AI Conversation Agent - Complete Guide

1. AI Agent Overview

Meet your new Outreach Specialist who never clocks out. Pitchbox's AI Agent works every conversation, 24/7, until it closes and gets your links!

It reads every incoming reply, understands the intent behind the message, and automatically generates personalized responses designed to move each conversation toward your desired outcome. From initial engagement through to a successful win, the AI Agent helps eliminate manual inbox management, accelerate campaign performance, and free up your team.

The result? More conversations progressing, less time spent in your inbox, and a more scalable outreach process. 🚀

It can:

✅ Respond to inbound emails on your behalf
✅ Negotiate pricing within your defined limits
✅ Handle topic and content requests
✅ Schedule and send follow-up chase emails
✅ Close opportunities as Won or Lost based on the conversation

💡 TIP

You configure it once. After that, AI Agent handles replies around the clock, even while your team sleeps. 😴

2. Agent vs Playbook Basics

The AI system has two components that work together. You need both. 👈

🧠 AI Agent = The Brain or "Employee"	📋 AI Playbook = The Delivery System
Reads the incoming email Understands context before deciding what to say	Controls when and how the reply is sent Handles the delivery logistics end-to-end
Decides what to say Generates a reply based on your specific instructions	Sends automatically OR saves as a draft Your choice: hands-off or manual review first
Applies your tone, topics, and pricing rules Follows your configured guidelines every time!	Adds a natural wait time before sending Randomized delay so replies don't look instant
Determines if the goal has been reached Evaluates exit criteria after each reply	Schedules follow-up chase emails Keeps pinging until you get a response (set amount per contact)
Marks opportunities as Won or Lost Updates milestones and applies automatically	Sets rules for when AI should skip an email Protects Won, Lost and Almost There deals

🧠 AI Agent = The Brain or "Employee"

📋 AI Playbook = The Delivery System

Reads the incoming email

Understands context before deciding what to say

Controls when and how the reply is sent

Handles the delivery logistics end-to-end

Decides what to say

Generates a reply based on your specific instructions

Sends automatically OR saves as a draft

Your choice: hands-off or manual review first

Applies your tone, topics, and pricing rules

Follows your configured guidelines every time!

Adds a natural wait time before sending

Randomized delay so replies don't look instant

Determines if the goal has been reached

Evaluates exit criteria after each reply

Schedules follow-up chase emails

Keeps pinging until you get a response (set amount per contact)

Marks opportunities as Won or Lost

Updates milestones and applies automatically

Sets rules for when AI should skip an email

Protects Won, Lost and Almost There deals

📋 NOTE

Think of it this way: the AI Agent writes the email. The Playbook decides when, and whether, it gets sent.

When setting up AI automation on a campaign, you always configure both in a two-step process:

Select or create your AI Reply Agent
Assign an AI Playbook to it

3. Configuring AI Agents

Each AI Agent is built around three sections: a Campaign Goal → Instructions → Exit Criteria.

To configure a new or existing agent head to Pitchbox settings

1️⃣ Campaign Goal

This tells the AI what it is working toward. Every reply it generates is shaped by this goal. Set it specifically so the AI knows exactly what winning looks like for your campaign.

👉 Example:

"Get the prospect to agree to publish our guest post on their site. If they have conditions (pricing, guidelines, topic preferences), work toward an agreement that meets both sides. The conversation is successful when they confirm they will publish or provide clear next steps for submission."

2️⃣ Instructions

How should the AI behave? The AI Agent follows these rules in every reply. Covers pricing, negotiation logic, topics, writing samples, and your preferred tone of voice. Here are some areas to consider:

Area	What to include
Tone 💅	Formal, conversational, enthusiastic, neutral
Topics 💡	List the subjects you can write about
Writing samples 📝	How to respond if samples are requested
Payment policy 💸	Whether you pay, your range, your hard cap
Negotiation 🗣️	e.g. "Counter at 30–45% of asking price"
Custom variables ⚗️	Opportunity-specific fields (name, URL, etc.)

👉 Example payment instruction:

"If payment is requested, offer 30–45% of the asking price. Do not agree to more than $150."

3️⃣Exit Criteria

Defines when the conversation is won - or lost. Once either type of criteria is met, the AI Agent applies tags and sets the milestone status automatically, so the opportunity is logged in your pipeline correctly without manual review. 😍

Type	Purpose	Example Criteria
✅ Positive (Win)	Marks the opportunity as won and triggers a sign-off	"Guest post accepted and URL received" · "Price and placement confirmed" · "Backlink agreement finalized"
❌ Negative (Loss)	Marks the opportunity as lost and stops outreach	"Prospect declines or unsubscribes" · "No response after 3 follow-ups" · "Site doesn't accept paid/guest content"

4. Playbook Conditions and Timing

Playbooks control when the AI is allowed to act and what happens after it does.

