How to Build a Lead Scoring Model Without Code

You can build a lead scoring model with no code: upload past leads labeled converted or not, describe the goal in chat, and TensorTurn trains a probability classifier (XGBoost or scikit-learn) on a cloud GPU, then deploys a /predict API so your CRM receives a score for each new lead. With real behavioral and firmographic data, expect a ROC-AUC around 0.70–0.85; the score ranks and prioritises leads, it does not guarantee any individual outcome.

What a lead score really is

A lead score is the model's estimated probability that a lead will convert, based on how similar leads behaved in the past. You use it to prioritise: sales works the highest-probability leads first and nurtures the rest. The practical value is not perfect prediction but better ordering — putting the leads most likely to close at the top of the queue so your team spends time where it pays off.

What data you need

• One row per historical lead with a converted outcome (1/0) you already know.
• Firmographics: industry, company size, region.
• Source and channel: how the lead arrived (paid, organic, referral, event).
• Behavioral signals: page views, email opens, demo requested, trial usage, response time.
• Both converted and lost leads, ideally hundreds or more, and no leakage columns like deal-closed date or final contract value.

How to build it on TensorTurn (no code)

• Import your leads export (CSV, Excel or from a URL).
• Run the tabular health check for leakage, outliers, correlations and mixed types.
• In chat: 'Score leads by conversion probability from the converted column; return a 0–1 probability.'
• TensorTurn trains the classifier and self-heals any failing cells automatically.
• Review ROC-AUC, precision, recall, a lift/gain chart and feature importance.
• Deploy a /predict endpoint that returns a probability, and map it to A/B/C tiers for the sales team.

Read the lift chart, not just AUC

For sales prioritisation, the most useful view is a lift or gain chart: it answers 'if we only work the top 10% of scored leads, what share of all conversions do we capture?'. A model that concentrates most conversions into the top deciles is directly valuable even if its overall AUC is modest. Feature importance is worth reviewing too, because it often reveals which behaviors (like requesting a demo) genuinely predict conversion.

How accurate can it be?

With strong behavioral signal, a ROC-AUC of 0.78–0.85 is realistic. With only firmographics and no engagement data, expect 0.65–0.72. As with churn, a suspiciously high score usually means leakage — for example a field that is only populated after the deal closes — so trust the health check. And remember the score is probabilistic: a 'low' lead can still convert, so use scores to allocate effort, not to write anyone off entirely.

Deploy as an API

Publish an authenticated /predict endpoint on Modal that scales to zero when idle, with a SHA-256-hashed Bearer key and per-call logging. Using the cURL, JavaScript, Python or Rust snippets from the playground, you can score leads as they enter your CRM and route them by tier automatically.