Notes on scoring, features, and why reps stop trusting their pipeline.
Short essays on predictive scoring, CRM data quality, SHAP explainability, and what we're learning building ax1om.
Clay's AI scoring is still rules-based. Here's what's missing.
Prompt-based lead scoring in Clay workflows is still rules-based, just written in natural language. Here's how a predictive API endpoint upgrades the scoring step from subjective to statistical.
Read postThe 7 CRM fields that actually predict B2B lead conversion
Feature importance analysis on 16,600 real B2B leads reveals which CRM fields drive conversion predictions. Spoiler: it's not page views or email opens.
Read postFeature Stability Score: the metric your model's explainability is missing
Accuracy tells you the model works. SHAP tells you why. Feature Stability Score tells you whether that why will still hold next quarter. A 2026 paper introduced the formula. Here's why it matters for GTM scoring.
Read postHow a B2B software company used ML scoring to fix its pipeline
A B2B software company replaced its traditional lead scoring with a Gradient Boosting model trained on 4 years of CRM data. 98.39% accuracy. Here's what happened.
Read postWhy rules-based lead scoring fails, and what to replace it with
Rules-based scoring is an opinion in a spreadsheet. ML scoring is what your data actually shows. A case study from a B2B software company with 16,600 lead records explains why.
Read post