Smart Conversations triggers deliver AI analysis results and redacted message content when Sinch's Smart Conversations feature is enabled. Smart Conversations uses machine learning to analyze message content for sentiment, intent, PII, offensive content, and more. These triggers require Smart Conversations to be enabled on your Conversation API app.

The two Smart Conversations triggers are:

SMART_CONVERSATION — AI analysis results for messages
MESSAGE_INBOUND_SMART_CONVERSATION_REDACTION — Redacted message content (PII removed)

Prerequisites

Smart Conversations must be enabled on your Conversation API app. Contact Sinch support to enable this feature, as it's not available by default.

SMART_CONVERSATION

Overview

Delivers AI-powered analysis of message content, including sentiment, intent classification, PII detection, offensive language detection, and custom entity extraction. Analysis is performed on all inbound messages when Smart Conversations is enabled.

When It Fires

Inbound message received from contact (after MESSAGE_INBOUND)
Analysis completed by Smart Conversations engine (typically 100-500ms after message received)
Only fires for messages where analysis completes successfully

Callback Structure

{
  "app_id": "01H1234567...",
  "accepted_time": "2024-06-15T14:25:00.456Z",
  "project_id": "PROJECT123",
  "smart_conversation_notification": {
    "message_id": "01H9876543...",
    "conversation_id": "01H5555555...",
    "contact_id": "01H3333333...",
    "analysis": {
      "sentiment": {
        "score": -0.65,
        "label": "NEGATIVE",
        "confidence": 0.89
      },
      "intent": {
        "primary": "order_status_inquiry",
        "confidence": 0.92,
        "secondary": ["refund_request"]
      },
      "pii_detected": true,
      "pii_entities": [
        {
          "type": "EMAIL",
          "text": "john.doe@example.com",
          "start": 45,
          "end": 66
        },
        {
          "type": "PHONE_NUMBER",
          "text": "555-1234",
          "start": 85,
          "end": 93
        }
      ],
      "offensive_content": {
        "detected": false,
        "confidence": 0.95
      },
      "language": "en",
      "categories": ["customer_support", "order_inquiry"]
    }
  }
}

Analysis Fields

Field	Description
`sentiment.score`	Sentiment score from -1.0 (very negative) to +1.0 (very positive)
`sentiment.label`	`POSITIVE`, `NEGATIVE`, or `NEUTRAL`
`sentiment.confidence`	Confidence score (0.0 to 1.0)
`intent.primary`	Primary intent classification
`intent.confidence`	Confidence in primary intent
`intent.secondary`	Array of possible secondary intents
`pii_detected`	Boolean indicating if PII was found
`pii_entities`	Array of detected PII items with type and location
`offensive_content.detected`	Boolean indicating if offensive content detected
`language`	Detected language code (ISO 639-1)
`categories`	Array of message category classifications

PII Entity Types

EMAIL — Email addresses
PHONE_NUMBER — Phone numbers
CREDIT_CARD — Credit card numbers
SSN — Social Security Numbers
PASSPORT — Passport numbers
DRIVER_LICENSE — Driver's license numbers
PERSON_NAME — Person names
ADDRESS — Physical addresses
DATE_OF_BIRTH — Dates of birth
BANK_ACCOUNT — Bank account numbers

Common Use Cases

Content Moderation — Detect and filter offensive language
Sentiment Routing — Route negative sentiment messages to senior agents
Intent Routing — Auto-route messages to departments based on intent
PII Compliance — Detect sensitive information for GDPR/HIPAA compliance
Analytics — Aggregate sentiment and intent data for business insights

MESSAGE_INBOUND_SMART_CONVERSATION_REDACTION

Overview

Delivers a redacted version of inbound messages with PII automatically removed or masked. This trigger fires in addition to MESSAGE_INBOUND and provides a sanitized message that's safe to store or process without PII compliance concerns.

When It Fires

Inbound message with PII detected (after PII detection analysis completes)
Only fires if PII was found in the message
Fires shortly after MESSAGE_INBOUND (typically 100-500ms later)

Callback Structure

{
  "app_id": "01H1234567...",
  "accepted_time": "2024-06-15T14:25:30.789Z",
  "project_id": "PROJECT123",
  "message": {
    "id": "01H9876543...",
    "conversation_id": "01H5555555...",
    "contact_id": "01H3333333...",
    "channel": "WHATSAPP",
    "direction": "TO_APP",
    "contact_message": {
      "text_message": {
        "text": "Hi, my email is [EMAIL_REDACTED] and my phone is [PHONE_REDACTED]. Can you help with order #12345?"
      }
    },
    "channel_identity": {
      "channel": "WHATSAPP",
      "identity": "46732001122",
      "app_id": "01H1234567..."
    },
    "accept_time": "2024-06-15T14:25:30.789Z",
    "metadata": "{\"redacted\":true,\"original_message_id\":\"01H9876543...\"}"
  }
}

Redaction Format

PII is replaced with standardized tokens:

[EMAIL_REDACTED] — Email addresses
[PHONE_REDACTED] — Phone numbers
[CREDIT_CARD_REDACTED] — Credit card numbers
[SSN_REDACTED] — Social Security Numbers
[NAME_REDACTED] — Person names
[ADDRESS_REDACTED] — Physical addresses
[DOB_REDACTED] — Dates of birth
[BANK_ACCOUNT_REDACTED] — Bank account numbers

Common Use Cases

GDPR Compliance — Store redacted messages instead of originals to minimize PII exposure
Training Data — Use redacted messages for ML training without privacy concerns
Audit Logs — Log redacted messages for compliance audits
Third-Party Integration — Share redacted messages with external systems safely

Key Points

SMART_CONVERSATION

Feature Flag Required — Smart Conversations must be enabled on your app; contact Sinch support
Async Analysis — Fires 100-500ms after MESSAGE_INBOUND; don't rely on it for immediate responses
Confidence Scores — Always check confidence scores before acting on analysis results
Language Support — Analysis quality varies by language; English has best support
Cost — Smart Conversations may incur additional charges; check with Sinch
Sentiment Scale — Ranges from -1.0 (very negative) to +1.0 (very positive); 0.0 is neutral

MESSAGE_INBOUND_SMART_CONVERSATION_REDACTION

Conditional Firing — Only fires if PII was detected in the message
Original Still Delivered — MESSAGE_INBOUND delivers original message; this trigger delivers redacted version
Storage Strategy — Consider storing only redacted versions for compliance
Irreversible — Redaction is one-way; PII cannot be recovered from redacted messages
Token Format — Use standardized tokens for easy identification of redacted content

General

Combined Use — Use both triggers together: analysis for routing/escalation, redaction for storage
Real-Time Actions — Use SMART_CONVERSATION for real-time routing decisions
Long-Term Storage — Use MESSAGE_INBOUND_SMART_CONVERSATION_REDACTION for compliant data retention
False Positives — PII detection may have false positives; review periodically
Custom Intents — Intent classifications can be customized; work with Sinch to train custom models
Privacy by Design — Redaction happens automatically; no PII ever leaves Sinch unprotected when using redacted messages