AI Reply Classification

When to Use

Use when processing inbound replies from outreach campaigns to auto-classify and route responses.

Classification Categories

Category	Description	Action
Interested	Wants to learn more, asks questions	Route to SDR
Meeting Booked	Confirms meeting or asks for time	Update CRM, notify AE
Not Now	Timing isn't right	Add to nurture, snooze 90 days
Not Interested	Clear rejection	Remove from sequence
Wrong Person	Not the decision maker	Ask for referral
Out of Office	Auto-reply, OOO	Snooze, retry later
Unsubscribe	Wants to opt out	Remove immediately

Prompt Template

Classify this email reply into one category:
interested, meeting_booked, not_now, not_interested,
wrong_person, out_of_office, unsubscribe

Reply text: """
{{reply_body}}
"""

Respond with JSON:
{ "category": "...", "confidence": 0.0-1.0, "reason": "..." }

Confidence Thresholds

>0.85: Auto-route, no human review
0.60-0.85: Route + flag for review
<0.60: Manual classification required

Implementation Notes

Use DeepSeek for cost efficiency on high volume
Fall back to GPT-4 for low-confidence classifications
Log all classifications for model evaluation
Review accuracy weekly, retune prompts monthly