Conversation Flow

To design the conversation flow, navigate to the Conversation Flow tab within your new agent and begin.

---

​

Primary Configurations

The primary attributes (name, persona, voice, objective, variables, advanced prompt, etc.) of your agent are defined in the first left panel Conversation Flow tab when selecting an agent.

​

Configuring Name

You can edit the internal name for this workflow (agent). This is not the name of the agent when it is speaking. To define the name of the agent for conversation purposes, specify this information in the Persona.

​

Configuring Persona

You can define the personality and persona of your agent here. An example persona is as follows:

You are Penny, a digital assistant working for Strada Insurance, an agency selling homeowners insurance in Florida. You have the persona of a seasoned customer support agent in his early 30s, combining deep technical knowledge with a strong sense of emotional intelligence.

​

Configuring Voice

You can both test and select a voice out of the list of all choices available within Strada’s platform here. This is also where you can adjust the speed of the voice agent along with adding in any custom pronunciations.

​

Configuring Objective

The objective is the overall purpose of the conversation and what the agent should focus on accomplishing. An example objective is as follows:

Your goal is to call {{first_name}}, who just submitted a request to get a homeowners insurance quote on your website, ask them qualifying questions and schedule a call to review their quotes and options.

​

Variables

In any text area when building your agent, you can define variables that will be provided when creating a call. Simply reference the name and wrap it around double braces ({{ }}). The variable name must not have any whitespaces

​

Referencing Actions

Similar to variables, actions you define for your agent can be referenced by specifying them directly in any text area. Be sure to copy the action name correctly and wrap it around a single back quote (`). For example (`Action_Name`).

​

Advanced Prompt

To provide any additional guidelines or instructions, you can use the Advanced Prompt section to do so. This can be used to:

• Provide an FAQ
• Specify additional guardrails
• Clarify specific instructions and rules for how to handle a specific scenario or objection

The advanced prompt can also be used to override the entire conversation builder. This can be useful, if you are importing a prompt from a different provider. Simply enable the checkmark Only use advanced prompt and all additional instructions will be ignored.

​

Knowledge Bases

This is where you can both select a relevant knowledge base specific to your agent as well as set the version you want to use for the agent. Right underneath, you will also see a Enable RAG toggle. If your knowledge base is rather large and covers multiple edge cases (in which case it may not always be super relevant), it helps to utilize RAG in order to not harm the latency of the agent. That said, if your knowledge is rather small and super relevant, it’s better to include it into the context of the call anyways and therefore, not activate RAG.

---

​

Conversation Nodes

To control the flow of the conversation, use the drag-and-drop conversation builder to design any flow.

​

Conversation Start

The Conversation Start node is responsible for defining what the agent says in the call opener. The Initial Message is used to provide a script for the agent to speak at the beginning. Variables can also be specified in the initial message. For example:

Hi! Am I speaking with {{first_name}}?

Additionally, the Initial Message Mode specifies turns i.e. whether the agent starts speaking first or the user/customer that is being called. General practice is to leave this as “Agent Waits for User” for outbound calls and “Agent Speaks First” for inbound calls.

​

Outcomes

Outcomes are a crucial part of defining the conversation flow. They can model any potential outcome that the user may say. When you add an outcome, you add a handle to the node, such that you can connect it with other conversation nodes. Outcomes can be vague such as “User responds” which will encompass any sort of response, or more explicit “User says yes”. The AI agent is intelligent enough such that a loose description can be adequate, though when handling multiple scenarios, it is best to be more specific.

​

Agent Turn

The Agent Turn node is to specify what the agent should say during the call. It has an input handle which indicates which previous node(s) it is coming from or connected to. It also has outcomes that provide handles to define where the conversation can go from this point forward. The agent turn node has 3 types of configurations:

​

Instructions

Instructions are a description of what should happen in this conversation turn. Instructions can be treated as a very focused prompt for that specific turn of conversation. Full Markdown is supported such that you can provide lists or bullets as part of your instructions.

​

Script

A script is used to tell the agent what to say verbatim. This is helpful, if you know exactly what needs to be stated and you want the agent to state it verbatim. For example:

Hi {{first_name}}! This is Penny, a digital assistant with Strada Insurance! … I saw that you submitted a request from home insurance! … Can I ask you a couple of questions to make sure we get you the best rate possible?

​

Notes

Notes are additional context that can provided in this turn. These can be thought of similar to Advanced Prompt but they are more focused to the specific conversation turn. You can use these to provide additional context or disclaimers. This is helpful for example when providing additional context about an action that the agent must invoke.

​

Outcomes

Additionally, Agent Turn node also has outcomes that are defined the same as outcomes in Conversation Start.

​

Conversation End

While this node is similar in configurations to the Agent Turn node, it lacks outcomes. This is because, the purpose of the Conversation End node is to wind up the conversation and end the call. These nodes usually come at the end of the flow, either when the purpose of calling has been fulfilled or there is nothing more left to be done/said.

---

​

Testing

To test the conversational flow you just built, utilize the Chat or Test Outgoing Call buttons on the top right corner of your page to conversate with your agent via chat or call (only if a phone number has been setup already). You will first be prompted to fill in variable values, and then the agent is activated based on your chosen action.