Skip to main content

AI lead qualification

This workflow scores a lead as soon as it is submitted via the form, records qualified leads in Snowflake and notifies the salesperson by email.

Loading the diagram…

The need​

Every lead submitted via the form must be qualified to know whether it deserves the salesperson's attention. An AI agent scores it automatically: qualified leads are recorded in Snowflake and the salesperson is notified by email, the others are published as unqualified on a Google Pub/Sub topic consumed by the CRM.

What the workflow does​

  1. New lead (form) — the Trigger node captures the form submission (triggerId form.submitted). The lead data arrives in data.
  2. Score the lead (AI) — the AI Agent node evaluates the lead with the gemini-1.5-pro model (structured output). The score is stored in context.agentOutput.answer.score.
  3. Score ≥ 60? — the Condition node routes the lead according to its score.
  4. Record the qualified lead — if the score is high enough, the Snowflake node inserts the lead in the leads_qualifies table.
  5. Notify the salesperson — the Send an email node sends the result to the sales team.
  6. Publish the unqualified lead (CRM) — otherwise, the Send Google Pub/Sub node publishes lead_rejected on the CRM's Pub/Sub topic.

What you need​

Construction​

Copy the workflow definition and paste it into the editor (the Trigger node is recreated automatically, it is not in the snippet):

[
  {
    "id": "agent-103",
    "name": "Score the lead (AI)",
    "config": {
      "type": "AGENT",
      "model": {
        "provider": "GEMINI",
        "name": "gemini-1.5-pro"
      },
      "systemPrompt": "You score B2B leads from 0 to 100, with a score field and a summary.",
      "message": "Here are the lead details: {{ data.company }}",
      "outputKey": "agentOutput",
      "responseFormat": "{\"type\": \"object\", \"properties\": {\"score\": {\"type\": \"number\", \"minimum\": 0, \"maximum\": 100}, \"synthesis\": {\"type\": \"string\"}}, \"required\": [\"score\", \"synthesis\"]}",
      "tools": []
    },
    "parents": [
      "trigger"
    ],
    "position": {
      "x": 0,
      "y": 200
    },
    "next": "cond-103"
  },
  {
    "id": "cond-103",
    "name": "Score ≥ 60?",
    "config": {
      "type": "CONDITION",
      "conditionCode": "return context.agentOutput.answer.score >= 60;",
      "conditions": [
        {
          "value": true,
          "next": "db-103"
        },
        {
          "value": false,
          "next": "event-103"
        }
      ]
    },
    "parents": [
      "agent-103"
    ],
    "position": {
      "x": 0,
      "y": 400
    }
  },
  {
    "id": "db-103",
    "name": "Record the qualified lead",
    "config": {
      "type": "DATABASE_SNOWFLAKE",
      "query": "INSERT INTO leads_qualifies (company, score) SELECT '{{ data.company }}', {{ context.agentOutput.answer.score }}",
      "outputKey": "snowflakeResult"
    },
    "parents": [
      "cond-103"
    ],
    "position": {
      "x": 0,
      "y": 600
    },
    "next": "email-103"
  },
  {
    "id": "email-103",
    "name": "Notify the salesperson",
    "config": {
      "type": "SEND_EMAIL",
      "to": [
        "ventes@agence-leads.fr"
      ],
      "subject": "New scored lead: {{ data.company }}",
      "content": "AI score: {{ context.agentOutput.answer.score }}",
      "sendAsHtml": true
    },
    "parents": [
      "db-103"
    ],
    "position": {
      "x": 0,
      "y": 800
    }
  },
  {
    "id": "event-103",
    "name": "Publish the unqualified lead (CRM)",
    "config": {
      "type": "SEND_EVENT",
      "provider": {
        "type": "GCP_PUBSUB",
        "projectId": "mon-projet-crm",
        "topicId": "leads"
      },
      "body": "{\"event\": \"lead_rejected\", \"company\": \"{{ data.company }}\"}"
    },
    "parents": [
      "cond-103"
    ],
    "position": {
      "x": 800,
      "y": 600
    }
  }
]

Check that it works​

  • Expected data: the agent message references {{ data.company }} — make sure the trigger payload actually exposes this data. The score is read from the agent's output key (default context.agentOutput); the condition and the email use context.agentOutput.answer.score.
  • Deploy the workflow (Deploy a workflow), then trigger the real scenario: every execution generates a trace you can view (View traces).

Variations​

  • If you change the agent's output key, update all the references that use it (condition and email).
  • The structured output forces the agent to return a reliable JSON object (score, summary) usable by the condition; without it, the agent's response is plain text.
  • Without a configured tool, the AI Agent node behaves in conversational mode: it answers but does not trigger external calls.

Going further

See also: Examples and integrations