The Canadian small-business AI privacy checklist (2026)

Most AI vendors will tell you privacy is “handled” because they hold a SOC 2 report. That badge describes their internal controls. It says nothing about whether your use of their tool meets Canadian privacy law — and it is your business name on the intake form when a customer asks who has their information.

I build AI agents that owners keep and control, so the data question lands on my desk constantly. Here is the honest version for a Canadian small business.

Short answer: A Canadian small-business AI privacy checklist comes down to five things: get meaningful consent before the AI collects personal data, know exactly where that data is stored, limit what you keep and for how long, be able to delete it on request, and have a plan to report a breach to the Office of the Privacy Commissioner. PIPEDA applies to almost every business with no small-business exemption, and Quebec’s Law 25 is stricter still.

Does PIPEDA even apply to a business my size?#

Yes. PIPEDA applies to any organization that collects, uses, or discloses personal information in the course of commercial activity, and there is no carve-out for being small. A solo salon booking clients through an AI agent is covered the same way a 40-person firm is. Alberta, British Columbia, and Quebec have their own private-sector laws deemed substantially similar, so which statute governs depends on where you and your customers are — but “we’re too small to matter” is not a category the law recognizes.

The practical read: the moment your AI captures a caller’s name, number, or reason for calling, you are the organization responsible for protecting it. Not the vendor. You. That frame sits under everything below, and under the vendor questions worth asking anywhere.

Where does your customer data actually go once AI touches it?#

Before you can protect the data, you have to know its path. Map it once, in plain terms, the same way I map any deployment: trigger, action, system of record, and the human who steps in when something is off.

• Trigger: a customer calls, texts, or fills out a form.
• AI action: the agent transcribes the call, pulls out the name, number, and request, and drafts a booking or a follow-up.
• System of record: the structured note lands in your CRM, calendar, or a shared sheet — the place you already trust.
• Human escalation: anything sensitive or unusual (a complaint, a health detail, a legal question) routes to you instead of being auto-handled.

Write down every place a copy lives: the phone/SMS provider, the transcription step, the CRM, the vendor’s own logs. That last one is where owners get surprised — many subscription tools retain transcripts for 30 to 90 days or longer, in a US cloud, by default. You cannot protect a copy you did not know existed.

What PIPEDA actually requires before you deploy#

PIPEDA is built on ten principles, but four of them do the real work for an AI deployment: consent, limiting collection, safeguards, and breach response. Here is what each means on a Tuesday morning, not in a law review.

• Meaningful consent. The customer should know their information is being collected and why. For an AI agent, that means a short spoken or written disclosure — “you’re speaking with an automated assistant for [business]” — and a privacy line on your site. Consent you cannot honestly claim is consent you do not have.
• Limit collection and retention. Collect what the booking needs, not the customer’s life story, and set a deletion window. If you do not need call recordings after the appointment is booked, do not keep them for a year.
• Safeguards. Reasonable security appropriate to the sensitivity of the data — access controls, encryption, and knowing which subprocessors can see it.
• Breach reporting. This is the one owners miss. Under PIPEDA’s mandatory rules, if a breach creates a real risk of significant harm — identity theft, financial loss, reputational damage — you must report it to the Office of the Privacy Commissioner and notify affected people as soon as feasible. You must also keep a record of every breach for at least 24 months, including the minor ones you decided not to report.

One note on the legal weather: Bill C-27, which would have replaced PIPEDA with the CPPA and added a dedicated AI law, died when Parliament was prorogued in January 2025. So in 2026, PIPEDA is still the federal rule. Do not build to a law that never passed.

PIPEDA vs Quebec’s Law 25: the parts that differ#

If any of your customers are in Quebec, you are also under Law 25, and it sets a higher bar. Build to the stricter of the two and you are covered everywhere.

Requirement	PIPEDA (federal)	Quebec Law 25
Breach notification	Report to the OPC + individuals when there’s a real risk of significant harm	Report to the CAI + individuals, expected within days
Privacy impact assessment	Recommended, not mandatory	Mandatory for certain projects, including some tech transfers
Max penalty	Up to $100,000 for certain offences	Up to $10 million or 2% of worldwide revenue
Breach record retention	At least 24 months	5 years

The pre-deploy checklist#

Run this before you connect a single customer record. If you cannot check a box, that is your next task, not a reason to skip it.

1. Consent script written for the call and a matching line in your privacy policy.
2. Data map listing every place a copy of customer data lives, including the vendor’s logs.
3. Retention window set — how long transcripts and notes live before deletion.
4. Deletion path you can actually execute when a customer asks to be forgotten.
5. Storage region confirmed (Canada vs US) if data residency matters to your clients.
6. Breach plan naming who reports, to whom, and where the 24-month record lives.
7. Subprocessor list from the vendor — who else can see the data, and can they train on it.

This slots directly into the broader roadmap for automating a small business without getting ahead of your own compliance.

Own the deployment, and privacy gets simpler#

Here is the wedge nobody selling a $99/month subscription will say out loud: with most SaaS AI tools, your customer data sits in their cloud, on their retention schedule, sometimes feeding their model training. You are renting the risk along with the software.

When you own the deployment, the transcripts and CRM notes live in accounts you control. You pick the region. You set retention. No third party quietly holds three years of your customers’ calls. It is the same reason ownership matters for what you’re actually buying — the data is yours, not a line item in someone else’s analytics. A hand-built AI Receptionist can run so no customer conversation ever leaves infrastructure with your name on it.

When this isn’t the right move yet#

Do not deploy AI against customer data if any of these are true. Fix them first.

• You handle health information and have no PHIPA or provincial health-privacy plan — clinics, therapists, dental. The bar is higher and the stakes are real.
• You can’t write an honest consent line because you are not sure what the tool collects. If you can’t explain it, you can’t consent to it on the customer’s behalf.
• Nobody owns the breach plan. If a breach happened tomorrow and you don’t know who reports it or where the record is, you are not ready.
• Your vendor won’t confirm storage region or training in writing. A vague answer here is the answer.

None of this is a reason to avoid AI. It is a reason to deploy it deliberately.

If you want to see what a compliant, Canada-aware setup looks like for your specific business, send the details through a free audit — it’s a short form, and I’ll reply with a plain-English data map and deployment plan within 24 hours. No call required.