect the AI Agent node to the Evaluate node. Double click the Evaluate node and enter this code:
```
var ANItoTime = Date.now() + "";
var lookbacktime = ANItoTime - 30000;
var ANIfromTime = lookbacktime + "";
1;
```
- In the same node, enter the number 1 in the Script Output field and in the Branch Name enter the word Success. Hit Save.
JDS_Identity Evaluate IMG
- Let's add another node, look for a node called WxCC API and bring it to the canvas. Connect the Evaluate green output to this new node. Double Click the WxCC API node and enter the following information:
  
  From: $(ANIfromTime)
  
  To: $(ANItoTime)
  
  Corrid: $(corrid)
- Hit Save.
JDS_Identity WxCC API IMG
- Now we are ready to use the CJDS node from Webex Connect, in the node palette find the node called Journey and bring it into the canvas.
- Connect the success output of the WxCC API node into the Journey node. Double click it and fill out the fill out the following:
  
  Method Name: Get Identity by Aliases
  
  Node Authentication: JDS_CPaaS
  
  Project Name: Customer-Journey-Widget
  
  Aliases: $(n4.Customer_ANI)
- Click the Save button.
JDS_Identity Journey IMG
- Go into the flow Settings by clicking the gear icon on the top right corner. From here, select the Flow Outcomes tab and expand the Last Execution Status menu.
- In this menu you can configure what is going to be sent back to the AI Agent when the action is executed. For this scenario, we want to send back the customer phone number, first name and last name. In the key value section, add the following variables:
  
  Key Value
  
  firstName $(n5.firstName)
  
  lastName $(n5.lastName)
  
  phoneNumber $(n4.Customer_ANI)
JDS_Identity Outcome IMG
- Click the Save button. Also, click the Save button for the flow on the top right corner and then click on Make Live. A window will pop up, click the Make Live option.
- Exit the flow from the top left corner.
- As an explanation of the flow, the AI Agent will trigger this flow to collect the user details from CJDS. In order to do this, you configured the Evaluate node to get the current time in epoch format and then you are using those values to run a Search API with the WxCC API node. This Search API returns the Customer ANI number, then this is used in the Journey node to collect the person profile details.
JDS_Identity Final Flow IMG

Now let's build a Fulfillment flow to check the status of an order from an external DB:

Select the option Create Flow and add the name CheckOrder. Click the "Create" button.
This opens the trigger category page, select the AI Agent integration trigger.
Now you should see the Configure AI Agent Event window open up. In the sample JSON box, enter the following information:
```
{
"phone": "",
"accountNumber": ""
}
```
Click the Parse blue button and hit save.

Check Order Trigger IMG

From the Node Palette on the left, drag and drop an HTTP Request node into the canvas. Connect the AI Agent node to the HTTP Request node. Double click the HTTP Request and setup the following fields:
Method: GET
Endpoint URL: https://66d8867237b1cadd8054f426.mockapi.io/CustomerDatabase?AccountNumber=$(n2.aiAgent.accountNumber)
Connection Timeout: 10000
Request Timeout: 10000

Output Variables:

Output Variable Name Response Entity Response Path

FirstName Body $[0].First

LastName Body $[0].Last

Product Body $[0].Product

InitialArrival Body $[0].initialArrival

CurrentArrival Body $[0].currentArrival

PaymentType Body $[0].paymentType

OrderStatus Body $[0].orderStatus

Click the Save button.

Check Order HTTP Request IMG

Drag and drop the Journey node into the canvas. Connect the success output of the HTTP Request node into the Journey node. Double click it and fill out the fill out the following:

Method Name: Write to CJDS

Node Authentication: JDS_CPaaS

Project Name: Customer-Journey-Widget

Request Body Object: Complete Object

Request Body Object JSON:

{
   "id":"$(n2.aiAgent.transId)",
   "specversion":"1.0",
   "type":"Order Status",
   "source":"IVR",
   "identity":"$(n2.aiAgent.phone)",
   "identitytype":"phone",
   "datacontenttype":"application/json",
   "data":{
      "taskId":"$(corrid)",
      "firstName":"$(n3.FirstName)",
      "lastName":"$(n3.LastName)",
      "phone":"$(n2.aiAgent.phone)",
      "product":"$(n3.Product)",
      "paymentType":"$(n3.PaymentType)",
      "channelType":"IVR",
      "initialArrival":"$(n3.InitialArrival)",
      "currentArrival":"$(n3.CurrentArrival)",
      "uiData":{
         "title":"Check Order Status",
         "iconType":"",
         "subTitle":"$(n3.OrderStatus)",
         "filterTags":[
            "Order Status"
         ]
      }
   }
}

Click the Save button.

