Data
Build a lead scoring model
Scores fitted to what actually converted, and validated.
Fill it in
With their outcomes.
A model scored on its training data is a description.
A score nobody understands is a score nobody uses.
Your prompt
Build a lead scoring model. Fit against what CONVERTED, not against what sales liked. Most scoring models encode an existing bias and then justify it, because the training signal was "marked qualified" rather than "closed". Hold out a time-based slice and validate on it. Scoring on the data you fitted is a description of the past, and a random split leaks the future into the training set. Check for leakage. A feature like "requested a demo" predicts conversion perfectly and is useless, because by then nobody needs a score. Make it explainable: which factors moved this score and by how much. A black-box score is ignored by the team it was built for.
Use Build a lead scoring modelOpens with everything above already filled in.
Why this works
Most scoring models encode an existing bias and then justify it, because they were trained on "marked qualified" rather than "closed". This fits to outcomes and validates on a holdout.