Data Warehousing for Lending: Origination, Portfolio, and HMDA in One Curated Layer

A lender builds a data warehouse and the compliance team doesn't trust it. The loan count differs from the LOS because application classification differs (which applications count as denied vs withdrawn vs approved-not-accepted). The HMDA view uses different code values than what was submitted. The vintage delinquency numbers don't reconcile to the servicing system because performance definitions differ. Each difference is small. Together they make the warehouse a parallel reality the compliance, credit risk, and finance teams treat with suspicion. By month four, the HMDA submission team is back to pulling data directly from the LOS, and the ALLL committee is back to pulling from servicing.

Lending warehousing done right encodes the LOS semantics and servicing definitions correctly from day one. Application and loan classification matching the LOS rules (denied, withdrawn, approved-not-accepted). HMDA LAR field values matching what gets submitted. Vintage and delinquency definitions matching the servicing system. Loan-level economics with consistent cost allocation methodology. Credit bureau data with stable tradeline linkage. Point-in-time correctness for underwriting model backtesting. Reconciliation against the LOS and servicing system after every load. Done with this discipline, the warehouse becomes the trusted source for compliance, credit risk, finance, and operations. Done generically, it stays parallel.