An inbox that prepares replies by itself
Automation 12 min
0/21
Business problem
The inbox becomes a chaos center when campaigns run: replies mix with announcements, the deal context is in the CRM (but no one looks), follow-ups slip away, and tasks ('check the price', 'ask about the decision maker') aren't saved anywhere. AI Inbox Automation tidies this up: every reply → context + draft + task + CRM update.
Proven method
The inbox workflow in 7 steps:
- FETCH: a reply enters the system (from module 9 — after classification),
- ANALYZE: AI reads the message — what the client wrote, what they want, what questions they have,
- CLASSIFY INTENT: the same classification as module 9 (INTERESTED/QUESTION/OBJECTION/...) — it decides the reaction,
- CRM CONTEXT: the system takes the deal data from the CRM (stage, history, notes, what was agreed) — so the reply isn't from a vacuum,
- GENERATE DRAFT: AI writes a response draft (CRM context + the message content + the company voice),
- ASSIGN A TASK: AI creates a task (e.g. 'check the price for X' or 'book the demo') with a deadline,
- UPDATE CRM: the system updates the deal (a note from the reply, a stage change, a new next step). HUMAN APPROVAL: a draft is NOT sent automatically (basic version) — it goes to an approval queue; the salesperson clicks 'send' or edits. Rules:
- CRM context is a quality condition (a reply without context is generic),
- a task without a deadline doesn't exist,
- the CRM updates with every reply (not 'on weekends'),
- human approval — at least sampling at large scale.
Market example
- BAD
- Manual, inconsistent execution or chaotic automation lacking data validation.
- GOOD
- A reply: 'Hello, how much does implementation cost for a company with 5 people in sales?' System: Analyze → 'asks about the price for 5 people'. Classify → QUESTION. CRM context → 'deal: MebelSoft, discovery on Wednesday, they need email integration'. Draft: 'Hi Anna, implementation for a 5-person team with email integration: [X]/month — during discovery we agreed that email integration is key, it's in the scope. The demo on Wednesday will show it specifically. Let's book 15 min?' Task: 'book the demo with MebelSoft — by Wednesday'. CRM: a note + stage + next step. You review and click send. All in 2 minutes of work instead of 20.
Ready-to-use prompt
You are an inbox automation architect. Context: I get [X] replies/day. CRM: [CRM]. My voice: [VOICE DESCRIPTION]. Tools: [AI + WORKFLOW + EMAIL TOOL]. Task: Design my AI inbox: 1. A schema: Fetch → Analyze → Classify → CRM context → Draft → Task → CRM update (what each step does) 2. CRM context: what the system takes from the CRM into the reply (fields) 3. A response draft prompt: a template (context + message + voice) 4. Task management: how AI creates tasks (a template + deadline + owner) 5. CRM update: what to update (note, stage, next step) and how often 6. Human approval: how to organize approval (a queue, sampling) 7. An end-to-end example: a reply → everything (describe on 1 example) Format: 7 sections.
Model rekomendowany: Claude 3.5 Sonnet / GPT-4o Copy and paste into ChatGPT, Claude or Perplexity
ALTERNATYWNE SCENARIUSZE
Variations & alternative scenarios
Basic version: draft + task, without auto-CRM (a human pastes).
Full version: full CRM integration (notes, stages, next steps automatically).
Practical exercise
Design the inbox workflow (7 steps). Write a response draft prompt (with CRM context). Write the rule: what you approve before sending.
Zapisano w przeglądarce