Iter/cases/whatsapp-ai-triage-engine
Messaging . Sample workflow . Open source
01
case /02

WhatsApp AI Triage Workflow

A sample workflow for businesses that still sort sales, support, scheduling, and follow-up messages by hand. It shows how AI can classify intent, prepare context, draft next steps, and bring in the right person when judgment matters.

Next.js 15Meta Cloud APIOpenAISupabasen8nTypeScript
Pattern
AI-assisted inbox triage
Best fit
Sales, support, booking, follow-up
Build path
4-week implementation sprint
Proof
Live sample + source available
Manual sorting
Reduced
AI handles first-pass classification
Response flow
Faster
drafts and routing are prepared sooner
Team context
Visible
intent, history, confidence, and next step
Control
Human-led
people decide on sensitive or unclear cases
/ problem

If WhatsApp is your front door, manual triage becomes the bottleneck.

A common small-business pattern is one shared WhatsApp number where sales questions, support issues, appointment requests, payment follow-ups, and random context all arrive together. Someone has to read each message, decide what it means, find the right information, and send it to the right place.

The issue is not that the team is slow. The issue is that repetitive sorting steals attention from the conversations that actually need care, nuance, and judgment.

/ approach

Use AI as a first-pass coordinator, not as the owner of the relationship.

The workflow reads each new message, classifies the intent, checks useful business context, and prepares the next step. Simple messages can receive a suggested or automated reply. Sensitive, unclear, or high-value conversations are routed to the right person with a summary, confidence score, and suggested action.

Architecture . WhatsApp -> intent -> context -> draft/action -> team review
WA
WhatsApp Cloud
AI
Intent + confidence
CTX
Customer context
ACT
Draft or action
TEAM
Team review
/ build

How the workflow works.

  • 01
    Capture the message

    The Meta Cloud API webhook receives the WhatsApp message, normalizes the payload, stores the conversation event, and keeps a clean record before any AI decision is made.

  • 02
    Classify intent and urgency

    An LLM labels the message as sales, support, booking, billing, follow-up, or another business-specific category. It also returns confidence, urgency, language, and missing information.

  • 03
    Prepare the next step

    The workflow can look up CRM context, previous messages, availability, or order status, then draft a reply, create a task, update a record, or trigger an n8n automation.

  • 04
    Route with context

    If the case needs judgment, the right person receives the summary, suggested reply, source message, confidence score, and audit trail instead of a bare forwarded chat.

/ outcome

Potential improvements when this replaces manual sorting.

The benefit is not a guaranteed metric; it depends on message volume, team habits, tools, and how much of the workflow is safe to automate. In the right setting, this pattern can reduce repetitive inbox sorting, make replies more consistent, keep context attached to every decision, and make escalations easier to review.

Business impact map

The financial upside is indirect but real: less time spent sorting messages, fewer leads or support requests left waiting, and better follow-through on conversations that already reached the business.

Operations
Less inbox drag

The team spends less time deciding where messages go and more time resolving the conversations that matter.

Customer experience
Earlier useful replies

Sales, support, and booking messages can get a first useful response or prepared next step sooner.

Revenue protection
Fewer stalled opportunities

New leads, payment questions, and follow-ups are less likely to sit unnoticed in a shared inbox.

Management visibility
Clearer handoffs

Intent, confidence, suggested action, and conversation history stay attached when a person takes over.

/ takeaways

Where AI fits.

  • AI is useful for classification, summarization, drafting, routing, and repetitive follow-up.
  • People should stay in charge of sensitive conversations, pricing exceptions, complaints, and unclear intent.
  • Confidence scores and audit trails make the workflow easier to review and improve over time.
  • The best version connects to the tools the business already uses: CRM, calendar, support desk, payments, or inventory.