AI agents in prospecting: automating without becoming spam

LinkedIn feeds are full of vendors promising an "AI salesperson" that sends a thousand personalized emails a day. For a Quebec SMB considering a custom AI agent, the pitch is tempting: prospecting is the work everyone postpones to Friday, and automating it sounds like the definition of a good investment.

The problem is that these playbooks almost all come from the United States, where the law lets you email first and apologize later. In Canada, the anti-spam legislation (CASL) flips the logic: no consent, no email. Importing the American manual here, with or without AI, means industrializing a violation.

This article draws the line we apply ourselves: automate the research aggressively, control the sending carefully, and turn the legal constraint into a competitive advantage.

The short answer, for the busy

Canada's Anti-Spam Legislation (CASL) governs every commercial email sent to an address in Canada, including messages written or dispatched by an AI agent. Three obligations apply: obtain consent (express or implied), identify yourself clearly, and provide a working unsubscribe mechanism, honoured within 10 business days and kept functional for at least 60 days after sending. Unlike the United States, where CAN-SPAM lets you email first under an opt-out regime, the Canadian rules are consent-based: the 1,000-cold-emails playbook imported from a US guide is a potential violation here, with penalties of up to $10 million per violation for a business. The profitable and legal way to use an AI agent for prospecting in Quebec: automate the research, enrichment, and drafting, and keep the sending under human control, backed by a documented consent basis, never a purchased list.

Two automations with nothing in common

Prospecting splits into two halves that tool marketing deliberately blurs. The first half is research: identifying companies that look like your best clients, spotting buying signals (a job posting, an announced move, a website redesign, a new executive), enriching each record with public information, then drafting a first message informed by all of it. That is repetitive intellectual work, high value and low risk. No message goes out, so CASL is not even in play. It is exactly the kind of task an AI agent excels at, like the other AI agent use cases for Quebec SMBs we have documented.

The second half is sending. The moment a commercial email leaves for an address, the law applies, whether the text was written by a human, a language model, or an intern. And automated sending multiplies precisely what the law counts: each email is a separate potential violation. An agent that sends 1,000 non-compliant messages does not make one mistake. It makes 1,000.

That is the honest line the "AI SDR" vendors leave out: automate the research without restraint, automate the sending with extreme caution.

What CASL actually requires

For every commercial email, the law imposes three obligations, laid out in the CRTC's documentation.

First, consent. It can be express (the person took a positive action: "yes, write to me") or implied (the law infers it from specific situations, more on this below). Express consent never expires unless withdrawn, but it must be provable: the burden sits with the sender, not with the recipient. A pre-checked box does not count as a positive action.

Second, identification. The message must say clearly who is sending it and on whose behalf, with contact information that stays valid for at least 60 days after sending.

Third, unsubscribe. Every message needs a simple opt-out mechanism, functional for at least 60 days, and any request must be honoured within 10 business days. It is the dumbest obligation to fail, and it is the one that cost Hudson's Bay $120,000 in 2024.

Implied consent: the real room to manoeuvre in B2B

The Canadian regime would be unworkable for business development if it demanded express consent before any first contact. It does not: implied consent exists, and it is built for serious prospecting.

First source: the existing business relationship. A client who bought or leased something within the last two years, or who made an inquiry within the last six months, can receive your commercial emails without ever ticking a box. A written contract still in force, or expired less than two years ago, counts too.

Second source, and this is the door for targeted B2B: conspicuous publication. If someone publishes their professional email address (on their company's website, in a professional directory) without a statement refusing commercial messages, you may write to them, on one condition that changes everything: the message must be relevant to their role. Writing to a manufacturer's director of operations about operations software is defensible. Scraping 10,000 addresses and pushing the same offer to everyone is not, because relevance does not survive volume.

As for purchased lists: the vendor swears the contacts are "opted in"? In front of the CRTC, you are the one who has to prove it, with evidence you do not have.

Read the intent behind this architecture. CASL does not ban prospecting. It bans laziness: it makes the researched, targeted, relevant contact perfectly legal, and mass spraying expensive.

What it costs when it goes wrong

The administrative monetary penalties in the law go up to $1 million per violation for an individual and $10 million for a business. Those are ceilings, not price tags, but enforcement is real: in June 2024, Hudson's Bay paid $120,000 under an undertaking with the CRTC over emails with a defective unsubscribe mechanism. In 2023, an individual received a $40,000 penalty for violations of section 6, the one that governs commercial messages. The CRTC publishes its decisions: appearing on that page is a kind of publicity nobody wants.

And there is a second bill, one the law does not send but Google and Microsoft do: deliverability. Mailbox providers measure complaints and bounces. A mass campaign without consent degrades your domain's reputation, and then it is your quotes, your invoices, and your project follow-ups that land in the junk folder. Spam is a debt the whole domain repays.

What a compliant prospecting agent looks like

Once the line is drawn, the requirements for a useful agent almost write themselves.

It watches signals: new companies in your niche, telling job postings, leadership changes, aging websites. It enriches each prospect with the relevant public information and prepares a briefing a human reads in 30 seconds.

It keeps the list like a legal register. Every contact carries its consent basis (express, business relationship, conspicuous publication) and its expiry date. This is the detail manual processes always drop: implied consent expires, six months after an inquiry, two years after a purchase. Tracking those clocks by hand is unrealistic. For software, it is trivial. Compliance becomes a data problem, and data problems get solved.

It drafts, it does not press "send". Every first contact goes through human review, at a volume low enough for the review to be real. It also processes unsubscribes immediately and keeps the proof of consent.

The most interesting part is the side effect: a system forced to aim well performs better. Twenty informed, relevant emails a week generate more conversations than a thousand generic ones, and reply rate is the only metric that pays. The same foundation serves elsewhere in the business too, from triaging inbound email to customer service.

Where to start

If prospecting is eating your hours or, worse, not happening at all, the first step is not buying a mass-sending tool. It is mapping your consent situation: who your clients and past clients are (business relationship), who wrote to you recently (inquiries), and which public segment your offer genuinely concerns. An agent gets built on that foundation.

That is the kind of scoping we do in a first conversation: identifying where automation creates value in your business development, and where it would create risk. 30 minutes, no commitment.

→ First conversation, no commitment

This article explains Canada's anti-spam legislation to help an SMB ask the right questions; it is not legal advice. How CASL applies depends on the facts of each situation: the implied consent categories, in particular, carry precise conditions. Consult a lawyer before launching a campaign, and rely on the official CRTC and Government of Canada texts.