Lead Intelligence Agent¶

The Lead Intelligence Agent delivers actionable insights for SDRs & AEs to close leads with better efficiency. It aggregates visitor behavior, conversation history, and intent signals into a comprehensive intelligence report.

Status¶

Beta - Available via n8n integration

Overview¶

The Lead Intelligence Agent provides detailed insights about individual visitors who have engaged with your website. It helps sales teams understand:

What pages the visitor explored
How engaged they were (session duration, page views)
What topics they discussed with the Website Agent
Their position in the conversion funnel
Their calculated intent score

Data Access¶

Supabase RPC Function¶

The agent retrieves data via the get_visitor_for_lead_intelligence RPC function.

Parameters¶

Parameter	Type	Required	Description
`p_site_domain`	TEXT	Yes	Your site domain (e.g., "demo.userose.ai")
`p_person_id`	TEXT	Yes	PostHog person_id (visitor identifier)

Example Call¶

SELECT * FROM get_visitor_for_lead_intelligence('demo.userose.ai', 'posthog-person-id-123');

n8n Integration¶

Prerequisites¶

Database credentials - Get these from your Rose admin:
PostgreSQL host: db.xxx.supabase.co
Database: postgres
User: n8n_service (dedicated read-only user)
Password: Provided by admin
Port: 5432

!!! warning "Do not use Service Role Key" The Service Role Key has full database access. Always use the dedicated n8n_service user which only has execute permission on these specific RPC functions.

n8n PostgreSQL credential - In n8n, go to Credentials → Add Credential → Postgres and enter the connection details above.

Getting the Person ID¶

The person_id is the PostHog identifier for the visitor. You can obtain it from:

PostHog webhook: When a visitor triggers an event, PostHog sends the distinct_id (same as person_id)
Rose webhook: Configure a webhook in Rose to receive visitor events with the person_id
Supabase query: Query the visitors table to find visitors by email or other criteria

Step-by-Step Workflow Setup¶

Option 1: Triggered by PostHog Event (Recommended)¶

[PostHog Webhook] → [Supabase RPC] → [Send to Slack/Email/CRM]

Add Webhook node as trigger
Copy the webhook URL and add it to PostHog as a destination
Configure PostHog to send events like lead_qualified or demo_booked
Add Postgres node
Credential: Select your n8n_service Postgres credential
Operation: Execute Query

Query:

SELECT * FROM get_visitor_for_lead_intelligence(
  '{{ $json.properties.$current_url.replace("https://", "").split("/")[0] }}',
  '{{ $json.distinct_id }}'
);

Add output node (Slack, Email, CRM, etc.)
Use the response fields to craft your message

Option 2: Scheduled Batch Processing¶

[Schedule Trigger] → [Postgres Query] → [Loop] → [Postgres RPC] → [Output]

Schedule Trigger: Run daily/hourly
Postgres node: Query visitors table for recent high-intent visitors
Loop: Iterate through each visitor
Postgres node: Call get_visitor_for_lead_intelligence for each

Postgres Node Configuration¶

-- n8n Postgres node query
SELECT * FROM get_visitor_for_lead_intelligence(
  '{{ $json.site_domain }}',
  '{{ $json.person_id }}'
);

Example: Slack Notification for High-Intent Leads¶

Complete workflow that sends a Slack message when a visitor has intent score ≥ 4:

[Webhook] → [Supabase RPC] → [IF intent_score >= 4] → [Slack Message]

Slack message template:

🔥 High-Intent Lead Alert

*Email:* {{ $json.email || 'Anonymous' }}
*Intent Score:* {{ $json.intent_score }}/5
*Funnel Stage:* {{ $json.funnel_stage || 'Browsing' }}
*Sessions:* {{ $json.total_sessions }}
*Time on Site:* {{ Math.round($json.total_session_duration_seconds / 60) }} minutes

*Topics Discussed:*
{{ $json.topics.join(', ') }}

*Pages Visited:*
{{ $json.pages_visited.map(p => '• ' + p.title).join('\n') }}

Response Fields¶

Field	Type	Description
`visitor_id`	UUID	Unique visitor identifier
`email`	TEXT	Visitor's email (if captured)
`pages_visited`	JSONB	Array of pages with URL and title (see below)
`total_page_views`	BIGINT	Total number of page views
`total_session_duration_seconds`	BIGINT	Total time spent on site (seconds)
`total_sessions`	BIGINT	Number of separate browsing sessions
`funnel_stage`	TEXT	Current conversion stage (see below)
`conversations`	JSONB	Full conversation transcripts (see below)
`topics`	TEXT[]	Topics discussed across all conversations
`keywords`	TEXT[]	Keywords extracted from conversations
`last_activity_at`	TIMESTAMPTZ	Most recent activity timestamp
`total_messages`	BIGINT	Total message count across conversations
`intent_score`	INTEGER	Calculated intent (1-5 scale)

