How to Train Your AI Salesperson: Complete Guide for 2026
How to train an AI salesperson: map the process, give it your own material, set limits, test on real conversations and review every week.
Table of contents
How do you train your AI salesperson to become useful?
Training an AI salesperson is not the same as training an employee. An AI does not learn by being around for a few weeks. It does what it has been told to do, with the material it has been given access to. Training therefore means four concrete things: describing the task precisely, giving it your own material, setting limits on what it may and may not do, and testing it on real conversations before a customer meets it.
It is rarely the technology that fails. It is the setup. A company switches on a tool, gives it a generic script and wonders why it sounds like everyone else. A new salesperson would never be allowed to start like that, and neither should an AI salesperson.
TL;DR:
- An AI salesperson is only as good as the description, material and test conversations it is built on.
- The order is: product knowledge, objections, conversation flow, limits, testing. Not the other way round.
- It does not get better by itself. Someone has to read the conversations and correct them, on an ongoing basis.
- Start with the AI drafting and a person approving. Only give it more independence for what you have seen it get right many times over.
What is the difference between training an AI salesperson and a human one?
A new human salesperson forgets some of what they learn, reads the mood in the room and can ask a colleague. An AI does not forget what it has been told, but it guesses when the instruction is imprecise, and it does not ask on its own if it is unsure. That is why it needs more precise instruction than a person, not less.
| Human salesperson | AI salesperson | |
|---|---|---|
| What "training" is | Onboarding, coaching, experience built over months | Description, material, rules and test conversations |
| Who corrects mistakes | The salesperson, after feedback | Whoever owns the setup, once the conversations have been read |
| What it does without instructions | Uses judgement | Guesses, and guesses with confidence |
| Consistency | Varies from person to person | The same rules every time, if the rules are written down |
| When it is ready | When the manager thinks so | When it has passed your own test conversations |
The point in the last row is the most important: you decide when it is ready, and you decide that on your own examples, not on a vendor's demo.
How do you prepare the company before you even open a tool?
Before you configure anything, the foundation has to be in place. An AI salesperson reproduces what is in the material. If the material is messy, so will the answers be.
Step 1: Map your real sales process
Write down how your best salespeople actually work: How do they open? Which questions do they ask? How do they respond when the customer says "send an email" or "we already have a supplier"? The best source is your own conversations. Listen to them, or read transcripts of them if customers have been told the call is recorded. What salespeople say they do, and what they actually do, are rarely the same.
Step 2: Describe who you call or write to
The one speaking for you must know who it is talking to: which industries, which sizes, which roles, which problems you typically solve, and what makes a moment right. The more concrete you are, the less generic the result will sound.
Step 3: Gather your own knowledge base
Give the AI access to what you yourselves stand behind:
- Product and service descriptions
- Answers to the questions customers keep asking
- Your best objection responses, written from real conversations
- Examples of past customers you are allowed to name
- Rules for what may be said about prices, delivery times and guarantees
Without that material the AI answers from general knowledge. With it, the AI answers on your behalf.
What should the instructions contain?
The instructions are what make the AI salesperson yours. Five things must be in there:
- Who it is. Role, tone and style of language. Danish, your own professional vocabulary and the courtesy you yourselves use with customers.
- What a conversation looks like. Opening, uncovering needs, proposal, objections, agreement on the next step.
- Which rules apply. What it may promise, what it may not promise, and who must always be asked first.
- When it hands over to a human. Upset customers, legal questions, prices outside the list, anything where a wrong message could cost something.
- What it must never say. Price guarantees, promises about delivery, negative statements about competitors.
Keep each part short and concrete. Long, messy instructions give answers just as messy as long, messy job descriptions give to people.
How do you test an AI salesperson before it meets a customer?
Testing is the most important step, and it is also the one most people skip. The method is simple:
1. Write test conversations based on reality. Take the conversations you have had, and the objections you meet most often. Use real wordings, not polished, invented ones.
2. Cover the difficult types. For example:
- The busy one: "I only have a minute."
- The sceptic: "We have tried AI before, and it did not work."
- The one who does not make the decision.
- The one who asks for material to get you off the phone.
- The one who says: "Stop contacting me."
3. Try to make it fail. Ask impossible questions. Interrupt. Change the subject. Ask for a discount. The failures you find now are the ones the customer would otherwise find for you.
4. Give each test conversation a grade. Use a simple scorecard: Is the tone right? Is the content correct? Does it follow the rules? Does it hand over at the right moment? Compliance is passed or failed, not "almost".
5. Correct, and run the tests again. When a change lifts one conversation and drags another down, you only see it if you run all the tests every time. Save them so they can be reused.
There is no magic number of test conversations. The number depends on how much is at stake. If the AI sends a draft which a person reads, it needs fewer tests than an AI that writes to customers itself. And an AI that can promise something that costs money needs the most.
What are the most common mistakes when training an AI salesperson?
Mistake 1: A generic script
A copied sales script gives you a generic AI salesperson. It needs your industry, your customers and your offer.
Mistake 2: No rules for legislation
Marketing by telephone, email and SMS is regulated, including under the Danish Marketing Practices Act and the GDPR. An AI that contacts people without a clear basis is your responsibility, not the vendor's. See the section on rules below.
Mistake 3: Too few tests, and too polished
If your test consists only of conversations that go well, you have not tested anything.
Mistake 4: No ongoing follow-up
The product changes, the prices change, and customers find new objections. A setup that was right in January can be wrong in June if nobody reads the conversations.
Mistake 5: No connection to your systems