Check Order Journey IMG

Go into the flow Settings by clicking the gear icon on the top right corner. From here, select the Flow Outcomes tab and expand the Last Execution Status menu.
In the key value section, add the following variables:

Key Value

product $(n3.Product)

paymentType $(n3.PaymentType)

InitialArrival $(n3.InitialArrival)

CurrentArrival $(n3.CurrentArrival)

OrderStatus $(n3.OrderStatus)

Check Order Outcome IMG

Click the Save button. Also, click the Save button for the flow on the top right corner and then click on Make Live. A window will pop up, click the Make Live option.
Exit the flow from the top left corner.

Check Order Final Flow IMG

We are down to our last fulfillment flow, this will be a SMS Deflection flow. The AI Agent will offer customers to use our SMS digital channel:
- Select the option Create Flow and add the name SMS_Deflection. Click the "Create" button.
- This opens the trigger category page, select the AI Agent integration trigger.
- Now you should see the Configure AI Agent Event window open up. In the sample JSON box, enter the following information:
```
{
  "phone": "",
  "firstName": "",
  "lastName": ""
}
```
- Click the Parse blue button and hit save.
SMS Deflection Trigger IMG
- From the Node Palette on the left, drag and drop a SMS node into the canvas. Connect the AI Agent node to the SMS node. Double click the SMS and setup the following fields:
  
  Destination Type: msisdn
  
  Destination: $(n2.aiAgent.phone)
  
  From Number: Select the SMS number assigned to your POD
  
  Message Type: Text
  
  Message: Thank you for accepting to try our customer support service over digital channels! If this is a good moment to connect you to an agent, please reply with the phrase "GoToQueue".
- Click the Save button.
SMS Deflection SMS IMG
- Drag and drop the Journey node into the canvas. Connect the success output of the SMS node into the Journey node. Double click it and fill out the fill out the following:
  Method Name: Write to CJDS
  
  Node Authentication: JDS_CPaaS
  
  Project Name: Customer-Journey-Widget
  
  Request Body Object: Complete Object
  
  Request Body Object JSON:
```
{
  "id":"$(n2.aiAgent.transId)",
  "specversion":"1.0",
  "type":"SMSDeflection",
  "source":"IVR",
  "identity":"$(n2.aiAgent.phone)",
  "identitytype":"phone",
  "datacontenttype":"application/json",
  "data":{
     "taskId":"$(corrid)",
     "firstName":"$(n2.aiAgent.firstName)",
     "lastName":"$(n2.aiAgent.lastName)",
     "phone":"$(n2.aiAgent.phone)",
     "channelType":"IVR",
     "uiData":{
        "title":"Digital Channel Deflection",
        "iconType":"icon-sms_16",
        "subTitle":"SMS",
        "filterTags":[
           "Digital Channel Deflection"
        ]
     }
  }
}
```
- Click the Save button.
SMS Deflection Journey IMG
- Click the Save button. Also, click the Save button for the flow on the top right corner and then click on Make Live. A window will pop up, click the Make Live option.
- Exit the flow from the top left corner.
SMS Deflection Final Flow IMG

Lab 4.3 AI Agent Instructions and Actions

In this section, you will setup AI Agent actions that use the fulfillment flows created in the previous section. Also, you will add the instructions on when to use those actions.

Create Actions

Switch to the Actions tab and click the New action button. Proceed to name the action JDS_Identity, add a description and select the Action Scope option called Slot filling and fulfillment.
In the Webex Connect Flow Builder Fulfillment section, select the POD XX service and the flow JDS_Identity. Click the "Add" button.

JDS Identity Action IMG
Click the New action button again, and proceed to name the action Check_Order, add a description and select the action scope option called Slot filling and fulfillment.

Create new input entities for the data required to check the order status, these are the details for each entity:

Entity Name Type Value Description

accountNumber number - account number, 8 digits

phone phone Use default regex A valid phone number with country code. This phone number will be returned from the JDS_Identity action.

