
Share
BMO says a quantum computing experiment with IBM identified the most relevant inputs for machine learning models using far less data than classical methods. The claim raises real questions about timing, replication, and payoff.
BMO says it has crossed a meaningful threshold in applied quantum computing. The bank announced on November 19 that it demonstrated "quantum utility" in a machine learning application, using IBM quantum hardware to perform work at a scale that exact classical methods cannot practically replicate. The specific task: identifying which inputs in a dataset actually matter to an AI model's output, a problem known as feature selection.
This is not a research curiosity. Feature selection sits at the center of nearly every serious machine learning pipeline in finance. Models built on bloated, redundant, or irrelevant inputs tend to overfit, cost more to run, and perform worse out of sample. If a quantum approach can reliably strip that noise out using less data, the implications touch credit risk models, fraud detection, and client segmentation across the industry.
BMO's claim is specific and testable. Working with the Democritus University of Thrace and using IBM's latest Nighthawk r2 processor, the bank adapted something called the Proxy Quantum Games framework. Each feature in a dataset becomes a "player" in a game. Features that move together with the target outcome score high. Features that are redundant or uninformative score low. The scoring runs on a quantum computer, and BMO says the resulting method matches the quality of traditional feature selection while requiring one to two orders of magnitude fewer samples.
One to two orders of magnitude is a meaningful number, not a rounding error. If accurate and reproducible, that is the difference between needing thousands of data points and needing tens or hundreds. For institutions constrained by data scarcity, whether due to privacy rules, cold-start problems with new products, or genuinely rare events like fraud, that gap could translate directly into better models built faster.
Kristin Milchanowski, BMO's Chief AI and Quantum Officer, framed this as an early but important step. "Demonstrating quantum utility is an important milestone for BMO and an exciting step forward for the broader financial services industry," she said, adding that the bank remains "in the early stages of application development." That qualifier matters. Quantum utility is not quantum advantage. It means the computation ran reliably at a scale beyond practical classical reproduction, not that it beat every classical alternative on cost or speed.
Scott Crowder of IBM Quantum called it progress toward "real-world quantum use cases in the financial services industry," which is the kind of validation vendors like to see from their enterprise partners. IBM has an obvious interest in showcasing bank-grade applications as it pushes toward utility-scale quantum computing. That does not make the result wrong, but it is worth remembering IBM's hardware and research team were directly involved in reviewing the work.

The risk management angle is where this gets interesting for investors and risk officers alike. Better feature selection, done with less data, means banks could build more accurate credit and fraud models in situations where historical data is thin. Emerging markets, new product lines, and novel fraud patterns are all classic cases where data is scarce and models are weak. A technique that extracts signal from smaller samples addresses a real structural limitation in current risk modeling, not a theoretical one.
Context matters here too. BMO joined the IBM Quantum Network in February 2025, a relatively recent commitment, and has since built out partnerships with the Chicago Quantum Exchange and Quantum Industry Canada. The bank also stood up an internal Centre of Excellence, the BMO Institute for Applied Artificial Intelligence & Quantum, to govern how these technologies get deployed. That governance layer is not incidental. Quantum results that cannot be audited or reproduced by compliance teams are of limited use to a regulated bank, no matter how elegant the underlying math.
External validation adds some weight to the announcement. BMO ranked ninth among financial institutions globally in the 2026 Quantum Index Finance ranking, the highest-ranking Canadian commercial bank on that list, which measures workforce, research and development, and adoption. Separately, Evident AI placed BMO among the top 20 banks globally in AI capability, ranking it second of 50 banks for AI use-case outcomes and first in Responsible AI Process, covering governance, risk management, training, and operational controls. Those rankings suggest BMO's quantum push sits inside a broader, reasonably well-resourced AI strategy rather than standing alone as a one-off pilot.
Steve Tennyson, BMO's Chief Technology and Operations Officer, described the milestone as the product of "years of investment in innovation, quantum computing, artificial intelligence and strategic partnerships." That is a fair characterization of the timeline. It is also a reminder that quantum computing programs at major banks are long-horizon bets, measured in years of R&D spend before any client-facing product emerges.
The headline claim deserves scrutiny rather than skepticism. A one-to-two order of magnitude reduction in data requirements for feature selection, if it holds up under independent replication and scales beyond the specific dataset tested, would be a genuinely useful tool for risk modeling in data-constrained environments. But this remains a single demonstration, reviewed by the vendor whose hardware it ran on, not an independently audited, widely reproduced result. Investors should treat this as evidence that BMO's quantum investment is producing tangible research output, not as proof that quantum computing is about to reshape bank risk models at scale. The more useful signals to watch going forward are whether BMO or peers publish the underlying methodology for outside review, whether the technique holds up on larger and more varied datasets, and whether any of this translates into deployed production models rather than demonstration projects. Until then, this is a well-documented research milestone, not a commercial product.
Tags
Original Sources
BMO Uses IBM Quantum Computers to Identify Which Inputs Matter Most to AI Models - HPCwire
↗ https://www.hpcwire.com/off-the-wire/bmo-uses-ibm-quantum-computers-to-identify-which-inputs-matter-most-to-ai-models
About the author
Marcus began tracking AI's market implications in 2016, noticing AI-related patent filings accelerating ahead of earnings upgrades before most of the sell-side had caught on. A former fixed-income quantitative analyst, he spent two decades building models that priced risk across emerging markets before pivoting to cover the economic impact of AI full-time. His writing translates opaque technical developments into clear risk/reward terms — and he's rarely diplomatic about the gap between AI valuations and underlying fundamentals. He believes most market participants still underestimate AI's long-run deflationary effect on knowledge work.
More from The Analyst →This Week's Edition
20 November 2026
29 articles
Related Articles

Elsevier Gives ClinicalKey AI Full EHR Access, Betting on Patient Context Over Chatbots
Products & Applications · 6 min

National Academies Report Calls for Faster, Tougher Retraction Standards
Models & Research · 5 min

Who Decides If Your Treatment Is Medically Necessary? Industry Leaders Say It Shouldn't Be AI Alone
Policy & Regulation · 6 min
Related Articles

Elsevier Gives ClinicalKey AI Full EHR Access, Betting on Patient Context Over Chatbots
Products & Applications · 6 min

National Academies Report Calls for Faster, Tougher Retraction Standards
Models & Research · 5 min

Who Decides If Your Treatment Is Medically Necessary? Industry Leaders Say It Shouldn't Be AI Alone
Policy & Regulation · 6 min
More Stories
© 2026 Cedar & Bloom. All rights reserved.