Skip to content

    Methodology · 2026 Q4

    How the Canadian Bank AI Index works

    Short answer

    Each institution scores out of 100 across five equal pillars of 20 points: customer deployment, governance disclosure, model-risk readiness, transparency to customers, and AI talent. Every point is a yes-or-no criterion met only with a public, dated source. We count AI as customer-facing only when the source says customers use it.

    Figures as of November 1, 2026.

    The Index scores public evidence. It answers one question: how much has each institution shown, in sources anyone can check, about using AI with customers and governing it? It does not judge whether that AI is accurate, fair or safe.

    Mitey, data editor · Evidence checked

    1. Who is included

    35 institutions that hold Canadians’ deposits or move their money: the Big Six, other domestic banks, Desjardins, digital banks and bank brands, the largest credit unions by member count, Alberta’s ATB, and the fintechs and payment networks millions of Canadians use. We add institutions each quarter; inclusion is not an endorsement and no institution pays to be listed or scored.

    2. Five pillars, 19 criteria, 100 points

    Each pillar is worth 20 points. Equal weights are deliberate: shipping AI without explaining its oversight should not top an index about trust.

    Customer deployment 20

    Does AI reach customers in sourced, live products?

    • Live customer use: Assistants & chat3 pts
    • Live customer use: Money insights3 pts
    • Live customer use: Fraud & identity3 pts
    • Live customer use: Credit & lending3 pts
    • Live customer use: Wealth & investing3 pts
    • Live customer use: Business banking3 pts
    • Customers can use an AI assistant directly2 pts

    Governance disclosure 20

    Does it publicly explain who oversees AI and how?

    • Publishes responsible-AI principles5 pts
    • Discloses board or committee oversight of AI5 pts
    • Names AI as a risk in its annual reporting5 pts
    • Names an executive accountable for AI5 pts

    Model-risk readiness 20

    Does it say its model-risk controls cover AI (OSFI E-23 or equivalent)?

    • Says its own model-risk framework covers AI/ML12 pts
    • Names OSFI E-23, or the AMF equivalent for Quebec-regulated institutions8 pts

    Transparency to customers 20

    Are customers told when AI is used, and does it report openly?

    • Tells customers when AI or automated decisions are used10 pts
    • Signed Canada’s Voluntary Code on generative AI5 pts
    • At least 3 first-party AI disclosures in the last 24 months5 pts

    AI talent & capacity 20

    Does it disclose an AI team, headcount or active AI hiring?

    • Has a named AI lab or dedicated AI team8 pts
    • Discloses AI or data-science headcount6 pts
    • Shows active AI hiring6 pts

    3. Bands

    Leader (80+)
    Broad, sourced customer deployment and detailed public governance.
    Advanced (60+)
    Several live customer uses and most governance disclosures in place.
    Developing (40+)
    Some live uses or disclosures, with clear gaps.
    Emerging (20+)
    A few public signals; little detail on oversight.
    Little public evidence (0+)
    We found little or nothing published. That is not proof AI is unused.

    4. How we label who AI reaches

    A 2026 review of our earlier tracker found it counted consumer surveys and a foundation model as “customer-facing”. We fixed that with four strict labels:

    Customers use it directly
    Customers or business clients use or see it: chat, in-app insights, AI advice shown to them, an ID check they go through.
    Runs on customer accounts
    It acts on customer accounts or decisions without the customer interacting: fraud and AML screening, credit models, pricing.
    Staff tools
    Used by employees: advisor assistants, contact-centre tools, coding assistants, back office.
    Research & governance
    Labs, foundation models not yet in a named product, principles, training programmes, partnerships and surveys.

    5. What counts as evidence

    1. Annual reports, MD&A, annual information forms and other regulatory filings.
    2. Earnings calls and investor-day materials.
    3. The institution’s own newsroom, press releases and product or help pages.
    4. Reputable news reporting that names the institution and the use.
    5. Vendor-issued announcements (labelled), job postings and patents we could open.

    Every row records the source type, the publisher, the date printed on the source, its launch date (see 5a), and the date we last opened it. Today the database draws on 65 press release, 59 newsroom / blog, 26 annual report, 20 news report, 13 report / study, 10 patent, 6 job posting, 6 earnings call / investor day, 3 product page, 2 regulatory filing sources. We never infer a criterion: a dash means we looked and found nothing public.

    5a. Rules we apply the same way to everyone

    Model risk is two separate questions
    Does the institution say its own model-risk framework covers AI or machine-learning models (12 points)? Does it name OSFI Guideline E-23, or the AMF’s equivalent for Quebec-regulated institutions (8 points)? Mentioning E-23 earns only the 8; describing the rule without saying the framework covers AI earns neither 12 nor a share of it.
    Brands and subsidiaries
    A brand or subsidiary (Simplii, Tangerine, Manulife Bank, PC Financial, Canadian Tire Bank) is scored on evidence it publishes or that names it. A privacy notice that covers its customers counts, because that is what those customers are told. Parent-group disclosures appear as context on the profile and score on the parent’s own profile.
    Launch date, not page date
    Each entry has a launch date separate from the source date: when it launched or was announced, when it was first disclosed (reports and filings that give no launch date), or “not published” when the only date is page metadata or the day we read the page. Charts by year and the “newest” lists use launch dates; entries with no published date are counted but not charted.
    Where AI applies
    Entries for U.S.-only operations (TD Bank, BMO U.S.) are tagged “U.S. only”. They count toward the institution’s record, but the account tool leaves them out because they never run on a Canadian account.
    Hiring is counted, not guessed
    AI hiring is the number of postings with AI, machine-learning or data-science terms in the title on the institution’s own careers board, counted with the same searches for every board, on the date shown. If a board cannot be read, we say so and count it later rather than scoring a zero as if nothing was there.

