Know Your Customer processes are where compliance burden meets customer experience. 80% of banks cite compliance as their top challenge, and KYC is often the worst offender: manual document review, multiple screening checks, and SAR drafting that takes 4-6 hours per case.
A single customer onboarding requires verifying identity documents (passports, driver licenses, utility bills), screening against PEP lists and sanctions databases, checking adverse media, and assessing risk. Each step is manual, siloed, and slow.
The solution uses three specialized agents orchestrated by Agent Bricks Multi-Agent Supervisor:
ai_parse_document to extract identity fields from customer documents, then ai_extract for structured data. Cross-references fields across documents for consistency.ai_classify to produce a composite risk score.ai_gen to draft Suspicious Activity Reports with required regulatory fields and evidence citations.Unity Catalog is the governance backbone: PII masking on customer data (SSN, DOB, address), column-level access controls, function registry, model governance, and the complete audit trail required for regulatory examination.
All models are accessed via AI Gateway on Databricks Model Serving. No external model access needed. AI Gateway logs every request and response for compliance via payload logging to inference tables in Delta Lake.
Batch inference via ai_query in SQL handles bulk screening for periodic customer reviews.
KYC onboarding time drops by 60%. False positives in screening decrease by 50% because Vector Search-based adverse media matching is more precise than keyword matching. SAR drafting saves 4-6 hours per case. Every decision has a complete audit trail via Unity Catalog and MLflow.