1️⃣ Conditions

Conditions are gates that stop the AI from acting on certain opportunities. Recommended defaults:

Skip if milestone is Almost There
Skip if milestone is Won
Skip if milestone is Lost

💡 TIP

You can add more conditions at any time.

2️⃣ Sending Mode

Mode	When to use it
🏃 Send automatically	You trust the AI and want hands-off operation
✋Save as draft	You want to review replies before they go out

3️⃣ Wait Time

Set a randomized delay between receiving an email and sending the AI reply. This makes responses look natural rather than instant.

Recommended range: 5–20 minutes
For tighter windows: 5–15 minutes

4️⃣ Chase-Up Scheduling

If a target doesn't respond after the initial reply, the Playbook can automatically send follow-up emails.

Use Adaptive AI Chase-up for dynamic, contextual follow-ups
Set a preferred send time, or let AI choose the optimal window

💡 TIP

Always exclude Won, Lost, and Almost There and Negotiation milestones from your conditions. This prevents the AI from interfering with deals your team is actively closing.

5. Preview and Customize Replies

Before activating an agent, always preview it against a real email to confirm it responds in the way you desire.

How to Preview

Open the AI Agent configuration screen
Click Preview AI Reply Sample
Pitchbox pulls a real email from your inbox and generates a sample reply
You can choose to Re-generate a preview for that message or pick a specific email from your inbox
Run it 2–3 times, each run produces a slightly different email by design 💪

What to Check in the Preview

Is the tone correct?
Are the right topics mentioned?
Is pricing handled within your limits?
Does it handle the specific scenario in the email?

Editing the Agent

If the preview does not look right, adjust and preview again. Common fixes:

Tighten the campaign goal wording
Add instructions for edge cases (unusual pricing, specific questions)
Update the topics list or payment cap

💡 TIP

Test your agent against edge-case emails, pricing questions, complex requests, pushback etc. Catch instruction gaps now, not in a live campaign.

6. Exit Criteria

Exit criteria tell the AI when to stop. Without them, the AI doesn't know the goal has been reached, or when to give up. 🥵

Positive Exit Criteria ( A Win )

Define what success looks like. Examples:

Target agreed to your price
Target accepted your offer or negotiated one you accepted
Target confirmed they want to move forward

When triggered, the AI will:

Send a sign-off email
Set the milestone (e.g. Almost There)
Apply your chosen tags (e.g. "AI Won")
Stop sending further replies

Negative Exit Criteria ( A Loss )

Define when to stop pursuing an opportunity. Examples:

Target said they are not interested
Price negotiation failed beyond your hard cap
No response after all chase attempts

When triggered, the AI will:

Mark the opportunity as Lost
Apply your chosen tags (e.g. "AI Lost")
Stop all further automated replies

💡 TIP

Use tags like "AI Won", "AI Lost", and "AI Negotiated" so you can track AI performance over time and identify which campaigns benefit most from automation.

7. Wins and Losses

✅ A Win — What Happens

1	AI detects the positive exit criteria was met e.g. "Target confirmed she wants to move forward and will publish"

2	AI sends a sign-off email e.g. "Fantastic! Really great to hear that."

3	Milestone is set to Almost There Your team picks it up from here - e.g. to write and submit content

4	Tags are applied and AI stops replying e.g. "AI Won" tag is added to the opportunity

❌ A Loss — What Happens

1	AI detects the negative exit criteria was met e.g. "Target said they are not interested"

2	Opportunity is marked as Lost AI applies your configured tags and stops all further activity

📋 NOTE

For link inserts and niche edits, the AI can drive the conversation further, all the way to placement confirmation, before marking as Won. Guest posts stop at Almost There because content still needs to be written.

8. Running an Agent on an Existing Inbox

When you activate an AI Agent, it processes new emails from that point forward. Emails that arrived before activation are not touched, unless you trigger it manually.

How to Run the Agent on Existing Emails

Go to your campaign inbox
Select the emails you want the AI to process (you can select all)
Click Run AI Reply Agent
Wait approximately one minute for the agent to process

✅ Result

Emails that disappear from your inbox have moved into the Agent Flow, the AI is now managing them. Emails that remain were skipped (see Chapter 10 for how to handle these).

