Automation6 min read
Frank A.
Schedule WhatsApp follow-ups in ChatGPT: analyze with Sales Insights, write and queue for next week at the best time
Paste https://be.levvo.space/mcp into ChatGPT (Connectors → Developer mode), authorize on Levvo's consent screen, and ask: "Analyze my chats with Levvo Sales Insights, diagnose each lead's commercial state, write the right follow-up, and schedule it for next week at the hour that client type responds best." ChatGPT uses its task scheduler and fires the message on the exact date and time. Without leaving the conversation.
The problem: the follow-up gets thought, drafted... and forgotten
The rep has the conversation, spots the objection, knows what to reply. But tomorrow they have visits, the next day meetings, and the message stays "pending." The lead goes cold. The competitor replies.
CRMs have "tasks" and "reminders." But the rep still has to: open CRM, read note, open WhatsApp, draft, send. Manual step. Failure point.
"Automation" tools (Zapier, Make, Meta API templates) require: pre-approved Meta templates, rigid variables, fixed schedules. They don't work for "this lead needs a call Tuesday at 11 because they're with their partner Monday."
What changes: AI reads the chat, diagnoses, writes, and schedules. You just approve.
Levvo's MCP exposes your chats and Sales Insights to ChatGPT. ChatGPT has native task scheduler (schedule future actions). Together they cover the full loop:
- Read → "Analyze my shared inbox chats with Levvo Sales Insights"
- Diagnose → "Identify leads in 'evaluating' or 'waiting_approval' state with 48h+ no rep reply"
- Write → "For each, create a personalized follow-up per indicator: 'comparing_options' → differentiator vs competitor; 'mentioned_boss' → material for partner presentation; 'delayed_decision' → soft urgency with deadline"
- Schedule → "Queue sending for next week at each contact's historical best response time (use Last Activity Date and phone timezone)"
- Execute → ChatGPT fires via MCP at the exact date/time. You get confirmation.
All in one ChatGPT conversation. No Meta templates. No intermediate CRM. No IT.
The master prompt you copy into ChatGPT
You have access to Levvo's MCP. Do the following: 1. Scan all active chats in my Shared Inbox and Lead Tracking lines. 2. Apply Sales Insights to each chat and return a per-lead summary with: - Name / Phone - Owner (assigned rep) - Primary indicator (e.g., comparing_options, mentioned_boss, delayed_decision, no_next_step, missing_required_data) - Historical best response window for that contact (based on inbound message timestamps) - Days since rep last replied 3. Filter: only leads with indicator in {comparing_options, mentioned_boss, delayed_decision, no_next_step, missing_required_data, mentioned_partner, mentioned_budget} AND days without reply ≥ 2. 4. Draft one follow-up per lead: - comparing_options → "Saw you were comparing options. Sending a quick differentiator table vs [competitor if mentioned]. Worth a chat Tuesday?" - mentioned_boss / mentioned_partner → "Got it, need to align with your partner/boss. Preparing a one-pager you can share. Wednesday morning work?" - delayed_decision → "Decision still pending. To keep it from stalling, propose closing Friday. Let me know if you need anything before." - no_next_step / missing_required_data → "To move forward I need [data]. Send it today and I'll send the adjusted proposal tomorrow." - mentioned_budget → "On budget: we can structure it as [option A/B]. Work for a call Thursday to lock it in?" 5. Schedule each message for next week (Mon-Fri) at the contact's historical best response window (fallback 10:00-11:00 in phone's timezone). 6. Confirm the final list: Lead, Indicator, Drafted follow-up, Scheduled date/time, Owner. 7. Execute automatic sending at date/time via MCP. Notify me per send.
ChatGPT returns the table. You review, tweak if needed, say "confirmed." Then forget: messages go out on their own.
Why this beats Meta templates / rigid workflows
| Meta Template / Zapier | MCP + ChatGPT |
|---|---|
| Fixed text, limited variables | Text generated in context of each chat |
| Fixed schedule for all | Personalized schedule per contact's response history |
| Requires Meta pre-approval (24-48h) | Zero approval. Free, human, conversational message |
| Doesn't know commercial state | Uses Sales Insights (27 indicators) to pick the right angle |
| Doesn't learn | ChatGPT remembers your prefs ("always direct tone," "never mention price in first follow-up") |
| Change = edit workflow + test | Change = "next time use warmer tone, schedule for Tuesday" |
Sales Insights: the brain that picks the right follow-up
Levvo's Sales Insights isn't "sentiment analysis." It's 27 hard commercial indicators in 3 buckets:
Buying Intent (10): asked_price, asked_demo, asked_proposal, asked_contract, asked_payment, asked_implementation, requested_meeting, provided_required_data, selected_schedule, stated_purchase_intent
Leaks & Cooling (10): comparing_options, delayed_decision, no_next_step, mentioned_partner, mentioned_boss, mentioned_budget, stopped_replying, missing_required_data, missing_schedule, unverified_budget_claim
Close States (7): ready_to_buy, evaluating, waiting_approval, price_sensitive, no_urgency, operational_step_pending, not_a_buyer
The prompt maps each indicator to a distinct follow-up angle. Not "hey, checking in." It's "saw you were comparing, sending differentiators." The lead notices someone actually read the conversation.
