TL;DR
BMO says it and Democritus University of Thrace used IBM quantum computers to test a method for ranking machine-learning features. The bank reported similar results to conventional feature-selection techniques while using roughly one-tenth to one-hundredth as many data samples; practical banking benefits have not yet been demonstrated.
BMO says it has demonstrated a machine-learning feature-selection method on IBM quantum computers, in work developed with Democritus University of Thrace. The bank reports that the method produced results of similar quality to conventional techniques using roughly one-tenth to one-hundredth as many data samples, a research result that could inform future financial applications but does not yet establish a customer-facing 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 project addresses feature selection: deciding which variables in a dataset are most useful to a machine-learning model. BMO and DUTH adapted the Proxy Quantum Games framework, recasting feature ranking as a strategic game that runs on a quantum computer. In the framework, each dataset feature acts like a player. Features score higher when they move with the model’s target and lower when they offer little information or repeat information already present in other features.
BMO says the resulting rankings had similar quality to those from conventional feature-selection techniques while requiring one to two orders of magnitude fewer samples. The company describes this as a demonstration of quantum utility: work carried out reliably at a scale that, in its account, exact traditional computing methods cannot practically reproduce. The announcement does not provide the dataset size, full comparative results, or enough methodological detail to independently evaluate that characterization.
The work used IBM quantum computers, including the IBM Quantum Nighthawk r2 processor, and was reviewed by IBM’s quantum research team, according to BMO. The bank points to improved client modeling, risk assessment and detection, and more personalized services as possible future applications. Those are potential uses, not outcomes reported from a deployed banking system.
BMO tests quantum feature ranking for machine learning
BMO and Democritus University of Thrace used IBM quantum computers to rank useful model inputs. The reported sample savings are a research result; banking benefits have not been demonstrated.
How the method works
The team adapted Proxy Quantum Games to treat each dataset feature as a player.
Features score higher when they move with the model’s target.
Features score lower when they add little information or repeat other inputs.
The rankings were run on IBM quantum computers, including the IBM Quantum Nighthawk r2 processor.
Research result, not rollout
The result could inform future financial applications, but no deployed banking benefit was reported.
- Similar feature-selection quality to conventional techniques
- Roughly one-tenth to one-hundredth as many samples
- Improved live credit or risk decisions
- Personalized services or lower operating costs
Evidence still needed
The announcement leaves key details open for independent evaluation.
- Dataset sizes, sample counts and test conditions
- Which conventional methods were compared and how quality was measured
- Performance and costs on operational banking data
- Further testing against well-tuned classical approaches
Part of a wider research effort
BMO frames the work as early-stage application development supported by partnerships and internal research.
A Research Result, Not a Banking Rollout
The demonstration matters because feature selection is a core step in building machine-learning models: reducing unnecessary or repetitive inputs can affect how efficiently a model is trained and how useful its predictions are. If the reported sample reduction can be reproduced on relevant financial data, it could offer a way to investigate models when labeled or otherwise suitable data are limited. That prospect is particularly relevant to banks, where modeling and risk decisions can depend on complex datasets.
But the announcement establishes a technical research result, not a business advantage in production. BMO has not said that the method has improved credit decisions, risk detection, service personalization, model accuracy in live operations, or costs. The reported comparison concerns sample requirements and similar feature-selection quality; it should not be read as evidence that quantum computing has already made the bank’s models better or reduced operational data needs.
The work also reflects a broader industry question: whether quantum processors can offer practical value on tasks that matter to businesses, rather than only on carefully bounded research problems. BMO’s result adds a finance-related example to that exploration. Its practical relevance will depend on further testing, including how the approach performs on real-world datasets and alongside the classical systems banks already use.
Products Worth Considering

Apple Watch SE 3 GPS + Cellular 40mm Smartwatch

Jackery HomePower 3600 Plus Portable Power Station, 3584Wh, 3600W

Bose TV Speaker (2nd Gen) Compact Soundbar with Bluetooth and HDMI Connectivity
As an affiliate, we earn on qualifying purchases.
BMO’s Quantum Research Partnerships
BMO describes the project as part of a wider effort to apply quantum technology to artificial intelligence and machine learning. The bank joined the IBM Quantum Network in February 2025 and says it is also working through partnerships with the Chicago Quantum Exchange and Quantum Industry Canada. It has established the BMO Institute for Applied Artificial Intelligence & Quantum, which it describes as an enterprise-wide center focused on responsible AI application and governance and on developing quantum capabilities.
The announcement places the research within a developing field rather than a finished commercial service. BMO Chief AI and Quantum Officer Kristin Milchanowski said the bank remains in the early stages of application development and described the work as building on years of research and collaboration. BMO also cited its position in the 2026 Quantum Index Finance ranking, where it said it ranked ninth among financial institutions globally and was the highest-ranking Canadian commercial bank. Such readiness rankings measure areas including workforce, research and adoption; they do not independently verify the performance of this particular 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’s Chief AI and Quantum Officer
Performance and Deployment Questions
The announcement does not disclose the datasets, sample counts, test conditions, or detailed benchmark results behind the comparison. It also does not specify which conventional techniques were used, how the researchers measured “similar quality,” or whether the results have been published in a peer-reviewed paper. BMO says IBM’s quantum research team reviewed the work, but that is not the same as independent publication or replication.
It remains unclear whether the method will perform as reported on larger or operational banking datasets, how it compares with well-tuned classical approaches, and what computing resources or costs it requires. BMO has not announced a timetable for a pilot, a production system, or a client service based on the research. The stated applications remain possibilities rather than confirmed deployments.
Further Testing Before Banking Use
BMO says it intends to continue researching practical quantum applications through its partnerships and internal AI and quantum institute. The next meaningful evidence would be further technical reporting that spells out the benchmark design and results, followed by testing on financial datasets that reflect real operational constraints. Any move into banking use would also require evaluation of reliability, model governance, privacy, risk controls and whether the approach offers an advantage over classical methods.
The bank has not announced a specific next milestone or launch date for this feature-selection work. Until it does, the development is best understood as a research demonstration that may guide later experiments, not as a new tool available to customers or staff.
Key Questions
What did BMO demonstrate?
BMO says it ran a quantum-computing method for ranking useful machine-learning dataset features, developed with Democritus University of Thrace and using IBM quantum computers.
What does the sample reduction mean?
BMO reported similar feature-selection quality to conventional techniques using roughly 10 to 100 times fewer data samples. The announcement does not give the underlying sample counts or full benchmark details, so the comparison cannot be independently assessed from the release alone.
Is the method being used in BMO’s banking services?
The announcement does not say that it has been deployed. BMO names client modeling, risk assessment and personalization as potential applications, not as completed operational uses.
Which IBM quantum processor was involved?
BMO says the work used IBM quantum computers, including the IBM Quantum Nighthawk r2 processor. It does not specify which processor ran each part of the experiment.
What remains to be verified?
Key details include the datasets and benchmarks, the definition of similar quality, and whether independent researchers can reproduce the results. The release also does not establish a business benefit or deployment timeline.
Source: thequantuminsider.com
