#AI-Powered WhatsApp Group Monitoring: Detect Buying Intent in Real Time

The Problem with Keywords

Keyword monitoring is a blunt instrument. You add "buy", "price", "interested" to your rules and watch what happens. You get flooded with false positives — people asking about prices out of curiosity, not intent. You miss genuine buyers who phrase things differently: "Anyone dealt with [your competitor]?", "Thinking of making a move soon", "Need to sort this out before the end of the month."

Keywords match strings, not context. They cannot tell the difference between "Is this worth buying?" (high intent) and "I bought this last year and hated it" (zero intent). Every false positive wastes your time. Every missed signal is a lead that went to your competitor.

AI Group Monitoring replaces keyword rules with LLM-powered classification. Every message goes through an intent classifier that understands context, tone, and domain-specific language. The result: dramatically fewer false positives, dramatically more real leads.

How AI Group Analysis Works

When a message arrives in a monitored group, MoltFlow dispatches an asynchronous Celery task to classify it. The task sends the message body to your configured LLM provider (OpenAI, Anthropic, Groq, or Mistral) along with a structured classification prompt. The LLM returns a structured response with four fields:

• intent: one of 7 categories
• lead_score: 1-10 scale
• confidence: 0.0-1.0
• reason: human-readable explanation

The analysis runs asynchronously — it never blocks your webhook handler. Messages appear in your dashboard with scores attached, and a group.message.analyzed webhook event fires when analysis completes.

The whole loop — message arrives, gets classified, score appears in dashboard — typically completes in under 2 seconds on gpt-4o-mini.

Setting Up AI Monitoring

Step 1: Configure your LLM API key

Go to Settings > AI Configuration. Enter your API key from OpenAI, Anthropic, Groq, or Mistral. Select your provider and model, then click Test to verify connectivity. The key is encrypted at rest with per-tenant Fernet encryption and never leaves your account.

For most use cases, gpt-4o-mini is the right choice. It costs approximately $0.00006 per message analyzed — a group with 500 daily messages costs about $0.03/day. Anthropic's claude-haiku-3 is a solid alternative if you prefer Anthropic's models.

Step 2: Enable AI Analysis on a group

Go to Group Monitor, open a group, and set the Monitor Mode to ai_analysis. From this point forward, every inbound message in the group is sent to your LLM for classification.

# Via API: switch a group to AI analysis mode
curl -X PATCH https://apiv2.waiflow.app/api/v2/groups/monitored/{group_id} \
  -H "X-API-Key: mf_your_key" \
  -H "Content-Type: application/json" \
  -d '{"monitor_mode": "ai_analysis"}'

Step 3: Add a custom prompt (optional but recommended)

The Monitor Prompt field lets you give the AI context about your specific group. Without a prompt, the AI uses a general buying-intent classifier. With a prompt, it understands your domain and classifies far more accurately.

# Set a custom classification prompt
curl -X PATCH https://apiv2.waiflow.app/api/v2/groups/monitored/{group_id} \
  -H "X-API-Key: mf_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "monitor_prompt": "This is a real estate investment group in Tel Aviv. Focus on people asking about properties, prices, or looking to buy or rent. Messages about general neighborhood questions are off-topic."
  }'

Custom Prompts: Teaching the AI Your Context

The classification prompt is the most powerful lever you have over AI accuracy. Here are effective prompts for different industries:

Real estate:

This is a real estate investment group. Focus on people asking about properties, prices, availability, or expressing intent to buy or rent. Buyers often use indirect language — "thinking of making a move", "looking at options", "need something before Q2". Messages about neighborhood events or unrelated topics are off-topic.

E-commerce / dropshipping:

This is an e-commerce sellers community. A lead is someone asking about product sourcing, supplier recommendations, or tools to scale their store. People seeking software solutions, automation, or bulk pricing are high-intent. General discussion about marketplace policy or logistics is off-topic.

SaaS / developer tools:

This is a developer community. A lead is someone asking about WhatsApp automation, API integrations, or programmatic messaging solutions. Phrases like "need to send messages at scale", "looking for a WhatsApp API", or "anyone integrated WhatsApp with their CRM" indicate strong buying intent. General coding discussions are off-topic.

The prompt is appended to the system classification prompt. It does not override the base classifier — it adds context that helps the LLM understand your domain.

Understanding the Output

Every analyzed message gets an ai_analysis field in the database. Here is a complete example:

{
  "intent": "buying_intent",
  "lead_score": 9,
  "confidence": 0.94,
  "reason": "Sender is explicitly asking for pricing and expressing urgency about timeline, indicating high buying readiness.",
  "provider": "openai",
  "model": "gpt-4o-mini",
  "analyzed_at": "2026-02-20T14:32:11Z"
}

The 7 intent categories:

Lead score interpretation:

• 8-10: Hot lead. High buying intent, express interest or urgent need.
• 5-7: Warm lead. Research phase or soft interest. Worth monitoring.
• 1-4: Low priority. Some relevance but no clear purchase signal.
• 0: Not yet analyzed. Analysis is asynchronous; score updates within seconds.

Confidence:

Values above 0.7 are generally reliable. A confidence of 0.5 or below usually means the message is genuinely ambiguous — the AI is not sure, and you probably should not be either. Low confidence often improves with a well-crafted custom prompt.

Webhook Integration

AI Group Monitoring adds three webhook events. Subscribe to them when creating or updating a webhook:

curl -X POST https://apiv2.waiflow.app/api/v2/webhooks \
  -H "X-API-Key: mf_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "AI Lead Notifier",
    "url": "https://your-app.com/webhook",
    "events": ["lead.detected", "group.message.analyzed", "group.message"]
  }'

group.message — fires for every message in a monitored group, immediately on receipt. Payload includes sender, content preview, and group info. Use this to log all messages to your CRM.

lead.detected — fires when a lead is detected by your monitor rules. Now includes an ai_analysis field (when analysis has completed) with intent, score, and confidence.

group.message.analyzed (Pro+ only) — fires after the LLM completes analysis on any group message. This is the event to subscribe to if you want real-time AI scoring in your external system.

{
  "event": "group.message.analyzed",
  "timestamp": "2026-02-20T14:32:11Z",
  "session_id": "uuid",
  "data": {
    "message_id": "uuid",
    "wa_group_id": "[email protected]",
    "sender_phone": "+972501234567",
    "analysis": {
      "intent": "buying_intent",
      "lead_score": 9,
      "confidence": 0.94,
      "reason": "Explicit pricing inquiry with urgency"
    }
  }
}

Plan Requirements

AI Group Monitoring is available on Pro and Business plans only.

MoltFlow does not charge per AI call. You bring your own API key and pay your LLM provider directly. At gpt-4o-mini pricing, 500 calls/hour costs roughly $0.03/hour at peak load — the rate limit is there for abuse prevention, not billing.

The hourly limit resets every 60 minutes. If your group is extremely high-volume and you hit the limit, messages queued above the limit are skipped (not retried). For most monitored groups, 500 calls/hour is more than sufficient.

Stop Reading Every Group Message Manually

If you are manually scanning WhatsApp groups for leads, you are spending hours per week on a task that does not require human judgment. AI Group Monitoring does the reading, classifying, and scoring for you. You get an alert when a 9/10 lead appears. You spend your time on the conversation that closes the deal, not on sifting through noise.

Enable AI analysis on your highest-value group, add a custom prompt for your industry, and watch the signal-to-noise ratio improve. The setup takes five minutes.