In Canadian financial services firms, AI automation pays first in high-volume work with written rules: onboarding and identity checks, AML alert triage, reading loan and account documents, assembling regulatory reports and answering routine client questions. Named people keep credit decisions, suspicious transaction reports and filings. FINTRAC obligations apply as before; federally regulated institutions add OSFI Guideline E-23 from May 1, 2027.
Remolda builds these workflows for dealers, lenders, credit unions and fintechs, starting with one measured pilot. The two-week AI Opportunity Audit costs $2,900 CAD plus applicable taxes. Send a written request with your type of firm, regulator and the workflow you want to fix, or call +1 (613) 867‑7767.
Where AI fits in a financial services firm
| Workflow | What AI does | Who decides |
|---|---|---|
| Onboarding and KYC | Reads ID and account documents, checks required fields, records the verification method | Onboarding officer approves the account |
| AML alert triage | Gathers KYC, history and prior alerts, drafts a disposition with reasons | Analyst closes or escalates; compliance officer decides on a report |
| Sanctions and name screening | Explains likely false positives with the matching fields | Analyst confirms |
| Loan and account files | Extracts income, employment and property data, lists missing items | Underwriter decides credit |
| Regulatory and board reporting | Pulls figures from source systems, drafts commentary, flags gaps | Owner of each report signs off |
| Complaints and client questions | Logs and acknowledges, drafts replies from approved content | Staff send; complaint owner decides |
Advisory practices have their own guide: AI automation for financial advisors covers KYC updates, complaints and the 2026 cost report for dealers' advisors.
Onboarding and KYC under the PCMLTFA
FINTRAC lists the methods a reporting entity can use to verify a person's identity: government-issued photo identification, the credit file (Canadian bureau, file at least three years old), dual-process, affiliate or member, and reliance. An onboarding workflow can read the ID or documents, check that they are current, record which method was used and hold the account until every required field is filled.
The agent prepares; the onboarding officer approves. Records are kept for at least five years under FINTRAC's record-keeping rules. Reading identity documents, statements and forms is our intelligent document processing work.
AML alert triage
Many alerts from a transaction monitoring system turn out to be explainable, and closing each one starts with gathering context: the client's KYC profile, recent transactions, earlier alerts and notes. An agent can do that gathering and write a first disposition with the indicators it found and its reasoning.
Three design rules keep this defensible:
- The analyst decides. The agent drafts; it never closes an alert on its own.
- Reasons are logged. Each decision is stored with the evidence the analyst saw, so an examiner can follow it.
- Escalation stays human. FINTRAC expects a suspicious transaction report as soon as practicable once there are reasonable grounds to suspect; the compliance officer makes that call.
Measure the pilot on minutes per alert and on agreement between the agent's draft and the analyst's decision. Multi-step builds like this are our AI workflow automation work.
Loan files, reporting and client service
Loan and account files. Language models read pay stubs, notices of assessment, appraisals and bank statements, extract the figures an underwriter needs and list what is missing. The credit decision stays with the underwriter, and in Quebec a decision based exclusively on automated processing must be disclosed to the person (Law 25, s. 12.1).
Regulatory and board reporting. An agent pulls figures from the core system and data warehouse, fills the report template, drafts the commentary and flags figures that moved unusually. The owner of each report signs off. Dashboards on the same data are our AI dashboards work.
Client service. An assistant answers routine questions (hours, documents needed, status of an application) from approved content and passes the rest to staff. Advice and anything about a specific account go to people. This is our customer support chatbot work.
The rules that shape the build
| Rule | Applies to | What it means for an AI workflow |
|---|---|---|
| PCMLTFA and FINTRAC guidance | Reporting entities: banks, credit unions, securities dealers, money services businesses and others | Compliance program with an officer, policies, risk assessment, training and an effectiveness review at least every two years; identity methods; records at least five years |
| OSFI Guideline E-23, Model Risk Management | Federally regulated financial institutions, from May 1, 2027 | AI and ML models in a model inventory with risk rating, validation, monitoring and governance, in proportion to risk; vendor models included |
| OSFI Guideline B-10, Third-Party Risk | Federally regulated institutions, since May 1, 2024 | Due diligence, contract terms, monitoring and exit plans for AI and cloud vendors |
| OSFI Guideline B-13, Technology and Cyber Risk | Federally regulated institutions, since January 1, 2024 | Security, resilience and change control for the systems the AI runs on |
| PIPEDA and Quebec Law 25 | Client personal information | Purpose limits, safeguards, vendor agreements; Law 25 s. 12.1 notice of exclusively automated decisions and s. 17 assessment before data leaves Quebec |
| Provincial regulators (FSRA, AMF, others) | Credit unions, provincial insurers and lenders | Provincial rules apply; check whether your regulator has adopted or referenced OSFI guidance |
Mapping a workflow to these rules, including the model inventory E-23 expects, is part of our AI compliance work. The sector overview with packages is on AI consulting for finance, with pages for banks and credit unions and fintech.
How to start
- Pick one workflow with volume and a clear decision owner: alert triage, onboarding or loan file intake.
- Measure it for two weeks: cases, minutes per case, rework.
- Check it against FINTRAC, privacy and vendor rules before building; for federally regulated institutions, add the model to the inventory now, ahead of E-23.
- Pilot with review on every output, then decide from the numbers.
The AI Opportunity Audit ($2,900, two weeks) ranks your candidate workflows and checks them against the rules above. The AI Pilot Sprint ($9,800, six weeks) builds one and measures it against the baseline. In our experience the pace is set by access to the core system and by how quickly compliance can review the design; plan for that in week one.
Have a compliance-heavy workflow you want to fix? Send a written request with your regulator, systems and monthly volumes. We reply within one business day.
Sources
- FINTRAC: Methods to verify the identity of persons and entities
- FINTRAC: Compliance program requirements
- FINTRAC: Record keeping requirements for financial entities
- FINTRAC: Reporting suspicious transactions
- OSFI: Guideline E-23 Model Risk Management (2027), letter
- OSFI: Guideline B-10 Third-Party Risk Management
- OSFI: Guideline B-13 Technology and Cyber Risk Management
- LégisQuébec: CQLR c. P-39.1 (ss. 12.1, 17)

