Sales had more leads than attention.
A B2B SaaS client had steady inbound demand, but every new contact looked equally urgent in the CRM. Account executives relied on recency, gut feel, and manual research to decide who deserved follow-up first.
Case Study
Predictive Lead Scoring Engine
Representative B2B SaaS workflow for ranking leads by purchase intent, routing the best-fit accounts to sales, and reducing wasted outreach.
Metrics shown are representative placeholders and should be refined before presenting as verified client outcomes.
+34%Qualified-lead conversion
-28%Sales time on dead leads
0.89Model AUC
High-intent buyers were buried in noise.
Form fills, product-page visits, firmographic fit, lifecycle stage, and email engagement lived in separate tables. Without a unified score, sales spent too much time chasing low-fit leads while better opportunities aged.
Built a CRM-native scoring model from the full lead journey.
Data sources
Joined CRM lifecycle events, web behavioral signals, and firmographic enrichment into a weekly training set.
Feature engineering
Created recency, frequency, fit, engagement-depth, source-quality, and sales-touch features from SQL transformations.
Model
Trained gradient-boosted trees with calibrated probabilities and interpretable feature-importance checks.
Validation
Evaluated AUC, precision by score band, lift over random assignment, and holdout-period stability.
Deployment
Served a FastAPI scoring endpoint that writes lead scores and routing tiers back into HubSpot.
Result
Sales could prioritize the accounts most likely to convert.
The representative model separated high-intent leads from low-fit contacts, producing a clearer routing queue, fewer wasted touches, and a stronger conversion signal in the top score bands.
Lift by Score DecileRepresentative placeholder