An AI agent for sales takes over the administration around selling: preparing for meetings, drafting follow-ups, updating notes in the CRM and pointing out deals that have gone quiet. The relationship, the pricing and the decision to send stay with the salesperson. Start with one task that recurs every week, for example meeting preparation, and let a salesperson read every result before it is used.
What an agent can do in sales
- Meeting preparation. Collect what you know of the customer on one page: earlier emails, quotes, notes and what was promised last time.
- Follow-up. Write a draft after the meeting, with what was agreed, that the salesperson reads, edits and sends.
- CRM notes. Move the key points from a meeting or a call to the right place, so nobody has to write them twice.
- Deals that have stalled. See which deals have been still for too long and remind the person who owns them.
- Quote groundwork. Prepare a quote from information that already exists, which a salesperson goes through and sends.
What the salesperson keeps
- The relationship. The conversation with the customer, the trust and the feel for when it is time to get in touch.
- Price and terms. What you offer, at what price and on what terms, is decided by a person.
- The final read. Nothing goes to a customer before a salesperson has read and approved it.
- The selection. Who should receive an offer and when is a judgement, and the salesperson makes it.
A Monday morning with an agent
Picture a salesperson with five meetings during the week. On Monday morning five pages are ready, one per meeting, with what was said last time, which questions the customer usually asks and which documents were sent. She reads them with her coffee and corrects what is wrong.
After each meeting a draft follow-up is ready to read. The time she used to spend searching and writing goes into the conversations with customers. The decision on what is sent stays with her the whole time.
Start with one task
- 1. Choose one task. One that recurs every week and annoys the most, for example meeting preparation.
- 2. Decide what the agent may read. Only what the task requires. Do not give access to the whole customer register if a list is enough.
- 3. Decide where the salesperson reads. A stop point where a person sees the result before anything is sent or changed.
- 4. Run it for a week. Let two salespeople use it and note what needed correcting.
- 5. Look at the result. Adjust the task or add the next one. Move on only when the first step works.
Customer information and the rules
Customer information is often personal data. Then you need a data processing agreement with whoever handles it, and you need to know which information the agent may see. Write it down in your AI policy.
Marketing and contact with customers are covered by rules. Let the agent prepare and let a salesperson send. Ask a lawyer or your data protection officer what applies to you.
Keep a list of which customer information the agent may see and where it ends up. The list helps you answer if a customer asks, and it shows when something needs to change.
How to measure whether it works
Measure three things during the first month: how much time each salesperson spends on administration each week, how often follow-ups are sent on time and how often a draft has to be rewritten completely. If the three move in the right direction, the agent is worth building on.
Ask the salespeople themselves after the first month as well. They see quickly whether the preparation helps in the meeting and whether the drafts sound like them. What they say shows what to change in the next step.
What makes a good first task
A good first task recurs every week, has a clear result and can be checked quickly. Meeting preparation meets all three: the result is one page, the salesperson sees at once whether it is right and a mistake costs little.
Follow-up drafts are the next step. They reach a customer and therefore need a more careful read before anything is sent.
Common mistakes
- Starting too big. An agent that should run the whole sales process is hard to control. Start with one task.
- Giving too much access. Give the agent what the task requires. The whole customer register is rarely needed.
- Skipping the stop point. Drafts that are sent without reading make mistakes visible to the customer.
- Forgetting the notes. If the notes in the CRM are thin, the preparation will be thin too. Let the agent help fill them.
- Measuring too late. Write down how much time the administration takes before you begin, or there is nothing to compare with.
How we work with this
We call it Agentic Operations. We map how the work flows and build the agents that take over the manual work between your systems, on top of the systems you already have. The price is fixed per step, with a cap, and is set before the step begins. We sign a data processing agreement before we touch any information.
Common questions
Can an AI agent send emails to customers by itself?
Technically it can, but we recommend that the agent prepares and a salesperson reads and sends. Then the responsibility for what the customer receives stays with you.
Which CRM do we need?
Usually an agent can work on top of what you already use. We first map how you work and which systems are involved.
What does it cost?
The price is fixed per step, with a cap, and is set before the step begins. Read more in the guide on what an AI agent costs for a company.
How do we protect customer information?
By giving the agent access to what the task requires and nothing more, through a data processing agreement and by having a person read the result before anything happens.
Does an agent replace the salespeople?
No. It takes over the administration so that the salesperson has more time for customers. The relationship and the decisions stay with a person.
How quickly can we start?
A first step is one task with a stop point. You can start small and see how it works in daily life before you decide on the next step.