In the Webex Connect Flow Builder Fulfillment section, select the POD XX service and the flow CheckOrder. Click the "Add" button.

Check Order Action IMG
Click the New action button again, and proceed to name the action SMS_Deflection, add a description and select the action scope option called Slot filling and fulfillment.

Create new input entities for the data required to send a SMS to the user, these are the details for each entity:

Entity Name Type Value Description

firstName string - This variable was returned from the JDS_Identity action. Do not ask the user.

lastName string - This variable was returned from the JDS_Identity action. Do not ask the user.

phone phone Use default regex A valid phone number with country code. This phone number will be returned from the JDS_Identity action.

In the Webex Connect Flow Builder Fulfillment section, select the POD XX service and the flow SMS_Deflection. Click the "Add" button.

SMS Deflection Action IMG

List of Actions IMG

Go back to the Profile tab on the AI Agent configuration page. In the instructions section, enter the following:

# Instructions
  **Execute step 1 before answering any questions or triggering any other actions**
  1. Inform the user that you will fetch their details, do this by executing the action JDS_Identity. 
  2. After getting the user details from the JDS_Identity action, greet the user by their first and last name. 
  3. Answer the user questions. 
  4. If the user wants to check their order status, collect their account number and use the action Check_Order to get the details. 
  5. If there are any issues with the order (delays, lost in transit, etc), ask the user if they want to proceed with a refund or replacement. Use the {{initialArrival}} and {{currentArrival}} dates to verify how much time the order has been delayed or lost. **Follow policies from the KB to determine if you can process the refund** 
  6. **Voice Channel Only** If the user wants to talk to a human agent or the company policies require you to do the transfer, first offer the user if they would like to continue this transaction over SMS. Inform them that our digital queue is faster and this way they won't have to wait long times on queue.  
  7. If they accept, trigger the action SMS_Initiate and inform the user they will receive a SMS shortly.

Click the option Save Changes and then Publish, provide any version name. Your AI Agent is now ready for testing.

Final AI Agent IMG

Testing

In Lab 3 you configured a voice flow and selected a placeholder AI Agent in the VA V2 node, go back to your voice flow and update that node to use your new AI Agent. Once you make the changes, remember to validate and publish your IVR flow. Once this is done, here's what you can do to test:

Testing AI Agent Integration with CJDS

Place a call to your voice flow and select option 1 in the menu.
Ask a couple of questions that are included in the Knowledge Base file you uploaded, for example:
- Are there any ongoing promotions?
- How long does shipping normally take?
- Are your sneakers authentic?
After you get some answers, ask about your order status.
It should tell you the order is delayed, but that it can't help with a refund or replacement for you because of store policies.
Insist that you want a refund and the AI Agent should offer you to use the SMS channel to talk to a human agent. Say yes and you should receive a SMS (Only US numbers).
Go to your Agent Desktop and on the JDS widget do a search for your phone number, you should see the events pushed from the AI Agent for Order Status and SMS Deflection.

You have completed the JDS 2851 Lab, Congratulations!

Key	Value
firstName	$(n5.firstName)
lastName	$(n5.lastName)
phoneNumber	$(n4.Customer_ANI)

Output Variable Name	Response Entity	Response Path
FirstName	Body	$[0].First
LastName	Body	$[0].Last
Product	Body	$[0].Product
InitialArrival	Body	$[0].initialArrival
CurrentArrival	Body	$[0].currentArrival
PaymentType	Body	$[0].paymentType
OrderStatus	Body	$[0].orderStatus

Key	Value
product	$(n3.Product)
paymentType	$(n3.PaymentType)
InitialArrival	$(n3.InitialArrival)
CurrentArrival	$(n3.CurrentArrival)
OrderStatus	$(n3.OrderStatus)

Entity Name	Type	Value	Description
accountNumber	number	-	account number, 8 digits
phone	phone	Use default regex	A valid phone number with country code. This phone number will be returned from the JDS_Identity action.

Entity Name	Type	Value	Description
firstName	string	-	This variable was returned from the JDS_Identity action. Do not ask the user.
lastName	string	-	This variable was returned from the JDS_Identity action. Do not ask the user.
phone	phone	Use default regex	A valid phone number with country code. This phone number will be returned from the JDS_Identity action.