Keynote Session by Commbank with Sonal Surana
- Predictive AI
- H2O Document AI
Overview
Sonal Surana of Commonwealth Bank of Australia (CBA) describes how the bank scaled H2O's Document AI into a production capability processing millions of documents a day, four months after early access. The talk walks through concrete use cases: extracting data from payslips and bank statements for credit assessment, reading and verifying customer ID documents like passports and driver's licenses for KYC, and automating invoice processing for accounts payable. It closes with a live demo of signature detection and forgery comparison, run against a CBA colleague's own signature as a test case.
Key takeaways
- How Commonwealth Bank of Australia scaled H2O's Document AI into production in four months, now processing millions of documents a day
- Where Document AI is applied across the customer lifecycle: account origination, KYC, credit assessment, servicing requests, and non-customer documents like supplier invoices
- How payslip and bank statement data gets extracted for credit assessment despite wide variation in document formats and templates
- How KYC document processing combines classification, noise classification, and noise reduction models ahead of OCR and NLP, reaching 50 to 85 percent accuracy across ID document types
- A live demo of signature detection and forgery comparison run against a colleague's own signature as a test case
Start owning your AI