Data Engineering for Lending: LOS, Servicing, Bureau, and Compliance Pipelines

Lending data engineering navigates source systems and data structures most enterprise data engineers haven't seen. The LOS (Encompass, MeridianLink, Blend, nCino) holds application and loan data in models with hundreds of fields, many of which trigger HMDA reporting obligations with specific code values the CFPB defines. The servicing system (Black Knight MSP, Sagent, FICS) holds loan performance data with transaction-level detail and the interest accrual patterns GAAP reporting requires. Credit bureau data arrives as reports with tradeline-level detail that needs parsing and stable linkage. Plaid and alternative data provide real-time bank account information with schema drift. Appraisal data comes through UCDP. Each source has its own conventions, update frequencies, and data quality patterns. Generic enterprise data engineering doesn't address any of this.

Lending data engineering that works follows industry-specific patterns. LOS integration with the HMDA-aware field mapping that makes the downstream compliance data usable. Servicing data ingestion with loan-level detail that supports vintage analysis and CECL modeling. Credit bureau parsing with stable tradeline linkage across pulls. Plaid and alt data ingestion with schema drift handling. Loan-level economic data (cost allocation, capital charge) joined for profitability analytics. Point-in-time correctness so backtests of underwriting models don't suffer lookahead bias. Bronze-silver-gold medallion with gold models aligned to HMDA LAR, CECL inputs, and operational reporting. SOC 2 and state examination-aligned audit logging. Done with this discipline, the data platform supports underwriting, servicing, and compliance. Done generically, it produces a system nobody trusts for regulatory work.