Checking the Agent Flow

After running, go to the Agent Flow view to see:

Which emails were processed
The current milestone for each opportunity
Whether chasing has been enabled

📋 NOTE

Emails already responded to before the AI was configured will be evaluated by the AI as if new, it will send a reply or flag the email for manual review.

9. Reviewing Successful Wins

Every won opportunity in the Agent Flow includes a complete audit trail of what the AI did and why it considered the goal reached.

What You See in the Agent Flow for a Win

Event	What it means
Agent was executed	AI read the email and composed a reply
Draft was created	Reply was written and queued
Email was sent	Reply was delivered to the target
Chasing was active	Follow-ups were scheduled in case of no response
Goal was met	Exit criteria were satisfied — opportunity closed as a win

The AI's Reasoning

When you click into a won opportunity, you will see:

A plain-English summary of why the AI declared the goal reached
A direct quote from the target's email that triggered the decision
A record of the sign-off email sent by the AI

💡 TIP

Regularly review won and lost opportunities in the Agent Flow. If the AI is declaring wins too early, tighten your positive exit criteria. If it is missing genuine wins, broaden them.

10. Running Replies Marked as Skipped

When the AI cannot respond confidently, it skips the reply and flags the email in your inbox rather than risk sending something incorrect.

Common Reasons for a Skip

Reason	Example
Missing personalization fields	Target asked about topics but no topics are in the opportunity record
Missing opportunity details	Name, blog URL, or pricing not filled in
Instructions gap	Target asked a question your agent instructions do not address

How Skips Appear

Skipped emails stay in your inbox with a tag indicating 'AI Reply Failed'. Open the opportunity detail to see the exact skip reason.

📋 NOTE

Example skip reason: "Email asked about pricing and content topics, but personalization fields (name, topics) were empty, impossible to generate an accurate response."

How to Resolve a Skip

Read the skip reason in the opportunity detail
Fill in any missing fields on the opportunity record
Update your agent instructions to cover the missed scenario
Manually reply if the situation is too nuanced for automation

Pitchbox will sometimes suggest specific instruction edits to prevent the same skip from happening again.

⚠️ IMPORTANT

Skips are a safety feature. A skipped email in your inbox is far better than a wrong reply being sent to a potential link partner.

11. Building Your Own AI Agent

Pitchbox gives you a pre-configured library to start from. You can use these as-is, customize them, or build entirely new agents for specific niches.

How to Create a New Agent

Click AI Agents in the top-right corner
Click New AI Agent
Choose a template from the AI Prompt Library (Chase-ups or AI Replies)
Customize the Campaign Goal, Instructions, and Exit Criteria for your niche
Click Preview to validate the output
Click Create and assign the agent to a category

Customizing for a Niche - Example (Fitness)

Setting	What to change
Topics	List your specialisms: fitness equipment, weight loss, nutrition
Payment cap	"Do not pay more than $50 per placement"
Tone	Adjust to match your brand voice
Agent name	Use a clear name: "Guest Post — Fitness"

Applying to Campaigns

From the new agent's detail page, select your project
Choose the campaigns to apply it to
Select a Playbook to pair it with
Save — the agent activates immediately on new inbound emails

💡 TIP

Name agents descriptively: e.g. Guest Post - Fitness or Link Insert - Finance. This makes it fast to select the right agent when configuring new campaigns.

12. Preview and Next Steps

Before going live, use the Preview function to validate every new agent and playbook configuration.

Final Pre-Launch Checklist

Item	Status check
1️⃣ AI Agent configured	Campaign goal, instructions, exit criteria all set
2️⃣ Payment rules defined	Hard cap and negotiation range specified
3️⃣ Topics and tone set	Instructions match your brand voice and niche
4️⃣ Preview tested	Run 2-3 times and output looks correct
5️⃣ Playbook conditions set	Won, Lost, Almost There milestones excluded
6️⃣ Sending mode chosen	Automatic send or draft for review
7️⃣ Wait time configured	5-20 min window set for natural responses
8️⃣ Chase-up scheduling active	Adaptive chase-up enabled and timing set
9️⃣ Agent applied to campaign	Project and campaigns selected, playbook paired

Building Out Over Time

Once your first agent is running, expand systematically:

Create niche-specific agents for each vertical you work in
Review the Agent Flow weekly to refine exit criteria
Expand to link insert and niche edit campaigns where AI can confirm placement
Check skipped emails weekly to improve instruction coverage

✅ Result

Fastest way to get started: pick one active campaign, apply a pre-configured agent with the Send & Chase Playbook, and monitor the Agent Flow for one week. Use what you see to tune the configuration before rolling it out across all campaigns.