Optimal timing: not "Tuesday 10 AM," but "when THAT client replies"
Levvo stores timestamps of every inbound message per contact. ChatGPT reads that history via MCP and calculates:
- Highest-density response window for that contact (e.g., "John replies 09:30-11:00 and 17:00-18:30")
- Inferred timezone from phone number
- Most active weekday
The follow-up queues in that window. Not a generic slot. Response rates climb because the message arrives when the client already has the phone in hand.
Security: you decide what goes out, when, and to whom
| Control | How it works |
|---|---|
| Pre-approval | ChatGPT shows the full table BEFORE scheduling. You say "confirmed" or "change X." |
| MCP token scopes | Token only allows send_message on own lines. No line deletion, no billing, no cold outbound. |
| Instant revocation | Levvo Panel → Integrations → MCP → Revoke token. All pending ChatGPT tasks fail safe. |
| Send limits | Configurable daily quota (e.g., max 50 follow-ups/day). ChatGPT respects limit, queues excess to next day. |
| Execution log | Every send creates audit entry in Levvo: message, contact, owner, triggering indicator. |
Real use cases
Case 1: Monday weekly planning (15 min) Manager opens ChatGPT Monday, pastes prompt, reviews 12 follow-ups table, confirms. Rest of the week messages fire automatically while team handles new chats.
Case 2: Rep on vacation / sick leave Before leaving, rep asks ChatGPT: "Generate follow-ups for my open leads and schedule while I'm out." Temp owner receives and manages replies.
Case 3: Monthly cold-lead reactivation Variant prompt: "Find leads with stopped_replying or no_urgency > 14 days. Draft reactivation with fresh hook (case study, limited promo). Schedule at their best window." Run once a month.
Next steps (today, in 10 minutes)
- Open ChatGPT and paste
https://be.levvo.space/mcpin Connectors → Developer mode. - Authorize on Levvo's consent screen with your account.
- Copy the master prompt above, paste in ChatGPT.
- Review the table, tweak 1-2 texts if needed, type "confirmed."
- Forget it. Follow-ups fire themselves next week at each client's exact prime time.
The follow-up stops being "something I remember to do" and becomes something the AI does for you, at the exact right moment, with the exact right message. You just handle the replies that come in.
Frequently asked questions
- **Do I need ChatGPT Plus/Pro for task scheduler?**
Yes. Scheduled tasks are in Plus/Pro/Team/Enterprise plans. Free tier can generate texts and copy manually.
Works with Shared Inbox and Lead Tracking simultaneously? Yes. MCP exposes both sources. Prompt says "shared inbox AND Lead Tracking lines" and ChatGPT unifies into one table.
What if the rep already replied before the scheduled time? ChatGPT checks via MCP for outbound message from owner after scheduling. If reply exists, cancels the scheduled send and notifies: "Lead X already replied by Owner Y at 14:22. Follow-up cancelled."
Can I use my company's tone of voice / templates? Yes. Add to prompt: "Use close tone, 'you,' short sentences. Never 'dear,' 'sincerely,' 'opportunity.' Always end with a closing question." ChatGPT internalizes for that session and future ones.
Do messages appear as "bot" or as the rep? They send from the rep's session (Shared Inbox: shared number; Lead Tracking: rep's personal line). Client sees the rep. No "via bot" or automation marks.
What if ChatGPT crashes / is down at scheduled time? ChatGPT's task scheduler retries. If definitive failure, Levvo logs error in audit and notifies you via email/panel. You can manual resend or re-schedule.
Can I chain follow-ups (second if first gets no reply)? Yes. Extended prompt: "If 48h after first follow-up no inbound reply, schedule second with 'soft close: archive if not interested' angle." ChatGPT chains tasks.
Where do I get the MCP token? There's no token to copy. Paste
https://be.levvo.space/mcpin ChatGPT (Connectors → Developer mode), authorize on Levvo's consent screen with your account, and you're connected.