AI-Enriched Cold Outreach: Research → Draft → QA → Write-back
What this template does
Automates cold email drafting from a lead list by:
- Enriching each lead with LinkedIn profile, LinkedIn company, and Crunchbase data
- Generating a personalized subject + body with Gemini
- Auto-reviewing with a Judge agent and writing back only APPROVED drafts to your Data Table
Highlights
- Hands-off enrichment via RapidAPI; raw JSON stored back on each row
- Two-agent pattern: Creative Outreach Agent (draft) + Outreach Email Judge (QA)
- Structured outputs guaranteed by LangChain Structured Output Parsers
- Data Table–native: reads “unprocessed” rows, writes results to the same row
- Async polling with Wait nodes for scraper task results
How it works (flow)
- Trigger: Manual (replace with Cron if needed)
- Fetch leads: Data Table “Get row(s)” filters rows where
email_subject is empty (pending)
- Loop: Split in Batches iterates rows
- Enrichment (runs in parallel):
- LinkedIn profile: HTTP (
company_url) → Wait → Results → Data Table update → linkedin_profile_scrape
- LinkedIn company: HTTP (
company_url) → Wait → Results → Data Table update → linkedin_company_scrape
- Crunchbase company: HTTP (
url_search) → Wait → Results → Data Table update → crunchbase_company_scrape
(All calls use host cold-outreach-enrichment-scraper with a RapidAPI key.)
- Draft (Gemini): “Agent One” composes a concise, personalized email using row fields + enrichment + ABOUT ME block.
- Prep for QA: “Email Context” maps
email_subject, email_content, and email for the judge.
- QA (Judge): “Judge Agent” returns
APPROVED or REVISE (brief feedback allowed).
- Route:
- If
APPROVED → Data Table “Update row(s)” writes email_subject + email_body (a.k.a. email_content) back to the row.
- If
REVISE → Skipped; loop continues.
Required setup
Data Table: “email_linkedin_list” (or your own) with at least:
email, First_name, Last_name, Title, Location, Company_Name, Company_site,
Linkedin_URL, company_linkedin (if used), Crunchbase_URL,
email_subject, email_body,
linkedin_profile_scrape, linkedin_company_scrape, crunchbase_company_scrape (string fields for JSON).
Credentials:
- RapidAPI key for
cold-outreach-enrichment-scraper (store securely as credential, not hardcoded)
- Google Gemini (PaLM) API configured in the Google Gemini Chat Model node
ABOUT ME block:
Replace the sample persona (James / CEO / Company Sample / AI Automations) with your own.
Nodes used
- Data Table
- HTTP Request:
- AI Agent:
- Google Gemini Chat Model
- Split in Batches: Main Loop
- Set: RapidAPI-Key
Customization ideas
- Process flags: Add
email_generated_at or processed boolean to prevent reprocessing.
- Human-in-the-loop: Send drafts to Slack/Email for spot check before write-back.
- Delivery: After approval, optionally email the draft to the sender for review.
Quotas & costs
- RapidAPI: Multiple calls per row (three tasks + result polls).
- Gemini: Token usage for generator + judge per row.
Tune batch size and schedule accordingly.
Privacy & compliance
You are scraping and storing person/company data.
Ensure lawful basis, respect ToS, and minimize stored data.