AI Agents

AI Agents are fully autonomous and are able to service complex tasks and actions like troubleshooting, orders, managing bookings, refunds etc

Agents versus Assistants

When would I use an Assistant?

An assistant is there to capture information, answer quick questions and is an aid to your existing human agent team.

When would I use an Agent?

An agent is there to act as an human agent and fully service customers as much as possible, so not only be able to handle queries but manage cases, such as updating bookings and creating support tickets.

Creating an Agent

Define the Agent’s role

It’s useful to start by thinking of the role that you need your AI Agent to take on in your business, for example :

• Receptionist: Who will take information and direct calls

• Booking Coordinator: Who will take and update bookings from your customers

• 1st Line Support Engineer: Who will answer simple questions and raise support cases

Define the responsibilities

Begin by identifying what actions you need your agent to able to perform in order to fit that role and service your customers, such as:

• Taking information from the customer to either route the call or provide context for your human agents

  • Example - sending a troubleshooting query to the support team

• Fetching data and informing the customer

  • Example - reporting back on the status of a case

• Storing data back into your system

  • Example - updating the date for a booking

• Answering simple queries

Create Agent Tools

We refer to these actions as ‘Agent Tools’, to create an Agent Tool do the following:

1. Navigate to Natterbox Settings

2. Click ‘Agent Tools’, under AI Configuration’

3. Click Create

4. Choose the type of Tool, see more Tool Types here

5. Define when the AI runs the Tool within the Description

   1. i.e. ‘This tool can be used to create a calendar event in a target's calendar.’

6. Define the actions the AI is to carry out within Tool as the Tool Parameters

   1. i.e. Summary : ‘Write a summary of the conversation including any agreed next steps, topics discussed, the questions the caller asked and any overall feedback the caller provided’

7. Click Publish

   1. This will create your first Agent Tool version, see more on versions here

Tools

Tool Description : This is the prompt which tells the AI when and how to decide to use this tool

Tool Parameters : These are the actions within the Tool which the AI can carry out

Tools Types

The Tool Types reflect the different kind of actions you need your Agent to achieve.

Macro Tool

The macro tool is a general tool to carry out an action of your description that doesn’t rely on fetching or storing data.

Example:

Human Escalation

If you need your Agent to escalate the call onto a human then you can create a macro tool with the following:

The Escalation tool can be run when the customer needs to speak with a person instead of an AI or is showing clear signs of frustration with their experience so far. You should confirm with the customer if they'd like you to connect them with another member of our team. If they ask to speak with a named colleague you should let them know you will connect them with our general team who can connect you with a specific individual.

Knowledge base Tool

If you need your Agent to be able to answer simple queries using a knowledge base as a source of information.

In order to create one of these Tools, first you will need to have had configured your knowledge base see more on this here.

Rest Tool

If you need your Agent to be able to fetch and save data through an API you have access to.

Salesforce Tool

If you need your Agent to be able to interface with your Salesforce account and fetch, store and update records.

We support the following operations :

• Create

• Query

• Update

• Upsert

• Apex

For each of the record management options a query builder will be available for you to build out and select which Salesforce records you want to be able to update and how:

Create an Agent

Once you have your actions defined as Agent Tools, you can now create an Agent to assign those Tools.

1. Navigate to Natterbox Settings

2. Click ‘Agent Configuration’, under AI Configuration’

3. Click Create

4. Set the Name and Description of the Agent

   1. We recommend tying it back to the role they’re going to be fulfilling, example ‘Receptionist’, ‘Capture customer’s details and forwards calls to the most appropriate department‘

5. Define the Prompt

   1. This is a where you define the behaviour and how you want Agent to conduct it’s actions

6. Assign the Tools

   1. Click Add

   2. Select the Tool

   3. Select the Version to use

   4. Define the Task Completion Conditions

      1. Upon executing the Tool if you need Agent to progress the call in the Routing Policy (i.e. Human Escalation or Route to a department) then you need to choose Success With Data or Success With No Data depending on whether or not the Tool is returning back data

      2. If you need the Agent to keep hold of the call and continue servicing then leave the Agent Completion Conditions unselected

   5. Repeat for each Tool you want your Agent to have access to

7. Click Save Changes

8. Click Publish

   1. This will create your first Agent version, see more on versions here

User Prompt Prefix & Inference Settings

These are advanced optional settings and we recommend leaving them as default.

Versions

For both Agents and Agent Tools, you can version your configurations.

This allows you to quickly make incremental changes, test them without impacting the live service and roll them out how you want.

When saving an Agent or Agent Tool, these changes will always be saved as Working Draft.

These changes can then be tested, simply select the version Working Draft from either Agent component or the Tool configuration on the Agent itself:

Once you’re happy with your recent changes come back to the configuration and hit Publish to generate a version.

Then version be selectively used on a per Tool or per Agent basis, so you can choose to gradually roll out your changes.

Alternatively if you for your Agents and Tools to always use the latest and greatest, select the Version option Latest Published Version.

Prompt Tags

When creating Agents you may find you need to use the same sets of instructions or rules across multiple Agents.

Rather than duplicating these out into different prompts that can be difficult to manage, Prompt Tags can be employed to create shared instructions/rules that are applied to any and all Agents.

1. Navigate to Prompt Tags

2. Click Create

3. Populate the Name and Description of your Prompt Tag

4. Define your Prompt

5. Click Save Changes

6. Add the Prompt Tag to your Agent by using the Macro Picker

For example you can use Prompt Tags to define your Business guidelines or compliance rules and then add them into the System Prompts of your agents.

Creating an Agent in a Routing Policy

Now you have your Agent ready and configured the next step is to then add this Agent to your desired Routing Policy.

1. Create a Natterbox AI Component

2. Select Agent App

3. Select your Agent from the dropdown

4. Select the Version to use