Explainable lead scoring

Call the right leads first.

Signal trains a lead-scoring model right in your browser, shows how well it separates buyers from browsers, and gives reps a ranked call list with the reasons behind every score.

Upload leads CSV 1,500 synthetic leads · trained live in your browser

Try it: untick Demo requested and Source to see how much accuracy those signals carry, then move the call-list slider.

Conversions captured

–

Lift vs. random

–

AUC (test set)

–

Base conversion rate

–

Cumulative gainsShare of conversions found vs. share of leads called

SignalRandom order

ROC curveTrue vs. false positive rate

SignalCoin flip

What drives conversionStandardized model weights — right raises the score, left lowers it

Call list

#	Lead	Score	Outcome	Why

The model

Features. Text columns become one-hot indicators; numeric columns are standardized so their weights are comparable. ID-like columns are ignored.

Training. Logistic regression fitted by batch gradient descent with L2 regularization, on a random training split. Every chart reports on leads the model never saw.

Evaluation. AUC measures how often a converter outranks a non-converter. Cumulative gains shows the business version: how many conversions you find if reps only call the top slice.

Explanations. Each lead’s reasons are the features contributing most to its score (weight × value), which is exact for a linear model.

Upload any CSV with a 0/1 converted column. Sample columns: lead_id, source, industry, company_size, title_level, pages_viewed, email_opens, demo_requested, days_since_last_touch, converted. Sample data is synthetic.