TL;DR
BMO says it demonstrated quantum utility for selecting useful variables in a machine-learning dataset, in work developed with Democritus University of Thrace and run on IBM quantum computers. The bank reports that the method matched conventional feature-selection quality with roughly one-hundredth to one-tenth as many samples; practical banking benefits have not yet been demonstrated.
BMO says it demonstrated quantum utility in a machine-learning application using IBM quantum computers, with a method for selecting useful dataset variables that the bank reports matched conventional techniques using roughly one-tenth to one-hundredth as many data samples. Developed with Democritus University of Thrace (DUTH), the work is a research milestone, not a report of a deployed banking product or proven customer benefit.
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
The application adapts the Proxy Quantum Games framework to feature selection, the process of identifying which variables in a dataset are most useful to a machine-learning model. BMO describes the method as treating each variable as a player in a strategic game. Variables score more highly when they move with the model’s target, and less highly when they provide little information or repeat information in other features. The resulting scores rank the variables.
BMO says the method produced results of similar quality to conventional feature-selection approaches while requiring one to two orders of magnitude fewer samples. Put another way, the bank describes a reduction to about one-tenth to one-hundredth of the samples used by the comparison methods. The announcement does not provide the dataset, sample counts, detailed performance measures, or a full account of how the comparison was conducted.
The work used IBM quantum computers, including the IBM Quantum Nighthawk r2 processor, and IBM’s quantum research team reviewed it, according to BMO. The bank says the project was developed with DUTH. It frames the result as an exploration of how quantum computing might help machine-learning systems find useful patterns in complex datasets with fewer examples, rather than as evidence that the approach is already producing business outcomes.
BMO tests quantum utility for machine learning
BMO reports that a quantum method selected useful dataset variables with similar quality to conventional techniques using far fewer samples. Banking benefits remain unproven.
How the method works
The approach ranks dataset variables by how useful and distinct they are to a machine-learning model.
The Proxy Quantum Games framework adapts strategic-game ideas to feature selection.
Variables score higher when they move with the target and lower when they add little or repeat other features.
The resulting scores identify which dataset variables may help the model.
What the result could mean
BMO names these as possible future uses, not tested outcomes.
Client models
Fewer samples could matter when relevant examples are scarce or costly to assemble.
Assessment and detection
BMO identifies risk work as a potential area for future application.
Personalized banking
The announcement does not show the method has been tested for customer services.
Research partners and hardware
The project was developed with Democritus University of Thrace and run on IBM quantum computers.
Evidence still needed
The release leaves key details unavailable for independent assessment.
- Dataset and sample counts
- Baseline methods and numerical quality scores
- Uncertainty ranges and repeated-trial results
- Tests on relevant banking data
- Comparison of total runtime and cost
- Assessment of reliability and governance needs
Potential Uses Across Banking
If the reported result can be reproduced and adapted to operational data, using fewer samples for feature selection could be useful where relevant examples are scarce, costly to obtain, or difficult to assemble. BMO identifies client modeling, risk assessment and detection, and personalized banking as possible areas for future application. Those are potential uses named by the bank; the announcement does not say the method has been tested in those settings.
The claim matters because it connects quantum computing to a defined machine-learning task rather than presenting it only as a general research ambition. At the same time, the announcement does not establish that the quantum method is faster, less expensive, or more accurate overall than classical alternatives. Sample efficiency is one reported measure, and practical value would also depend on performance, computing costs, reliability, data requirements, and whether the method works on real banking problems.
For financial institutions, the distinction between a promising experiment and a usable service is material. Banks would need to evaluate a method against relevant data and existing systems, while considering model governance and risk controls. BMO describes this project as part of a broader effort to apply quantum technology responsibly, but has not announced a client-facing implementation or a timeline for one.
Products Worth Considering

