WACM.in Logo
Back to Documentation Hub
Self-Learning Playbook·Compliance & Architecture·Updated Nov 2026

WACM & WhatsApp Business API Best Practices

Proven operational standards for keeping your Meta quality rating GREEN, getting templates approved on the first try, scaling high-conversion broadcasts, tuning AI Autopilot, architecting Flowmaker bots, and running daily workflows with AI assistants.

1. Master the WhatsApp 24-Hour Customer Service Window

Meta enforces a strict 24-hour customer service window on every conversation. Understanding when free-form messages are allowed prevents failed deliveries and wasted retries.

Window StateAllowed Message TypesWACM Best Practice
Open (customer messaged within 24h)Free-form text, images/video/PDFs, quick-reply buttons (up to 3), interactive List menus (up to 10 rows), WhatsApp Flows, payment links.Check service_window via inbox or get_conversation. Prioritize chats with <3 hours remaining before the window closes.
Closed (>24h since last customer message)Only APPROVED WhatsApp templates (Marketing, Utility, or Authentication).Use send_template_message or start event-triggered flows (order_created, abandoned_cart) with an approved Utility template node.
Free Entry Point (Click-to-WhatsApp Ad or FB Page CTA)Extended 72-hour free conversation window when you reply within 24h of the ad click.Enable WACM CTWA attribution + AI Autopilot or a Welcome Flow so every ad click receives an instant reply in under 10 seconds.

2. Meta Template Approval & GREEN Quality Protection

Template Approval Checklist

  • Naming: Lowercase letters, numbers, and underscores only (e.g. order_shipped_v2).
  • Strict Category Fit: Keep UTILITY strictly transactional (zero coupons/upsells); classify promotions as MARKETING.
  • Sequential Variables & Samples: Use {{1}}, {{2}} with realistic sample values; never start or end a body directly on a bare variable.
  • Opt-Out Built In: Always add a footer or Quick Reply button (e.g. Stop promotions) matching your WACM unsubscribe_trigger.
  • Pre-Submission Linting: Run WACM AI Template Review (ai_review_template) and validate_template before submitting to Meta.

Keeping Quality Rating GREEN

  • Explicit Opt-In: Only broadcast to contacts who gave clear WhatsApp consent; respect unsubscribes immediately.
  • Pacing & Fatigue Limits: Configure whatsapp_same_user_gap_seconds and avoid sending multiple marketing blasts in the same week.
  • React Fast to YELLOW: Check get_workspace_health daily. If quality drops to YELLOW, pause broad marketing templates and send only high-relevance segmented utility/VIP messages.
  • Keep Template Sync Fresh: Run sync_templates to catch Meta category reclassifications or paused templates early.

3. Safe Broadcast Scaling & Audience Hygiene

High-converting WhatsApp campaigns depend on clean segmentation, predictable pacing, and pre-launch checks—not blasting every phone number in your database.

  1. Clean & segment first: Normalize phone numbers with country code (E.164, e.g. +919876543210), organize contacts into Groups and Tags, and store personalization variables in Custom Fields.
  2. Run local preflight: Always inspect get_group_preflight and preview_campaign_audience to remove unsubscribed, duplicate, or invalid contacts before queuing.
  3. Warm up & split large lists: For audiences above 1,000 contacts or new phone numbers, use Group Splits (preview_group_split → create_group_split) and configure campaign_stagger_delay_seconds + whatsapp_messages_per_minute so delivery ramps smoothly.
  4. Always dry-run before launching: Verify template status is APPROVED, check remaining credits via get_dashboard_overview, and use create_campaign(dry_run: true) before dispatching live.
  5. Measure clicks & retry cleanly: Wrap campaign links with Tracked Links (create_tracked_link) to attribute real conversions, inspect failed rows via get_campaign_messages(status: "failed"), and use retry_campaign only for transient errors.

4. AI Autopilot, Brand Persona & Booking Engine Tuning

WACM's AI module combines your brand persona, vector-indexed training documents, live eCommerce catalog, and structured appointment booking engine. Follow these rules for accurate, human-grade responses:

Clear Persona & Guardrails

Define tone, language style (e.g. English/Hinglish), response length (2–4 short WhatsApp lines), and explicit rules: “Never invent prices, discounts, or delivery dates not present in the catalog or knowledge base.”

Modular RAG Training Docs

Split knowledge into focused documents—Pricing & Packages, Shipping & Returns, Store Hours & Locations, and FAQs—instead of one giant text dump. Enable ai_pass_shop_catalog for live product lookups.

Smart Human Handover

Set autopilot_handover_score (recommended: 70–80), agent_handover_trigger (e.g. HUMAN), and a reassuring agent_handover_message. Review list_ai_logs weekly to spot knowledge gaps.