An AI salesperson that books a meeting but does not put it into your calendar and your customer records creates more clean-up than value. What happens in the conversation must end up in the same places as the rest of your sales work.
What does ongoing improvement look like?
An AI salesperson is not a one-off project. A simple weekly rhythm works well:
- Monday: Read a handful of conversations from last week. Where did it go wrong? Which objections came up most often?
- Tuesday and Wednesday: Correct instructions and material where the errors occurred.
- Thursday: Run all test conversations again and compare with last time.
- Friday: If it has improved, put the change into use. If not, roll it back.
The most important thing is that one person in your company has responsibility. Not because the task is heavy, but because nobody else reads the conversations.
How do you give the AI independence in a safe way?
A good principle is that you decide what the AI may do on its own, and that everything else waits for approval. In practice you can step it up:
- Drafts. The AI writes proposals for emails and notes. The salesperson reads and sends them.
- Defined tasks. The AI answers the questions you have said yes to on its own, and passes the rest to a colleague.
- Expansion. Only once you have seen it handle a task correctly many times in a row do you give it more.
It is the same path you would follow with a new graduate: first with side-by-side training, then independent on the simple things, then on more.
Which rules must you stay within?
Three frameworks are worth knowing before an AI contacts customers:
- GDPR. You must have a lawful basis for processing personal data, tell the data subjects what you are doing, be able to delete data on request, and have data processing agreements with suppliers who handle the data. Read the guidance from the Danish Data Protection Agency at datatilsynet.dk.
- The Danish Marketing Practices Act. It regulates when and how you may contact consumers and businesses with marketing, including by telephone, email and SMS. The Danish Consumer Ombudsman has guidance at forbrug.dk.
- The EU AI Act. The EU's rules on AI set transparency requirements, including that people must know they are talking to an AI.
The limit on what an AI voice may say in a sales call to a company is not settled enough for us to guess at it here. Have a lawyer assess it before you send anything live.
What can Agent360 do for you?
Agent360 builds AI solutions inside your own systems. We call it "Your AI department": a task you solve by hand today is built on top of the systems you already use, with you as the approver. The price depends on the task and is agreed before we start.
For sales work there are two task types many people will recognise:
- After the conversation: Notes, fields in the customer register and a draft follow-up, so the salesperson spends time on the customer instead of on paperwork. See notes and customer register.
- The phone: Calls taken, messages gathered and appointments put in the calendar. See phone. Voice agents are built by our partner JesperAI.
If you want to be able to train, correct and build on your own once we have left, there is also a personal AI mentoring programme. Examples of tasks we can help with are on use cases.
Frequently asked questions
Q: How long does it take to train an AI salesperson? It depends on how messy your starting point is and how much is at stake. The setup itself can be done quickly, but writing down your material and gathering good test conversations takes time. Do not expect it to be sorted in an afternoon, and plan time for the first few weeks of ongoing corrections.
Q: Can an AI handle objections as well as a human? On the objections you know and have written good answers to, it can answer consistently. On new, emotionally charged or unclear objections it should hand over to a human. Test with your own objections, not the demo's.
Q: What happens if the AI says something wrong to a customer? That is why you set limits, let it hand over when in doubt and read the conversations. If you start with drafts that a person approves, the error never reaches the customer.
Q: Should we fine-tune a model or use instructions and our own material? For most companies, clear instructions and access to your own material are both simpler and easier to correct than adapting the model itself. Fine-tuning only makes sense once you have many good conversations and a concrete problem that the instructions do not solve.
Q: Is it the same for B2C and B2B? The method is the same, but the rules are stricter for consumers. The Danish Marketing Practices Act sets stricter requirements for contact with private individuals than for contact with businesses, and more is at stake if something goes wrong.
Q: Can we use the same method to train our human salespeople? Yes. Test conversations, scorecards and a weekly review work just as well for coaching people.
Q: What is the difference between Agent360 and JesperAI? Agent360 builds AI solutions in the customer's own systems and offers a personal AI mentoring programme. JesperAI is a partner that builds voice agents.
Next steps
Three concrete steps you can take this week:
- Write down your sales process, with the real wordings from your best conversations.
- Gather ten to twenty of the hardest conversations and objections you have experienced. They will be your first test.
- Decide who owns the follow-up. Without an owner, no setup stays good.
Book 20 minutes of free consultation with Agent360 and tell us where sales work eats up your time. We will also tell you if the task is not worth building.
Also read: How to onboard your first AI salesperson and The Bionic Sales Rep: humans and AI in the same workflow.
Sources
Contents
Should your sales reps spend their time talking?
We have built it for our own sales: the call is transcribed, the AI suggests the outcome, and the email is ready. We look at how your sales run today.
Book a reviewShould your sales reps spend their time talking?
We have built it for our own sales: the call is transcribed, the AI suggests the outcome, and the email is ready. We look at how your sales run today.
How we built it · Sales & follow-up
Sales cockpit with AI
A call list for salespeople. The call is transcribed in Danish, AI suggests the outcome (e.g. “Call again”) and drafts the email based on what the customer said.
- The salesperson calls
- AI writes transcript and email
- The salesperson sends
Related articles

Own or rent software? A decision guide for Danish businesses
Own or rent software? Cost over time, data lock-in, GDPR and exit, self-host and fair source, plus a concrete decision framework for Danish businesses.

Claude Code in a business: what an AI developer agent can do, and what it requires
What Claude Code, Anthropic's agentic developer tool, can do in a business, and what it requires in security, governance and organisation.

AI strategy for businesses: a step-by-step guide
A practical guide to an AI strategy: map workflows, choose use cases, fix your data, run a pilot, bring staff along and know when not to automate.