Funnel Stages¶

Stage	Description	Trigger
`Conversion completed`	Visitor booked a demo	`demo_booked = true`
`Conversion initiated`	Visitor clicked CTA	`cta_clicked = true`
`Conversion offered`	Demo was proposed to visitor	`demo_proposed = true`
`null`	Not yet in conversion funnel	No conversion signals

Intent Score¶

The intent score (1-5) is calculated using:

60% Behavior signals: Based on funnel progression
5 = Demo booked
4 = CTA clicked
3 = Demo proposed
2 = 2+ message turns
1 = Anonymous/minimal engagement
40% Interest signals: Based on conversation engagement (lifetime_interest_score)

Pages Visited JSONB Structure¶

[
  {
    "url": "/pricing",
    "title": "Pricing - Rose AI"
  },
  {
    "url": "/features/chatbot",
    "title": "AI Chatbot Features"
  }
]

Conversations JSONB Structure¶

[
  {
    "id": "conversation-uuid",
    "started_at": "2024-01-15T10:30:00Z",
    "topics": ["pricing", "enterprise"],
    "keywords": ["API", "integration"],
    "messages": [
      {
        "role": "user",
        "content": "What's your enterprise pricing?",
        "created_at": "2024-01-15T10:30:00Z"
      },
      {
        "role": "assistant",
        "content": "Our enterprise plans start at...",
        "created_at": "2024-01-15T10:30:05Z"
      }
    ]
  }
]

Security¶

Webhook Authentication Required

If your n8n workflow is triggered by a webhook, you MUST add authentication to prevent unauthorized access. Without authentication, anyone who discovers your webhook URL can trigger the workflow and access your data.

Options:

Header Auth: Require a secret token in Authorization header
Query Param Auth: Require a secret in the URL (less secure)
IP Allowlist: Restrict to known IPs (e.g., PostHog servers)

Read-Only Access¶

The RPC functions are read-only (STABLE in PostgreSQL terms). They cannot modify any data.

Tenant Isolation¶

The p_site_domain parameter ensures data isolation:

Each call is scoped to a specific site domain
Users can only access data for sites they own
No cross-tenant data access is possible

Supabase Role Configuration¶

We use a dedicated n8n_readonly role with minimal permissions. This role can only execute the lead intelligence RPC functions—no direct table access.

1. Create the Role (run once in Supabase SQL Editor)¶

-- Create a dedicated role for n8n integrations
CREATE ROLE n8n_readonly NOLOGIN;

-- Grant execute on lead intelligence functions only
GRANT EXECUTE ON FUNCTION get_visitor_for_lead_intelligence(TEXT, TEXT) TO n8n_readonly;
GRANT EXECUTE ON FUNCTION get_account_for_outreach_agent(TEXT, TEXT) TO n8n_readonly;

-- Create a service user that uses this role
CREATE USER n8n_service WITH PASSWORD 'your-secure-password-here';
GRANT n8n_readonly TO n8n_service;

2. Generate a JWT for n8n¶

Use the Supabase dashboard or API to generate a JWT for the n8n_service user, or use the service with direct PostgreSQL connection.

3. Configure n8n¶

In n8n, you can either:

Supabase node: Use the anon key (functions are granted to anon)
PostgreSQL node: Connect directly with n8n_service credentials (more secure, no JWT needed)

PostgreSQL node configuration:

Host: db.xxx.supabase.co
Database: postgres
User: n8n_service
Password: your-secure-password-here
Port: 5432
SSL: Require

Recommended Setup¶

Method	Use Case	Risk Level
PostgreSQL with n8n_service	n8n integrations (recommended)	Low - Only RPC execute permission
Supabase with Anon Key	Quick setup, less secure	Low-Medium
Service Role	Never use externally	High - Full database access

Use Cases¶

Pre-call research: Review visitor's conversation history before a sales call
Lead prioritization: Use intent score to prioritize outreach
Personalized follow-up: Reference specific topics and questions in outreach
Handoff notes: Provide context when passing leads between team members

Website Agent - The agent that captures visitor conversations
Outreach Agent - Uses this data for personalized outreach
Data Sources Reference - Technical data architecture