AI lead scoring workflow

1. A webhook receives the form: company, role, team size and message.
2. The Decision node asks three questions in one call: fit and intent as scores, and whether the contact is a decision maker.
3. A scorecard weights the answers into one score, and score bands fire hot, warm or cold.
4. Hot leads post to Slack with a Connector node; warm and cold leads are tagged for nurture.
5. When confidence is low, uncertain fires instead and an Agent takes a closer look.

[
  { "key": "fit", "type": "score", "instructions": "How well does this lead fit our ideal customer?",
    "criteria": ["Poor", "Partial", "Strong"] },
  { "key": "intent", "type": "score", "instructions": "How ready to buy does this lead seem?",
    "criteria": ["Browsing", "Evaluating", "Ready"] },
  { "key": "decision_maker", "type": "noul", "instructions": "Is the contact a decision maker?" }
]

[
  { "port": "cold", "min": 0 },
  { "port": "warm", "min": 0.4 },
  { "port": "hot", "min": 0.7 }
]

The Lead score preset fills all of this in. Set Minimum confidence to 0.5 so borderline leads go to review.

An agent writing "8/10" in prose drifts from run to run and costs output tokens. Jev returns probabilities for each level, so the scorecard is plain arithmetic you can audit in the run, and output tokens are free.