Samsung Galaxy Tab A11+ Kids Edition, 128GB, Royal Blue

Jackery Solar Generator 1000 V2 with 100W Solar Panel

Jackery HomePower 3600 Plus Portable Power Station, 3584Wh, 3600W
As an affiliate, we earn on qualifying purchases.
BMO’s Quantum Research Program
BMO joined the IBM Quantum Network in February 2025, according to its announcement, and says it is investing in quantum research, talent development, and applications. It also names partnerships with the Chicago Quantum Exchange and Quantum Industry Canada as part of its work on financial-services applications. The feature-selection project with DUTH is presented as one effort within that wider program.
The bank says its Institute for Applied Artificial Intelligence & Quantum is an enterprise-wide centre of excellence focused on responsible AI application and governance, as well as advancing quantum capabilities. BMO also cites external rankings: it says it placed ninth among financial institutions in the 2026 Quantum Index Finance ranking and was the highest-ranking Canadian commercial bank. These rankings provide information about the bank’s reported research and adoption position; they do not independently verify the performance of this particular experiment.
The release places the project within early-stage quantum application development. Its central technical claim concerns the number of samples needed for feature selection, not a completed banking deployment. No customer, account, or risk decision is identified as having been made using the method.
““While we remain in the early stages of application development, these projects show how quantum computing can help solve business problems in new ways.””
— Kristin Milchanowski, BMO chief AI and quantum officer and director of the BMO Institute for Applied Artificial Intelligence & Quantum
Evidence Still Needed for Deployment
The announcement does not disclose enough experimental detail to independently assess the reported sample reduction. It does not name the dataset or give its size, the exact baseline methods, numerical quality scores, uncertainty ranges, or results from repeated trials. BMO reports that the method matched conventional techniques, but the release does not provide those comparisons in a form that allows readers to evaluate them.
It is also unclear how the method performs on real banking data, whether its reported sample efficiency carries over to other tasks, or how it compares on total runtime and cost. The release does not describe an operational deployment, independent publication, or client outcomes. IBM’s research team reviewed the work, according to BMO, but the announcement does not specify the scope of that review.
Accordingly, claims about improved risk assessment, more accurate modeling, or more personalized services remain potential applications, not demonstrated results of this project. Further technical details and tests would be needed to establish whether the approach has an advantage for production financial-services systems.
Further Testing Before Banking Use
BMO says the project forms part of its continuing research into quantum approaches for AI and machine learning, but it did not announce a next publication, test date, or launch plan. The next evidence readers would need is a fuller account of the experiment: the datasets and baselines used, the meaning of “similar quality,” and reproducible results across multiple tests.
Any move toward banking use would also require testing on relevant tasks and data, comparison with established classical methods, and assessment of computing costs and governance needs. BMO has named modeling, risk work, and personalization as areas of interest, but has not said which, if any, will be tested next. Until the bank provides those details, the confirmed development remains a reported research demonstration rather than a deployed financial service.
Key Questions
What did BMO report?
BMO reported a quantum-computing method for selecting useful variables in machine learning. The bank says it achieved similar result quality to conventional feature-selection techniques with roughly one-tenth to one-hundredth as many samples.
Who worked on the project?
BMO says the application was developed with Democritus University of Thrace and run on IBM quantum computers, including the IBM Quantum Nighthawk r2 processor. IBM’s quantum research team reviewed the work, according to BMO.
Does the result show quantum computing is better than classical computing?
Not on the information released. BMO reports comparable feature-selection quality with fewer samples, but it does not provide details sufficient to assess overall accuracy, speed, cost, or performance across other tasks.
Is the method being used for banking customers?
The announcement does not describe a deployed product or customer use. Client modeling, risk assessment and detection, and personalization are presented as possible future applications.
What remains unknown about the demonstration?
The release does not give the dataset, sample counts, detailed comparison results, or a deployment timeline. Further technical evidence would be needed to judge reproducibility and relevance to real banking tasks.
Source: thequantuminsider.com
