ANZ credits AI with shortening major AWS data migration
The bank says it moved 1.7 petabytes of data in six months, against an expected 18 months, while reducing engineering effort and rework.
By Quang Dinh

ANZ says AI-assisted engineering helped it complete the transfer of its enterprise big data platform to Amazon Web Services in six months, rather than the 18 months it had expected.
The bank shifted 1.7 petabytes from an on-premises Cloudera Hadoop system to Amazon EMR. The migration encompassed 142,000 tables and 300 million files, along with 4,500 automated workflows and 4,000 views. ANZ said workloads were largely transferred with minimal redesign.
ANZ estimated that agentic AI lowered the migration effort by 35 per cent and roughly halved defects and rework against traditional methods. Chief data and AI officer Kai Yang said business and technology employees worked together with AWS engineers, using coding agents and developing agents for troubleshooting.
An assistant built using Amazon Bedrock helped engineers identify and resolve performance bottlenecks. ANZ did not identify the underlying models, saying only that they had internal approval. AWS Professional Services and unnamed AWS partners also worked with the bank's engineers.
The platform serves ANZ's payments, customer remediation, business banking and private banking operations. To protect sensitive customer information before transferring it, the bank replaced sensitive values with tokens. Almost 80 terabytes has undergone that process.
ANZ said tokenisation ran about seven times faster than its initial approach. Daily operating costs exceeding US$10,000 fell to a monthly figure below US$3,000. Those figures apply only to tokenisation, not the entire platform; the bank did not describe the earlier method.
Yang said ANZ was also developing a reusable AI platform designed to operate both on-premises and through cloud services including AWS.



