Conversation Outline | Docs

The outline is the set of guidelines that tells your agent how to conduct each conversation. You never start from scratch — when you describe what you need, the design agent generates a complete outline with all sections filled in. You review it, refine anything that doesn't fit, and launch.

Think of the outline as briefing a very smart person before sending them into a conversation. You tell them the purpose, who they're talking to, how to behave, what data to collect, and what to do when it's over. They handle the rest.

The outline is organized into three tabs: Overview, Agent, and Data Capture.

Overview

This tab defines the purpose of the conversation and who it's for.

Research Question

The high-level purpose and objectives of the conversation. Why does this conversation exist? What is it trying to accomplish?

This is the agent's north star — everything it does serves this purpose.

Goals

The specific directions you want the agent to explore. Unlike a form or survey where you have to translate goals into questions and control logic, with Perspective you just point the agent in the right direction. It will get the conversation there naturally, ask follow-up questions, and probe as needed.

Goals are ordered by priority. If the conversation runs short, the agent focuses on the most important ones first.

Examples:

"Understand why they almost churned before renewing"
"Identify which competitors they evaluated and why"
"Learn what would make them recommend us to a colleague"

Participants

A description of who the agent is going to talk to. The more the agent knows about who it's speaking with, the better job it does — adjusting its language, depth, examples, and approach to match the audience.

Example: "Enterprise SaaS customers with 50+ seats who renewed in the last 90 days."

Agent

This tab controls how the agent behaves during the conversation — its behavior, flow, and style.

Interview Guidelines

Instructions that shape how the agent conducts the conversation. These cover three things:

Behavior — What the agent should and shouldn't do. "Don't ask about pricing unless the participant brings it up." "If they mention a competitor, follow up on that."
Flow — How the conversation should progress. "Start with their overall experience before going into specifics." "Wrap up after 5-7 minutes."
Style — The tone and energy of the conversation. "Be casual and conversational." "Match the formality of a colleague chat, not a formal interview."

Navigation Flow

Dynamic next steps after the conversation ends. Instead of one-size-fits-all, different participants can get different follow-ups based on what the agent learned during the conversation.

Examples:

A hot lead gets routed directly to your sales calendar
Someone just looking for information gets redirected to a resource page
A churning customer gets a personalized retention offer
Everyone else sees a thank-you message

By default, conversations end with a simple thank you. You can also enable a reflection — a summary of the conversation generated for the participant, so they see that their input was heard and understood.

Knowledge

Background information the agent can reference during the conversation — product details, industry context, internal data, competitive positioning. Think of it as a briefing document.

Use knowledge when the agent needs context that participants won't provide. For example, if you're collecting feedback on a new feature, add the feature description so the agent can ask informed follow-up questions and understand what participants are referring to.

Resources

Materials attached to the conversation. There are two types:

Public — Shared with participants. A link to a site you want them to review, a UX mockup to react to, a document to read before the conversation.
Private — Internal context for the agent only. A competitive analysis, a pricing sheet, internal notes — things the agent should know but not share.

Data Capture

This tab controls what structured data the conversation collects. Every conversation captures a full transcript and recording by default. Data capture lets you go further.

Output Fields

Structured, typed fields that the agent captures as part of the natural conversation. This is the key difference from traditional forms — you get consistent, structured input without burdening the participant with form-filling.

Each field has a type (text, number, boolean, file, array), can be required or optional, and can have predefined options. Required fields will always get captured. Optional fields get picked up when they come up naturally.

For example, instead of a dropdown asking "Company size: 1-10, 11-50, 50-200, 200+", the agent picks up company size from the natural flow of conversation and fills in the structured field automatically.

This means you can feed structured data into your CRM, trigger automations based on field values, and generate reports — all from a conversation that felt natural to the participant.

Mandatory Questions

For when you want a specific question asked exactly as written. No inference, no natural weaving — just ask this question, verbatim. Most of the time, output fields are the better choice.

Outline Versions

Every time you refine the outline, Perspective creates a new version with a version number and a change message. This gives you a full history of how the conversation design evolved — useful for collaboration and for understanding what changed between iterations.

Refining the Outline

You refine the outline by talking to the design agent in natural language. You never need to edit fields manually.

Examples:

"Add a goal about understanding their evaluation process"
"The agent should be more casual and use shorter questions"
"Add our product FAQ as private knowledge"
"If the participant is a decision-maker, redirect them to our sales calendar"
"Add a required field for company size"
"Make NPS score a mandatory question"

Each change generates a new outline version. Test the conversation after each change to see how it affects the experience.