How to build a five-agent prospecting workflow
This guide is based on The Case for Autonomous Marketing: Claim 13 Hours Back Each Week webinar, published on the AI Lab by ActiveCampaign.
Get the quick-start guide
What will I accomplish with this guide? By the end of this guide, you’ll have a five-agent outbound workflow that pulls fresh prospect data, drafts personalized emails in your voice, generates follow-ups from your call notes, and logs every touch into a single tracker. It’s based on the workflow Amanda Pressner Kreuser of Masthead walked through in this webinar—the same flow that gives her four to five hours back each week and 234 hours back each year.
Before you start, you’ll need:
- A LinkedIn Sales Navigator seat (for prospect search)
- A Phantom Buster account (for the data extraction phantom)
- A ChatGPT Plus or Team account that can build a CustomGPT
- 10–15 of your own past outbound emails that earned a response (these become the GPT’s training data)
- A short writing sample that captures your voice (a couple of paragraphs is enough)
- A Google Workspace account with permission to run an Apps Script
- A blank Google Sheet to act as the outreach tracker
Quick reference
- Total time: 4–6 hours to set up; runs on autopilot afterwards
- Tools needed: LinkedIn Sales Navigator, Phantom Buster, ChatGPT (CustomGPT), Google Sheets, Google Apps Script
- Key output: A multi-agent workflow that researches prospects, drafts outbound and follow-up emails in your voice, and logs every touch automatically
Watch this section
For full context on the following topics, watch these sections of the webinar:
- Why Amanda built this workflow and what changed for her: [15:06]–[16:11]
- Sales Navigator to Phantom Buster to CustomGPT chain: [16:11]–[17:14]
- Reviewing drafts and feeding the GPT call notes for follow-ups: [17:14]–[17:48]
- Logging every touch with a Google Apps Script: [17:48]–[18:19]
The workflow
Phase 1: Source prospects from Sales Navigator
After this phase, you’ll have: a saved Sales Navigator search returning your ideal prospects.
- Define your ICP filters: in Sales Navigator, build a lead search using the filters that match your ideal customer (industry, company size, role, geography, and any signal you trust).
- Save the search: name it something Phantom Buster can recognize later (e.g., “ICP: agency founders, US, 11–50”).
- Spot-check the results: scroll through 25–50 leads and confirm they look right. A noisy search compounds when an agent runs against it.
Phase 2: Pull prospect data with Phantom Buster
After this phase, you’ll have: a CSV of prospect records with job title, company info, recent posts, and mutual connections.
- Connect your LinkedIn account: in Phantom Buster, link your account so the phantom can run as you.
- Choose the Sales Navigator Search Export phantom: point it at the URL of the search you saved in Phase 1.
- Set the fields you want pulled: at minimum, name, title, company, recent LinkedIn post, and mutual connections. These are the same fields Amanda’s GPT reads in Phase 3.
- Schedule a recurring run: weekly is a good starting cadence. The output is a CSV you can hand to the GPT, or sync directly into a Google Sheet.
Phase 3: Train a CustomGPT on your voice and your winners
After this phase, you’ll have: a CustomGPT that drafts outbound emails in your voice, trained on the templates that have actually worked for you.
- Open the CustomGPT builder in ChatGPT: click “Explore GPTs” → “Create” → “Configure.”
- Upload your 10–15 past winning emails: these are the templates that earned real responses. Amanda specifically used hers because they were proven, not generic.
- Upload a writing sample: a paragraph or two that captures your voice (sentence length, tone, signature phrases).
- Write the system instructions: describe the goal (draft a first-touch outbound email per prospect), the inputs the GPT will receive (the Phantom Buster fields), and the constraints (length, tone, no clichés, never invent facts about the prospect).
- Test it on three real prospects: paste in three rows from your CSV and read the output. Refine the instructions until the drafts sound like you.
She trained it on 15 of her actual email templates, the ones that she knew worked because they had gotten real responses over the years. She also fed it writing samples so it could match her voice and tone.
Phase 4: Generate first-touch drafts and review every one
After this phase, you’ll have: personalized first-touch emails for each prospect, edited and ready to send.
- Hand the GPT one prospect at a time: paste the row from your Phantom Buster CSV and ask for a draft.
- Read every draft before it goes out: Amanda specifically does not auto-send. She edits each one and adds her human touch.
- Capture edits as feedback: when you change a phrase, paste the revision back into the GPT’s instructions over time so it drifts toward your voice rather than away from it.
Phase 5: Generate follow-ups from your call notes
After this phase, you’ll have: a follow-up email per prospect call that references specific things you discussed.
- Take notes during the call (or record it): Amanda uses either notes or recordings, whichever is faster for you.
- Paste the notes into the GPT: ask it to draft a follow-up that references the specific points you covered.
- Review and personalize: same rule as Phase 4. The AI handles the heavy lifting; you handle the human touch.
Phase 6: Log everything with a Google Apps Script
After this phase, you’ll have: an outreach tracker that updates itself with who you contacted, who responded, and where they sit in the pipeline.
- Create the tracker sheet: columns for prospect, company, date contacted, draft status, response, engagement level, and pipeline stage.
- Write (or generate) a Google Apps Script: ask the CustomGPT to draft an Apps Script that appends a row whenever you log a new send and updates the engagement column when a reply lands. Open Extensions → Apps Script in your sheet to paste it.
- Trigger the script from your sheet or inbox: depending on how the script is wired, you can run it on a button click in the sheet or on an inbox label/filter event.
- Review weekly: scan the tracker once a week to see who’s gone quiet and feed those names back into Phase 5 for a follow-up.
Related
More data from the AI Lab.