52 fields. 20+ screenshots. Per test case. That was QA for one of Australia's largest core banking migrations - done by hand.

‍
As part of its UNITE core banking migration, Westpac needed to verify that customer records - identity, verification, source-of-funds, compliance, preferences - had migrated accurately between systems. In practice, that meant testers manually reviewing 20+ screenshots and validating 52+ fields per test case. Slow, repetitive, and hard to scale.

‍
Working with ClearRoute, Westpac built an AI-enabled validation framework powered by Claude Sonnet 5. OCR extracts structured text from screenshots, Claude applies validation rules and compares source vs. target values, and the system generates field-level pass/fail results automatically.
‍

The result: validation time per test case dropped from 60-75 minutes to around 5 minutes - a ~92-93% reduction - while covering more fields (53 vs. 52) with greater consistency than manual review.

It's a pattern that extends well beyond this one migration: any process built on manual, screenshot-by-screenshot comparison is a strong candidate for the same approach.