Understand

AI agents for sales

In a sales team, most of the time lost is not spent in meetings but between them: requalifying an enquiry, tracking down where a quote stands, updating a record, following up at the right moment. That is exactly the work an AI agent takes on — not the customer relationship itself, but everything that makes it possible.

1. What an agent takes on

Four tasks recur in almost every team, and delegate well because they follow a statable rule and can be checked at a glance.

Qualify incoming enquiries — read what comes in, extract the need, the stated budget and the deadline, set aside what clearly falls outside your market, and present the rest ranked by priority.
Follow up on unanswered quotes — watch what went out, spot what stays silent beyond the delay you set, and propose a written follow-up in your tone.
Keep the customer record current — after each exchange, log what was said, the next step and its date, so the record stops being an approximate memory.
Prepare meetings — gather the history, previous exchanges and points left open, and hand them to whoever is taking the call.

What these four tasks have in common: none of them calls for commercial judgement, only for consistency. Which is precisely what a human does badly with twenty other things in mind.

2. What should not be handed to it

Three things stay on the human side, and not for technical reasons.

Negotiation. Granting a discount, arranging a payment schedule, arbitrating between two requests: these are commitments. They must go through human approval because they bind the company, not because an agent would be incapable of them.

The first contact on a major account. An agent writing in place of a director on a strategic deal saves ten minutes and can cost the business. The test is not difficulty, it is what a clumsy move would cost.

What nobody has put into words. If your qualification rule exists only in your best salesperson's head, the agent will not guess it. Writing it down is the real preparatory work — and it serves the human team anyway.

3. The follow-up case

Following up is the best first project, because it meets all four criteria of good delegation: it recurs constantly, its rule fits in one sentence, its result shows immediately, and a mistake is undone with a word of apology.

In practice the agent watches the quotes that went out, applies the delay you set, drafts the follow-up and submits it to you. You approve it, or you let it go without approval once the quality suits you — that is the autonomy boundary, and it moves as trust settles, never before.

The real gain is not writing faster: it is that no quote is ever left without a follow-up because the week was busy.

4. Measuring the result

Three figures are enough, and they must be read together.

Coverage — the share of enquiries handled within the deadline you set yourself. That is where the effect shows first.
Human rework time — how many minutes a day the team spends correcting the agent. If it does not fall, the rule is badly worded: that is what needs revisiting, not the agent.
Spend — the cost of the models consumed, measured against the work completed. It is capped per agent and per period.

A system that handles everything but demands systematic review automates nothing: it shifts the burden from the salesperson to the reviewer.

5. Frequently asked questions

Can an AI agent replace a salesperson?

No, and that is not the aim. It takes back the administrative half of the job — tracking, following up, updating, preparing — to give the salesperson back the time for conversations. Selling remains a matter of judgement and relationship.

Do you have to change CRM to use AI agents?

No. The agent plugs into the tools already in place and works them as a person would. Changing tools at the same time as introducing agents means running two projects at once, and no longer knowing which one failed when something goes wrong.

Will customers see that they are dealing with an agent?

That depends on what you decide. In the most common setup the agent prepares and a human sends: the customer receives a message from their usual contact. If you let the agent write directly, honesty calls for saying so — and several regulations now require it.

Which task should you start with?

Following up on unanswered quotes. Its rule fits in one sentence, the result shows within days, and a mistake there is harmless. The other tasks are added once this one runs unsupervised.