- Westpac says AI tools cut a data warehouse migration from a $70 million budget to around $35 million and finished it more than two years early.
- Westpac's CEO says the tools and methods can be reused. Leaders should ask whether the saving repeats on the next migration.
- Most benefits are still to come. Westpac expects to book them from FY27 and plans to decommission the old warehouse in FY27.
The hard part of migration is the dull part
Old systems rarely fail in dramatic ways. They stay in place because moving them is slow, expensive and tedious. Someone has to rewrite the code, prove the new version gives the same answers, and write down what the old one did.
Westpac says AI tools sped up that work. It is worth reading closely, because these are the tasks that plausibly make migrations slow and costly. That is this article's argument, not a finding from Westpac, which reports one project and no cost breakdown by task.
What Westpac reported
Westpac told investors it used AI to migrate the legacy data warehouse of its St.George business. A data warehouse is the central store a bank uses for reporting and analysis. According to iTnews, a slide in the bank's annual results showed an initial budget of $70 million. The slide now projects a final cost of roughly half that, around $35 million.
CEO Anthony Miller said the team "used AI tools to speed up code conversion, testing and documentation". In his words, the migration was completed "more than two years earlier at half the cost". The work sits inside Unite, Westpac's codename for a large technology simplification programme.
The migrated data is headed to a platform Westpac calls Adapt. iTnews summarises it as a cloud analytics system running on Azure. Westpac's slide puts the expected direct benefit of the migration at about $15 million a year. The bank plans to decommission the old warehouse in FY27, and Miller said it expects to start booking the benefits from that year.
How the saving might work
The source names three tasks: code conversion, testing and documentation. Westpac gives no more technical detail, so the following is general context, not the bank's account.
Code conversion means rewriting the logic that loads and reshapes data so it runs on a new platform. Testing means showing the new system produces the same results as the old one. Documentation means recording what an ageing system does, often because nobody wrote it down.
All three are large, repetitive and easy to underfund. They are also work where a tool that reads and writes code could plausibly help. The lesson here is that AI may prove most useful in maintenance work that holds a modernisation back. Today that rests on one bank's self-reported, independently unverified figures about one project.
The reusable method matters more than the saving
Miller said the project produced a "reusable AI data migration capability". He said the tools, methods and experience can be applied to other migrations. That is where a one-off saving could become a pattern.
iTnews notes that banks saw something similar with cloud computing. They built and refined models for moving applications from on-premises systems to the cloud. A method that can be refined with each use, as cloud migration models were, would be worth more than one good result.
This is one bank and one project. It does not show that AI halves every migration. It shows what one large institution is willing to put in front of investors.
What the numbers do not yet tell us
Several caveats apply, and all of them are Westpac's to answer.
The saving is measured against a budget, not against a rival method. Budgets can be generous or tight. The figures are Westpac's own, and iTnews does not report independent verification.
The $15 million a year in benefits has not been booked. Westpac says that starts from FY27. The old warehouse is still due to be switched off. A migration is not finished until the old system is gone and the bills for running it stop.
Questions to put to your own team
If your organisation has a legacy migration planned, start with how the budget was built. Ask how much of it covers code conversion, testing and documentation, since that is where Westpac says AI helped.
Ask who checks the output. Faster conversion means more code to prove correct, so testing capacity matters as much as the tool. Ask how the vendor or team will report savings against the original budget, and when the old system will actually be shut off.
Then ask the question Westpac has raised for itself: does the method carry over? A saving on the first migration is a result. A saving on the second is a capability.
Produced by the WebPulse Newsroom with AI assistance from the original reporting credited below, and checked against that source by our editorial review. How we use AI.
Original reporting: iTnews.
