Problem
70–80% of closings for Indonesian SMBs and B2C businesses happen over WhatsApp. Leads come in fast — then die because replies are late: chats idle >24 hours, no follow-up reminders, and sales forgetting who they already touched. What's rotten isn't the product, it's the response time.
Recognize this?
- 7pm chat answered next noon — the customer has gone cold
- Follow-ups live in the sales rep's head, not in a system
- You only chase the leads who replied first
- Weekly recaps built from memory, not from data
✕ Now
- Leads replied to hours late
- Follow-ups vanish with the rep's memory
- Nobody knows which ones are rotting
✦ With Hatysa
- First touch < 5 minutes, whenever it arrives
- H+1 / H+3 / H+7 sequences run themselves
- Automatic idle-chat detection + weekly reply-rate report
What the agent does
The agent watches your business WhatsApp inbox, detects conversations that have stopped moving, and sends staged follow-ups in your brand tone — stopping itself the moment the customer replies or declines.
✦ Handled by the agent
- detects stalled chats, drafts and sends the staged follow-ups, stops itself once the customer replies, and reports reply rate weekly.
◆ Decided by you
- take over the chat when a buying signal appears, and sign off on the message tone up front.
Pipeline
WA Business webhook (Fonnte/WATI/WhatsApp Cloud API) → store conversation + status.
rules: >24 hours without a reply after a price question / >3 days without a close → enters the follow-up queue.
LLM drafts the follow-up from chat context (product asked about, last objection) and scores send-worthiness (no spamming clear decliners).
H+1 gentle → H+3 value (testimonial/catalog) → H+7 polite close. Human handoff when a buying signal appears.
weekly: reply rate before vs after, conversations revived, revenue touched.
Integrations
- WA Business API (Fonnte / WATI / Meta Cloud API)
- Read-only CRM (Salesforce/HubSpot) optional; runs fine without a CRM (state in Postgres)
- Report output → WhatsApp owner (and Mission Control if present)
Model stack
- GLM 5.3 Flash for drafting + scoring (prompt cache = brand & product templates)
- Qwen3.8 Flash when customer-sent images need reading (product/receipt photos)
- Cost estimate: < US$10/month for 5,000 conversations/month
Success metrics (tracked weekly)
Sprint deliverables
Agent deployed + connected to 1 business WA number
✓Brand tone playbook (approved before the agent sends any message)
✓Weekly report dashboard via WA
✓Flow documentation + code handover (perpetual)
✓Ready to light up this star?
Free 30-minute discovery call: we'll show you exactly how many credits this takes for your business.
Book a Discovery Call →