    6. Limits and conflicts

    • Big, listed banks publish far more than credit unions. The Index rewards disclosure, so size helps. We say so on every score.
    • Institution claims (hours saved, users, “first in Canada”) are reported as claims, not verified.
    • AIBanks.ca is part of the MITEY Network, which runs consumer finance sites. No financial institution pays us, and no score can be bought.
    • We use the Evident AI Index only as cited context; it does not feed our scores.

    7. Updates and corrections

    We re-check evidence and publish a new edition each quarter, adding deployments as they are announced. Institutions and readers can send a dated, public source to hello@mitey.ca or through the tips form.

    8. Change log

    1. November 4, 2026Launch dates corrected from each entry’s own source text: TD mortgage/HELOC pre-approval (August 2023), TD Insurance term life (March 2023), TD nudges (November 2020), TD U.S. Virtual Assistant (April 2020), TD Insurance FRISS (June 2021), Desjardins phone assistant (December 2021), Scotiabank predictive prompts (end of 2025), RBC ATOM (in use since 2023) and three CIBC tools (2025). A test now fails if an entry’s text names an earlier launch than its launch date.
    2. November 4, 2026Policy decisions about AI, such as RBC’s block on third-party AI agents, moved out of the deployment database and its counts into a separate policies list.
    3. November 3, 2026Added RBC’s October 27, 2026 decision to block third-party personal AI agents from online banking, from RBC’s own page. Other banks’ positions are reported by media only and are not yet in the database.
    4. November 3, 2026AMF AI guideline page rebuilt from the AMF’s own guideline page (updated April 24, 2026) instead of a law-firm summary; the unverified draft-date row was removed. Unverified key dates are now dropped from every regulation page.
    5. November 3, 2026Method made consistent. Model risk is now two separate questions: does the institution say its own framework covers AI/ML (12 points), and does it name OSFI E-23 or the AMF equivalent (8 points)? Naming E-23 alone no longer earns 20. Scotiabank, Laurentian and EQ Bank lose the framework points; TD, BMO, ATB and Coast Capital lose the naming points.
    6. November 3, 2026One brand rule: brands and subsidiaries are scored on evidence they publish or that names them; a privacy notice covering their customers counts. Simplii (CIBC), Manulife Bank (Manulife) and Canadian Tire Bank (Canadian Tire Corporation) no longer inherit parent disclosures. Simplii is relabelled a bank brand.
    7. November 3, 2026AI hiring counted on Scotiabank’s (14 postings) and National Bank’s (15) careers boards, which could not be read on November 1; Tangerine gets credit for two postings named for it. Scores re-ranked.
    8. November 3, 2026Entries now carry a launch date separate from the source date. Rows dated only by page metadata (vendor case studies, “last updated” pages, undated job postings) are no longer shown as new or charted by that date. Re-dated: CIBC complaint responses (Q4 fiscal 2025), CIBC AIVA (Q4 fiscal 2024), National Bank’s AWS agents (2025), KOHO voice (2025). Questrade entry trimmed to what its source supports. U.S.-only TD and BMO rows are tagged and left out of the Canadian account tool.
    9. November 2, 2026Recategorised five customer ID and liveness checks (RBC, EQ Bank, Simplii, Interac, Wealthsimple) from Cybersecurity to Fraud & identity, the category the method defines for them. RBC, EQ Bank, Simplii and Wealthsimple each gain 3 deployment points; Interac’s is announced, not live, so its score is unchanged. TD’s named-AI-leader note updated after Luke Gee’s reported move to Scotiabank.
    10. November 1, 2026Rebuilt as the Canadian Bank AI Index (2026 Q4 edition): 35 institutions scored, 210 sourced deployment entries (up from 40), a regulation tracker with 19 verified items, and Evident AI Index context. Retired the chatbot checklist, budgeting sheet and readiness quiz; merged the scam guides into one.
    11. November 1, 2026Reach labels corrected: consumer surveys and foundation models no longer counted as customer-facing (four labels: customer-facing, customer-affecting, internal, research & governance).
    12. November 1, 2026Removed three rows whose sources do not say AI is used: Desjardins Alvie (its own report says analytics, not AI, for now), ATB’s 2026 virtual assistant, and RBC Genesis. Added AI job-posting counts for CIBC (12) and Desjardins (11); Scotiabank and National Bank job boards could not be counted and score 0 on that criterion until they can.
    13. October 30, 2026First sourced tracker published: 40 first-party announcements across 10 institutions.

    9. Data and reuse

    The method applied to Wealthsimple

    Wealthsimple meets 8 of 19 criteria. Each pillar is out of 20; the total is their sum.

    The method applied to Wealthsimple: inputs, steps and result
    Customer deployment11 / 204 of 7 criteria met
    Governance disclosure5 / 201 of 4 criteria met
    Model-risk readiness0 / 200 of 2 criteria met
    Transparency to customers15 / 202 of 3 criteria met
    AI talent & capacity6 / 201 of 3 criteria met
    Largest gapSays its own model-risk framework covers AI/ML (12 points)
    Total, 2026 Q437 / 100 (Emerging)Rank 9 of 35

    Open this comparison →

    Mitey the yeti at a desk, a checklist notebook open beside the laptop
    Mitey writes every rule down before scoring, so anyone can recompute a score from the CSV.

    Get the quarterly Canadian Bank AI Index update

    New scores, new deployments and rule changes, once a quarter from the MITEY Network. Unsubscribe any time.

    Educational information only, not financial, legal or privacy advice. Index scores measure public evidence, not service, safety or fairness. If you think you have been scammed, contact your financial institution right away and report it to the Canadian Anti-Fraud Centre at 1-888-495-8501. Last reviewed November 1, 2026.