Why intent without risk is incomplete
A high-intent consumer who will not pass underwriting consumes funnel capacity and degrades top-of-funnel CAC. A low-intent consumer who would pass underwriting is invisible to traditional lead products. The combined score concentrates spend on the cohort that both intends to convert and is likely to be profitable.
Signal sources
Behavioral risk signals (where consented) include: spending pattern indicators, account-balance volatility, employment-stability indicators, and credit-utilization patterns. These are merged with behavioral intent signals at score time. Decisioning use cases — actual underwriting — are handled with the appropriate consumer reporting frameworks via approved partners.
Tier-aware modeling
Prime, near-prime, and subprime cohorts have substantially different behavioral profiles and different decay curves. A single blended model under-serves all three. Tier-segmented models are the standard.
Signal half-life — production model
Predictive cohort vs. cold list
Citations
- · Hand, D. J. — Measuring classifier performance. Machine Learning, 2009.
- · Crook, J., et al. — Recent developments in consumer credit risk assessment. EJOR, 2007.