ZVA Use Specific Knowledge for an End User

[Gines]

Hello, I'm working with Zoom Virtual Agent and wanted to clarify two things:

1. User-Specific Knowledge:
   Is it possible to assign or tie specific knowledge to an individual end user, so that when they ask questions during a session, the bot references only (or prioritizes) that knowledge?
   Our intended workflow is:

• We receive the event that a chat has started.

• We look up relevant data for that user in our database.

• We want to pass that data to the bot as knowledge, so it can use it in responses during the same session.

1. Dynamic Knowledge via API (Mid-Session):
   If we add new knowledge via the Knowledge API during an active chat session, is the virtual agent able to immediately access and use that information in the same ongoing conversation?

[VanessaVaZoomie]

 -- Hello there!

Yes, it is possible, but the lexicon is a little different. I'm thankful you spent time describing the use case so that I could craft the correct answer! I'm going to assume that you are referencing a chat agent.

Step 1: You would want to set up a "greeting action" and create a tool (using the forward slash action and selecting "new tool") that has a GET request and stores the fields about the customer into custom variables with a "global" variable type. Use only one tool to store all customer fields from that database into unique variables.
>> You have to set up the custom-global variables before the tool. You can find variables in the preferences section of Virtual Agent. See image.

Step 2: Once the tool is set-up to GET the fields and store them, you will write the greeting prompt. I would recommend something like this:
Action 1: "Before greeting the customer, use the tool (do a forward slash and pick the tool from step 1) to gather important information about the customer from the database." 
Action 2: Once you have the data about the customer, greet them and explain your role and ask how you can help. 

Step 3: In the guidance, you will write a prompt that explains how and when to use these variables. Duplicate this prompt in the top and bottom of the guidance, to show importance. 
Example: 
**Information Available** 
1. You have information about the customer stored in your variables for you to use as context, personalization and decision criteria. 
2. The information you have is:
- The customer’s first name is stored in (forward slash) [select the variable that you stored the first name in]
- The customer’s service package is stored in (forward slash) [pick the variable]
- the customer’s email address is stored in (forward slash) [pick the variable]
- the customer’s region is stored in (forward slash) [pick the variable].


Step 4: In the guardrails, you will want to reference that it must locate the value in the variable and use instructions to provide the correct answer for that field value as it pertains to the customer.

Step 5: Add logic to (a)the knowledge base and (b) skills.

Step 5a: Load in your knowledge base and then click the three dots on that row in the UX and select “add segment.” See image. You will now select the variable and set the condition for when the bot should use this knowledge base. In our example above, you would select “variable = global.custom.variable.region" for the region and then the condition would be “equal to: North America” and then select the category or tags in the knowledge base that should only be used when that field value condition is met. **you MUST select a category or tag if the knowledge base has multiple answers for different segments. You can leave it blank if you have each segment in a unique knowledge base record.

Step 5b: In your skills, you must write an instructions prompt that explains what variable to look-up and what values to expect and what to do if the field of the value meets a condition.

For example: If you want the skill to be to use specific tools or give a specific answer depending if the field for “region” in “North America”, then you would list that in the instruction prompt for the skill. Use one skill and instruct on what variable to look-up and what value condition triggers the next action. Ensure your formatting is clear and distinct.

Step 6: Try this and let us know if it worked or if you have additional questions. Without knowing precisely the skill and dataset, the instructions above are generalized. If you have a specific use case that is not working with these instructions, respond with the details and we will craft a solution for you. 🙂

Thanks!

For images referenced above, click this link to go to a public Zoom Doc with the images referenced and descriptions.