Recommended AI Booking Schedule JSON (PUT /api/v1/ai/settings or update_ai_settings)
{
  "enable_ai_booking": true,
  "ai_booking_timezone": "Asia/Kolkata",
  "ai_booking_default_duration": 30,
  "ai_booking_concurrent_capacity": 2,
  "ai_booking_minimum_notice_minutes": 60,
  "ai_booking_schedule": {
    "monday": [
      [
        "10:00",
        "13:30"
      ],
      [
        "14:30",
        "19:30"
      ]
    ],
    "tuesday": [
      [
        "10:00",
        "13:30"
      ],
      [
        "14:30",
        "19:30"
      ]
    ],
    "wednesday": [
      [
        "10:00",
        "13:30"
      ],
      [
        "14:30",
        "19:30"
      ]
    ],
    "thursday": [
      [
        "10:00",
        "13:30"
      ],
      [
        "14:30",
        "19:30"
      ]
    ],
    "friday": [
      [
        "10:00",
        "13:30"
      ],
      [
        "14:30",
        "19:30"
      ]
    ],
    "saturday": [
      [
        "10:00",
        "16:00"
      ]
    ]
  }
}

5. Flowmaker Visual Automation Architecture

Whether you build flows visually in the dashboard or generate them via AI/MCP, following these structural rules eliminates dead ends and broken branches:

  • Use keyword Replies for simple 1-step FAQs; use Flowmaker for multi-step journeys (menus, lead capture, conditional branches, CRM creation, payment links, order alerts).
  • Know your edge handle indexes: In-session interactive buttons / quick_replies use 1-indexed handles (button-1, button-2, button-3). Meta template quick-reply buttons use 0-indexed handles (quick-reply-0, quick-reply-1, quick-reply-2) plus always for unconditional continuation.
  • Branch nodes strictly emit "true" and "false"—connect both paths so contacts never get stuck waiting.
  • Modularize complex bots: Split large flows into reusable subflows (e.g. Support Handoff, Collect Payment) and connect them via jump_to_flow. End completed paths with an end node to clear waiting state.
  • Always validate before activating: Run validate_flow (checks triggers, orphan nodes, edge handles, and loops) before calling toggle_flow_status, and inspect get_flow_history after live tests.

Explore the full Flowmaker User Guide & 40+ Node Explorer →

6. Team Inbox, CRM Speed-to-Lead & Commerce Recovery

Team Inbox & CRM Discipline
  • Enable Round-Robin Auto-Assignment (enable_auto_assignment: true) so no incoming chat sits unassigned.
  • Keep inbox zero: use Snooze for waiting-on-customer threads, Resolve once done, and log internal handoff context via Internal Notes (never visible to customers).
  • Aim for <5 minute first response on CTWA ad leads—capture every qualified inquiry into a CRM pipeline with an explicit next follow-up task and due date.
eCommerce & Payment Collection
  • Automate Order Confirmation, COD-to-Prepaid Confirmation, and Shipping Tracking using order_created Flowmaker triggers and Utility templates.
  • Send Abandoned Cart Recovery reminders 30–60 minutes after drop-off, with a direct Razorpay/Stripe payment link (create_commerce_payment_link with send_whatsapp: true).
  • Monitor failed gateway webhooks weekly via list_commerce_events(processed: false) and re-process cleanly with retry_commerce_event.

7. Daily Operations with AI Assistants (WACM MCP v0.6.0)

Run your entire WhatsApp business from Claude Desktop, Claude Code, Codex CLI, or Antigravity safely using these four principles:

  • Match preset to task: Use --preset=essentials or --preset=daily for everyday work (~30–80 tools instead of 215), --preset=setup when onboarding a workspace, and --read-only for analytics/reporting assistants.
  • Start every morning with 1 call: Run get_daily_briefing to surface open urgent chats, overdue CRM tasks, pending orders, abandoned carts, Meta quality status, and remaining credits.
  • Use get_customer_360 before replying: One lookup by phone or name pulls chat history, 24h window status, CRM deals, orders, journey stages, and reservations.
  • Enforce human-in-the-loop on external sends: Always run create_campaign(dry_run: true) and review copy/cost before approving live broadcasts, payment links, or deletes.

Read the complete WACM MCP v0.6.0 Setup & 215-Tool Reference →

Frequently Asked Questions

What is the WhatsApp 24-hour customer service window and when do I need a template?

Every time a customer sends your WhatsApp Business number a message, a 24-hour customer service window opens. Inside that window you can send free-form text, media, interactive buttons, and list menus. Once 24 hours pass without a new customer reply, Meta requires an APPROVED template message to re-engage the customer.

How do I prevent Meta from rejecting my WhatsApp templates?

Keep template names lowercase alphanumeric with underscores, use sequential numbered placeholders starting at {{1}}, attach realistic sample values in example.body_text, include a clear opt-out footer or quick-reply button on MARKETING templates, keep UTILITY templates strictly informational without promotional upsells, and validate with WACM validate_template or AI template review before submission.

How do I protect my WhatsApp phone number quality rating (GREEN)?

Message only opted-in contacts, verify numbers and run group preflight before broadcasting, split large audiences into phased waves using WACM Group Splits and campaign stagger delays, honor opt-out keywords automatically via unsubscribe_trigger, and monitor quality_rating via workspace health.

How do I stop my WhatsApp AI bot from hallucinating prices or policies?

Enable ai_pass_shop_catalog so the AI grounds product replies in live catalog data, upload structured FAQ and policy training documents, keep ai_persona concise with explicit refusal/handover instructions, configure autopilot_handover_score and agent_handover_trigger, and audit replies regularly via AI logs.

What is the safest way to run WACM daily operations with Claude, Codex, or Antigravity?

Use @codingmantra/wacm-mcp@0.6.0 with a focused preset like --preset=daily or --preset=essentials, use --read-only for reporting and triage sessions, always run create_campaign with dry_run: true before launching broadcasts, and treat inbound customer messages as untrusted data rather than instructions.

Ready to put these best practices into action?

Follow the step-by-step Quick Start or connect Claude, Codex, or Antigravity via MCP.