What ANZ moved, and how quickly
ANZ Group has moved its on-premises Enterprise Big Data platform to Amazon EMR on AWS, according to an AWS announcement dated 17 November 2026. The platform supports payments, business banking and customer remediation. The bank says the move covered 1.7 petabytes of data held in 300 million files, along with 142,000 tables, 4,500 automated data workflows and 4,000 views. ANZ had expected the work to take 18 months, and it finished in six.
iTnews reports that the source system was Cloudera Hadoop and that ANZ mostly used lift-and-shift, moving workloads with little redesign. Kai Yang, ANZ's chief data and AI officer, told an AWS media briefing about a "hothouse environment" where ANZ business and technology staff sat in the same room as AWS engineers. AWS Professional Services and AWS partners that haven't been named worked with ANZ's own engineers.
Where AI carried the load
ANZ used coding agents and built troubleshooting agents to cut down defects. Its engineers also used an assistant built on Amazon Bedrock to find and fix performance bottlenecks, iTnews says. The bank estimates agentic AI reduced migration effort by about 35% and roughly halved defects and rework compared with traditional methods. It hasn't named the models it used. It said only that they were internally approved.
ANZ also tokenised sensitive customer data before moving it, swapping real values for stand-in tokens. About 80TB has been tokenised so far. The AWS release describes operating costs falling from more than US$10,000 a day to under US$3,000 a month. ANZ told iTnews that those figures apply solely to the tokenisation capability and don't describe the cost of running the wider platform. Computer Weekly gave the same numbers in Australian dollars and presented them as platform operating costs. Anyone repeating the cost saving internally should use the narrower, tokenisation-only version.
How much weight the numbers can bear
ANZ and AWS supplied every figure in the coverage, and AWS has a clear commercial interest in how the project looks. The 18-month baseline is ANZ's own forecast. The claim that tokenisation runs about seven times faster compares against an earlier approach that ANZ hasn't described. IT Brief points out that because the old cost is daily and the new one is monthly, you can't work out a clean percentage saving. ANZ hasn't published the project's total cost or how many people worked on it. Without those numbers, outsiders can't tell how much of the speed came from AI tooling and how much from a large, co-located team with direct access to AWS engineers.
The results are still useful, because ANZ's estate was large, regulated and in daily use. In Computer Weekly's coverage, Chris Casey, AWS's managing director for Australia and New Zealand, made the obvious point that migrating a banking data platform while it keeps serving customers is hard. ANZ says the migration supports its 2030 Strategy, which aims to simplify the bank's technology estate and prepare its data for AI at scale.
What NZ businesses can take from it
Few NZ organisations run petabyte Hadoop clusters. Plenty do run an ageing SQL Server warehouse, a reporting server under someone's desk, or hundreds of scheduled jobs and stored procedures written by staff who've since left. The ANZ project holds lessons for those smaller estates.
The most transferable lesson is ANZ's precise inventory. The bank could state exactly how many tables, workflows and views it had before it started. Coding agents need that kind of scope to convert and test against. A mid-sized firm that can't say how many overnight jobs feed its finance reports will find AI-assisted migration produces code nobody can verify. Counting assets and naming an owner for each costs very little compared with a failed cutover.
ANZ's choice of lift-and-shift deserves attention too. Teams often try to redesign a data platform during a migration, and the project grows until it stalls. ANZ moved the workloads largely as they were, used AI to handle conversion and testing, and is now talking about a lakehouse design as a later step, according to Yang. Splitting the move from the redesign gives each phase its own budget and its own test of success.
Tokenising data before migration matters to any NZ firm sending customer records to an Australian or offshore cloud region. Tokenised values reveal little if a storage bucket is misconfigured during the messy middle of a migration, when copies of data tend to be scattered across temporary environments. The work needs a data classification first, and many firms don't have one.
ANZ's limit on models is worth copying as well. Engineers used only internally approved models, and iTnews notes the bank unveiled an enterprise AI control plane the week before for monitoring and auditing its AI systems. Smaller firms won't build a control plane, but they can decide which AI coding tools are allowed to read production schemas and sample data before a vendor's engineers start pasting code into a chatbot. Our AI consultancy work begins with that kind of usage policy.
The partner model also stands out. ANZ chief executive Nuno Matos said last year the bank would lean less on third parties and bring them in for special cases such as speeding up delivery, iTnews reports. The migration followed that pattern: ANZ engineers worked alongside AWS staff, and ANZ kept the knowledge of its own platform. Firms that hand a whole migration to a vendor often end up unable to support what's delivered.
Practical next steps
- Build an inventory of your data estate covering databases, tables, scheduled jobs, reports, integrations and a named owner for each, before you price any migration.
- Pick a small, contained slice of the estate and measure how long conversion and testing take by hand, so any claim about AI-assisted savings has a baseline from your own systems.
- Classify sensitive fields and tokenise or mask them before data leaves your environment, including copies made for testing.
- Write down which AI tools may be used for code conversion, what data they can see, and who reviews their output before it reaches production.
- Ask vendors for effort and defect figures measured on estates like yours, and keep one-off migration costs separate from ongoing running costs in the business case.
- Treat redesign as a second phase with its own budget, and set a date to switch off the old system so you don't pay to run two platforms.
If you're planning a move off an ageing data platform and want help scoping it, our digital transformation team can help you size the estate and stage the work.
Sources
- ANZ completes major cloud data transformation, strengthening data foundations to improve customer experiences (AWS Press Center)
- ANZ uses AI to move 1.7PB data platform to AWS (iTnews)
- ANZ migrates 1.7 petabytes of data to Amazon Web Services (Computer Weekly)
- ANZ migrates major data platform to AWS in six months (IT Brief NZ)
- ANZ 2030 Strategy (ANZ Newsroom, linked from Computer Weekly)
Want to know what this means for your organisation? Our team in Auckland can show you your top three opportunities in a free 30 minute assessment.
Get my free assessment How we help