DBS reported about S$1 billion in economic value from data analytics and AI in 2025, generated by more than 2,000 models across more than 430 use cases. The series began in the 2022 annual report as about S$150 million of revenue uplift, which the following year was restated as S$180 million of economic value. It then grew to S$370 million in 2023 and S$750 million in 2024. At 2026 exchange rates the latest figure is roughly US$770 million, not the billion dollars it is often quoted as.
Economic value, as DBS counts it, has three components: revenue uplift, cost avoidance and productivity, and losses averted from fraud and scams. Only one of those is a saving in the sense a CFO would use the word. The measurement method is the strong part. Nimish Panchmatia, DBS's Chief Data and Transformation Officer, has said a large part of it is test and control: each use case is compared against a control group, the difference is counted as AI value, and a business owner agrees the outcome before the model goes live.
What the figure leaves out matters as much. It sits outside the audited financial statements. No per-use-case breakdown is published. DBS has not said whether the number is net of build and run costs. And the target moved: the original S$1 billion goal set in 2022 was a five-year revenue aspiration, and by 2024 it had become a one-year economic value target on a broader definition. DBS disclosed the change, which is to its credit, but numbers are only comparable across years when their definitions match. DBS was one of only two banks, with JPMorgan Chase, to publish a total realised AI value in the 2024 Evident AI Index.
A board that wants its own version of this number should ask six questions before anyone quotes it. Is there a written definition of which components count, agreed before the first figure? Does each use case have a control group, or a documented baseline where a control is impossible? Which accountable business lead signs off each value claim? Is value reported net of data, compute, licence and people costs? Have changes to the definition or prior-year figures been disclosed side by side? And has an independent party, such as internal audit, tested a sample of the claims? DBS publicly meets the first, second and fifth.
The place to start is a one-page definition covering components, methods and owners, followed by a check on whether the three largest use cases have a control group or a recorded baseline. Nothing should be reported upward until a control has run for a full quarter.
Reference
This piece is adapted for Praxora Lab from the original. Originally published at terencekok.com ›
Sources
