AI in sales
Agentic AI in the sales playbook: start with account research
The first job I’d give a sales agent is deciding which accounts deserve a closer look. It’s a better starting point than asking it to send more emails.
I use AI in my own sales work, from prospecting and outreach drafts to organising my schedule. I review the language before it goes out. That last bit matters: my name is on the conversation.
For a company entering Canada, the useful question is where an agent can take work off a seller’s plate while making the account judgment better. A longer list, by itself, does neither.
Give it a commercial question
“Find Canadian companies” is a fairly expensive way to get a directory. I’d start with a narrower question: which Canadian operators have the problem we solve, and what public evidence makes a conversation worth having now?
Give the agent your definition of a good customer, the reasons an account gets ruled out, and a few examples you have already assessed. Ask it to return the evidence, the date, what it thinks that evidence means, and what it still cannot establish.
For example, a property operator announcing an acquisition could be worth researching. It does not follow that its operations team is unhappy, its software is inadequate, or a budget exists. Those are questions for discovery. They should stay questions.
Decide whether it needs to be an agent
An agent can choose its next step as it works towards a goal. A fixed workflow follows a route you have already specified. Anthropic makes this distinction in its engineering guidance, and recommends starting with the simplest approach that works.
If the task is pulling the same fields from the same approved sources, a fixed workflow may be enough. If researching an account leads to a subsidiary, a project announcement, and a different buying team, an agent has a more useful job. It still needs a stopping point. “Keep researching” is not a brief.
Make the Canadian assumption visible
I’d want the research to distinguish a Canadian head office from a Canadian operation. Who owns the purchase? Who will use the product? Is the local team allowed to choose a supplier, or is that decided elsewhere?
Ask it to flag uncertainty about implementation coverage, language needs, and where decisions sit. Don’t let a Toronto address stand in for a Canadian buying mandate. That distinction can change the account plan before anyone writes an opening line.
Review the accounts you know first
Run a small batch through the process, including accounts you would pursue and accounts you would reject. Read the linked evidence. Check whether the agent confused an owner with a manager, relied on an old announcement, or turned a possibility into a fact.
Then compare the finished research with how you would prepare yourself. Did it surface something useful? Did it miss the reason the account is a poor fit? How much correction did it need? Count that correction time when deciding whether the process helps.
I’d keep the seller responsible for prioritising the account and approving the outreach. A useful agent leaves you better prepared to have a conversation. If it leaves you fact-checking a beautifully written fiction, you’ve hired yourself a second job.
Next: the decisions to make before your sales agent hits send. For the account strategy underneath the automation, start with choosing your first Canadian target accounts.