Skip to content

    Guide · Credit & lending · Consumers

    How banks use AI in credit decisions

    Short answer

    Some banks say AI handles parts of a credit decision, alongside fixed policy rules and often a human underwriter. TD says an AI model grants some mortgage and HELOC pre-approvals, with underwriters reviewing the rest, and RBC says its ATOM model is used in credit adjudication. Announcements rarely say how many decisions are automated or which data models weigh.

    Figures as of November 3, 2026.

    When you apply for a mortgage, credit card or loan, your application passes through data checks, scoring models, policy rules and often a human underwriter. Some Canadian banks now say AI handles parts of that work. This guide explains the process, what TD and RBC have actually announced, and what you can check yourself.

    Reviewed

    Key takeaways

    • A credit decision is a chain of steps: data, scoring models, fixed policy rules and, often, human review. AI can sit in several of those steps.
    • TD says an AI model grants some mortgage and HELOC pre-approvals and that underwriters review the applications it does not approve. RBC says its ATOM model is used in credit adjudication.
    • Bank announcements rarely say how many decisions are automated, which data the models weigh, or how accurate they are.
    • OSFI's Guideline E-23 sets model risk expectations that include AI, but it does not take effect until May 1, 2027.
    • You can check and dispute your credit reports. There is no verified federal right to reasons when you are refused credit.
    How a credit application typically flows
    1. STEP 01

      Application

      You share income, employment, housing costs and debts, and usually consent to a credit check.

    2. STEP 02

      Data gathering

      The bank pulls your credit report and may use information it already holds, such as your account history.

    3. STEP 03

      Document checks

      Pay stubs, tax slips and property documents are reviewed. TD says AI agents now help with this for some mortgage files.

    4. STEP 04

      Risk scoring

      Statistical or machine learning models estimate how likely the loan is to be repaid.

    5. STEP 05

      Policy rules

      Fixed lending criteria and affordability checks apply. A good score doesn't override a hard rule.

    6. STEP 06

      Human review

      Underwriters review some files, especially complex ones or ones not approved automatically.

    7. STEP 07

      Decision

      You receive an approval, a decline or a counter-offer, such as a lower limit.

    Is a credit decision one verdict or a chain of steps?

    It is easy to picture one computer saying yes or no. In practice, a lending decision is usually a series of steps. Data is collected, models estimate risk, the bank applies its lending policy, and a person may review the file before an offer goes out.

    Machine learning is one way to build the scoring step. Instead of a hand-written formula, a model learns patterns from past data, such as which kinds of applicants repaid on time. Newer tools also use AI to read documents, pull out figures like income, and summarize a file for a human reviewer.

    So "the bank used AI" can mean very different things. A model might score your application, check your paperwork, or simply prepare notes for an underwriter. Knowing which step is involved helps you ask better questions.

    How does an application move through the system?

    The exact process differs by bank and by product, and banks don't publish full details. The general pattern looks like this:

    • Application. You provide details such as income, employment, housing costs and debts, and usually consent to a credit check.
    • Data gathering. The bank pulls your credit report from a credit bureau and may use information it already holds about you.
    • Document processing. Pay stubs, tax slips and property documents are checked. This is one area where banks have described using AI agents.
    • Risk scoring. Models estimate how likely the loan is to be repaid. These may be traditional statistical models or machine learning models.
    • Policy rules. Fixed rules apply, such as the bank's lending criteria and affordability checks.
    • Human review. Underwriters look at some files, especially complex ones or ones a model did not approve.
    • Decision and offer. You get an approval, a decline, or a counter-offer such as a smaller amount.

    What has TD announced about AI in lending?

    In a May 2024 release, TD said it introduced an AI model in August 2023 that can grant some mortgage and HELOC pre-approvals within seconds. According to TD, underwriters review the applications the model does not approve. The same release said TD Insurance launched a machine learning model in March 2023 to approve some eligible term life insurance applications.

    In May 2026, TD said it launched its first agentic AI model for real estate secured lending. According to TD's release, AI agents classify documents, extract information, calculate income and check figures against select policies, then produce a summary memo for underwriters. TD said its Trustworthy AI team oversees and monitors the models.

    Two details matter. The 2023 model covers some pre-approvals, not every application. And the 2026 release describes agents preparing work for human underwriters; it does not say the agents make final credit decisions.

    Mitey the yeti examines a speech bubble through a magnifying glass
    A press release tells you what a bank is proud of; your credit report tells you what the model actually sees, so start there.

    What has RBC announced about AI in lending?

    In July 2025, RBC announced ATOM, a proprietary foundation model built by RBC Borealis and trained on large-scale financial data, including client transactions. RBC said ATOM is used in credit adjudication (deciding on credit applications) and in Avion Rewards personalization, and that it was first used in 2023.

    In August 2026, RBC said its retail credit underwriting program, built on ATOM, lets it look beyond traditional credit models and tailor credit offers to individual clients' circumstances. That release announced an industry award; it was not a product launch. The early results RBC described, such as higher approval rates and faster decisions, are RBC's own figures and have not been independently verified.

    Other banks have described AI tools that touch lending too. You can follow dated, sourced announcements on our tracker.

    What don't the announcements tell you?

