Models / group-data-ai

group-data-ai/lead-priority-scorer

Lead Priority Scorer

ConfidentialLimited riskGroup ReuseScoringLightGBMN/Av2.3
186 212K/30d 142

AUC (out-of-time)

0.812

Conversion lift, top decile

3.4x

Precision@P1

0.63

Calibration error (ECE)

0.021

Median latency

42 ms

About this model

Scores every open sales lead in the CRM for its probability of converting within 30 days and sorts it into a priority band, P1 (call today) to P4 (nurture). Each score comes with its top drivers and a suggested next action, so relationship managers and branch staff work the leads most likely to close first. The model behind the optimised lead tracking use case in the Enterprise Data Platform programme.

Intended use

Prioritising and routing Consumer Banking and Business Banking sales leads from branches, RM referrals, the contact centre and digital forms, and suggesting the next follow-up. Leads without marketing consent are excluded. It ranks leads only; it never makes a credit or pricing decision.

Training data lineage

Trained on crm-leads-history (Oct 2023 to Jun 2026, 1.4 mil leads with converted or lost outcomes) joined to customer-360-mart relationship features and campaign-history contact outcomes, all read from the lakehouse gold layer in BigQuery. Trained and tracked in Vertex AI Pipelines and Experiments, registered in the Vertex AI Model Registry, and reviewed by Model Risk Management as a Tier 3 model.

Limitations & bias notes

Leads from brand-new channels (for example the 2026 partner referral scheme) have little history and score close to the base rate until about 3 months of outcomes accumulate. The model learns from past RM follow-up behaviour, so leads that were historically under-worked can be under-scored; a random 5% holdout of P3 and P4 leads is followed up to measure this. Scores are refreshed hourly, so a lead updated minutes ago may show a stale band.

Evaluation metrics

AUC (out-of-time)0.812
Conversion lift, top decile3.4x
Precision@P10.63
Calibration error (ECE)0.021
Median latency42 ms
#lead-scoring#crm#consumer-banking#next-best-action#lightgbm#flagship#edp-use-case

Try it

Live sandbox

L-2026-08814

Synthetic
Source
Branch walk-in · Bangsar
Interest
Home financing
Segment
Mass affluent, existing CASA
Relationship
Salary credited to the Bank for 4 years
Enquiry age
1 day

Used in workflows

Retiring this asset would require these chains to be re-pointed first.

Feedback

5.0(1)

Owning business unit

DAGroup Data & AI
Updated
2026-07-22
Latest version
v2.3
Licence
Group Reuse
Access
Open to all BUs