reply-classifier

name	reply-classifier
description	Read reply emails and classify them by intent (interested, objection, not now, unsubscribe) with recommended next actions.

Reply Classifier

Task

Read incoming reply emails from outbound campaigns and classify each by intent. Output a categorized list with recommended next actions so your team responds fast and appropriately.

Input Requirements

• Reply emails in one of these formats:
  • CSV with columns: from_email, from_name, subject, body, received_date, campaign_name
  • Raw email text (for single-reply classification)
  • File path to a folder of .eml or .txt files
• Copy frameworks from your CLAUDE.md (for generating suggested responses)

Instructions

1. Read each reply and classify into one of these categories:

Category	Description	Response SLA
Interested	Wants to learn more, asked a question, agreed to meet	Within 1 hour
Objection	Raised a concern but didn't shut the door (budget, timing, competitor, authority)	Within 4 hours
Not now	Timing isn't right but didn't say never ("circle back later", "not a priority right now")	Acknowledge + set reminder
Referral	Pointed you to someone else ("talk to my colleague", "reach out to X")	Within 2 hours
Unsubscribe	Wants to be removed ("remove me", "stop emailing", "unsubscribe")	Immediate - remove and confirm
Out of office	Auto-reply, OOO message	No action - re-engage after return date
Not relevant	Wrong person, wrong company, spam, bounce notification	Archive

2. For each classified reply, generate:

   • Category and confidence level (high / medium / low)
   • Suggested response using your copy frameworks:
     • Interested → propose a specific time, keep momentum
     • Objection → acknowledge, address with one line, soft re-ask
     • Not now → confirm, offer a resource, set a future date
     • Referral → thank them, ask for an intro or permission to use their name
     • Unsubscribe → confirm removal, professional close
   • CRM action: what to update (deal stage, next step, contact status, task)

3. Flag edge cases: replies that are ambiguous or could be read multiple ways → mark as "needs human review"

Output Format

CSV with columns:

from_email, from_name, company, received_date, category, confidence, suggested_response, crm_action, campaign_name

Plus a summary:

Total replies processed: N
- Interested: N (respond within 1 hour)
- Objection: N (respond within 4 hours)
- Not now: N (set reminders)
- Referral: N (follow up on referral)
- Unsubscribe: N (remove immediately)
- Out of office: N (no action)
- Not relevant: N (archive)
- Needs human review: N

Save to: ./data/replies/classified_[DATE].csv

Reply Routing

• Interested + Referral replies: Route to Slack #replies channel within 1 minute for fastest response. Include: sender name, company, reply snippet, suggested response, and CRM action.
• Objection replies: Route to Slack #replies with suggested rebuttal from copy frameworks.
• Unsubscribe: Auto-log in Google Sheets exclusion tracker (or CRM) - no Slack notification needed.
• Gmail integration: If using Gmail MCP connector, Claude Code can draft response emails directly in Gmail based on the suggested responses above. Review before sending.

Configure Slack webhook: set SLACK_WEBHOOK_URL in your .env file.

Example Usage

/reply-classifier

Input: ./data/replies/inbox_export_march_10.csv

Output:

Total replies: 24
- Interested: 5 → respond ASAP
  - Sarah Kim (DataFlow): "This looks relevant. Can you send a one-pager?"
    → Suggested: "Sure thing, Sarah. Here's our one-pager [LINK]. Happy to walk through it live - does Thursday 2pm work?"
    → CRM: Move to "Engaged", set task for Thursday

- Objection: 3
  - Mike Chen (Acme): "We just signed a 2-year deal with [COMPETITOR]."
    → Suggested: "Makes sense - [COMPETITOR] is solid for X. Where we're different is Y. Worth keeping in touch for when the renewal comes up?"
    → CRM: Set "Competitor" field, set reminder 18 months out

- Not now: 6
  - Alex Rivera (CloudCo): "Not a priority this quarter."
    → Suggested: "Totally get it. I'll circle back in Q3. In the meantime, here's a relevant case study if useful."
    → CRM: Set follow-up task for July 1

- Unsubscribe: 2 → remove from all sequences immediately
- Out of office: 4 → no action, earliest return: March 17
- Referral: 1 → follow up on referral to VP Engineering
- Not relevant: 2 → archive
- Needs human review: 1 → tone unclear, may be sarcastic