    Press releases describe goals and highlights. They are not technical disclosures. Read them for what they claim, and notice the gaps.

    None of these releases explains how a declined applicant can find out which factors mattered. That is a question to put to the bank directly.

    Reading bank AI credit announcements
    What the releases sayWhat they don't say
    TD: an AI model grants some mortgage and HELOC pre-approvals; underwriters review the restWhat share of applications are approved automatically, or which factors the model weighs
    TD: AI agents prepare a pre-adjudication memo for underwritersHow underwriters use, check or override that memo
    RBC: ATOM is used in credit adjudication and retail underwritingWhich decisions ATOM influences and how much weight its output carries
    RBC: early results include higher approvals and faster decisionsIndependent measures of accuracy, error rates or fairness across groups of applicants

    What does OSFI's Guideline E-23 expect?

    The Office of the Superintendent of Financial Institutions (OSFI) published its final Guideline E-23 on Model Risk Management on September 11, 2025. It takes effect on May 1, 2027, so as of October 2026 it is not yet in force. It applies to federally regulated financial institutions, including banks.

    E-23 explicitly covers AI and machine learning. It addresses explainability, "black box" models, autonomous decision-making and model drift (when a model's performance shifts as conditions change). It expects data to be accurate, representative and traceable, with attention to bias. Banks are expected to keep an enterprise inventory of models with non-negligible risk, rate each model's risk using factors such as complexity, autonomy and customer impact, and govern models across their whole lifecycle.

    E-23 is a supervisory guideline for institutions, not a consumer rights law. It doesn't give you a way to challenge a decision, but it shapes how carefully the models behind those decisions must be built and checked. Read more in our E-23 explainer.

    What can you do before and after you apply?

    You can't see inside a bank's model, but you can check the data that feeds it. An error on your credit report can affect how any lender assesses you.

    • Get your credit reports. Request your reports from Canada's two national credit bureaus, Equifax and TransUnion, and read them line by line.
    • Dispute errors. The Financial Consumer Agency of Canada (FCAC) suggests gathering proof, using each bureau's dispute form, contacting the lender involved, and escalating if needed. You can also add a free consumer statement to your file.
    • Ask the bank. If you are declined, ask what information was used and whether a person reviewed the file. There is no verified federal law guaranteeing reasons for a credit refusal, so the answer may be limited, but asking costs nothing.
    • Know the Quebec rule. Quebec's private-sector privacy law gives extra rights when a decision is based exclusively on automated processing. See your rights when an algorithm denies you.
    • Complain if something went wrong. Our complaint guide walks through the bank's process and the external ombudsman.

    Credit bureaus are regulated provincially

    FCAC notes that the federal government does not regulate credit bureaus. If a dispute with a bureau stalls, complaints go to your provincial or territorial consumer affairs office.

    From the Index: AI in credit decisions

    What the Canadian Bank AI Index shows on this topic, counted from the published data. Every number can be rechecked in the database.

    From the Index: AI in credit decisions: inputs, steps and result
    Institutions scored35Canadian Bank AI Index, 2026 Q4
    Entries in the database210Sourced and dated, as of 2026-11-01
    Entries filed under Credit & lending10
    Of those, reaching customers10Customer-facing or running on customer accounts
    Live (launched or disclosed in use)8Pilots and announcements excluded
    Institutions with a live, sourced customer use5 of 35 (14%)TD, Neo, RBC, CIBC, Vancity

    Open these entries in the database →Computed November 1, 2026.

    What this means for you

    • Get your Equifax and TransUnion credit reports before you apply and read them line by line; an error there can affect how any lender assesses you.
    • Dispute errors with proof through each bureau’s dispute process, contact the lender that reported the information, and consider adding a free consumer statement to your file.
    • If you are declined, ask what information was used and whether a person reviewed the file. There is no verified federal right to reasons, but asking costs nothing.
    • In Quebec, decisions based exclusively on automated processing carry extra rights; see your rights when an algorithm denies you.
    • If you think something went wrong, use the bank’s complaint process and then the external ombudsman, as set out in our complaint guide.

    Questions people ask

    Does AI decide whether I get a mortgage in Canada?

    It depends on the bank and the application. TD says an AI model grants some mortgage and HELOC pre-approvals and that underwriters review the ones it doesn't approve; its 2026 agentic AI release describes agents preparing memos for underwriters, not making final decisions.

    Can I find out why a bank's model declined me?

    You can ask, but there is no verified federal right to reasons for a credit refusal. In Quebec, s. 12.1 of the private-sector privacy law gives rights to information when a decision is based exclusively on automated processing.

    Is OSFI's Guideline E-23 in force?

    Not yet. OSFI published it on September 11, 2025, and it takes effect on May 1, 2027.

    What if my credit report has an error?

    Gather proof, use the dispute process at Equifax and TransUnion, and contact the lender that reported the information. FCAC also says you can add a free consumer statement to your file.

    Who regulates credit bureaus?

    Not the federal government, according to FCAC. Complaints about a credit bureau go to your provincial or territorial consumer affairs office.

    Educational information only, not financial, legal or privacy advice. Rules and bank features change; check the linked official pages and talk to your financial institution or a qualified professional about your situation.

    Keep reading

    Related guides

    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.