If decisions about how to do the work live only in your head, AI for business will become another new starter you have to explain everything to personally. Start with your answer to a question your staff bring to you again and again: save the decision itself and explain why you made it.
“It takes me half a year to train them. They work for a year and leave.” That was how the owner of a company selling car parts described training salespeople at the first meeting. The service side was doing well, but sales of parts had fallen. When the owner got involved, things picked up. When he stepped back, the figures fell again.
If the team only sells when you are involved, a ready-made course will not free up your time yet. AI will prepare the training material, but you or the head of sales will still have to check how the salesperson applies it with a customer.
You can recognise a good use of AI by the question you no longer have to explain the answer to all over again. Let’s look at the mistakes that give owners more work when using AI, starting with the most costly: the company quickly spreads the wrong way of working.
Mistake 1. Asking AI to invent your business rules
Write down what a salesperson is allowed to promise a customer and how they must act when there is a problem before asking AI to write instructions. Delivery terms and working rules come from company decisions — otherwise a new starter will learn from the tool’s guesses.
For example, you ask it to put together training for a salesperson. It produces a tidy programme: identify the need, present the benefits, handle objections. A new starter reads it, then comes back with the same old question: can they give the customer a delivery date if the supplier has not confirmed anything yet? There is no answer in the programme. But there is a section on sounding confident.
The cost of this mistake is promises you then have to keep. The customer is already counting on the delivery, the salesperson points to the instructions, and the owner is once again sorting out an exception personally. The faster you hand those instructions out across the team, the more identical problems you will get.
Start with a specific decision about the work. For example, the company has decided that, until the supplier confirms, the salesperson tells the customer when they will come back with an answer about timing and does not present a delivery date as agreed. You can give this rule to AI with an example of a correct reply. It will prepare an explanation for the new starter and several practice situations. The team leader will check that no new promises have appeared in them.
For a one-off draft like this, a chat tool that can work with files is suitable. For example, Alice AI answers questions about an uploaded document and highlights its main points — this is described in its help on working with files. But uploading a document does not in itself make every answer correct: you or the head of sales still have to check it against the rule.
Before trying it, remove names, phone numbers and details of specific customers from the material. For an exercise about delivery times, the rule itself and an anonymised conversation are enough. There is no need to transfer your entire working database just to get the first draft.
For the owner at the start of the article, each round of training took up more of his time. A saved explanation will not keep an employee at the company. But it will give the next new starter material they can begin with, without another lecture from the owner.
Mistake 2. Treating a polished report as proof that the work is done
You read a polished AI report and decide that the work is done. Yet the salesperson may have failed to keep a promise to a customer — and you will not even know. Check the facts behind it: what the employee has finished and what remains to be done.
Imagine your salesperson tells you in the evening that they worked with customers, discussed terms and sent proposals. Once processed, this becomes a businesslike paragraph. It is more pleasant to read, but you still do not know who was promised a quote and who has been waiting for it since yesterday.
Do not ask the employee to sound more formal. Ask them to name the action they completed, its result and what remains to be done. For example: the proposal was sent, the buyer requested a different configuration, the quote is not ready yet. Now there is something to examine. If AI describes this deal as agreed, you can spot the mistake in the original message.
Give it a narrow task: gather the results for each employee, list unfinished work and questions for the team leader separately. Keep the original message or a link to it beside each conclusion. If there is no deadline, the report must leave it that way: the employee did not give a deadline. Tomorrow, the team leader will clarify it instead of relying on an invented date.
Agree separately on what the figures mean. What do you count as a sale: the customer’s agreement, an invoice issued or payment? When employees mean different things by the same word, a combined report only hides the discrepancy. AI gets the counting rule from the company; the monetary total is checked against the accounts.
For regular collection, RENTROP has a separate service, AI Report: it takes reports by voice or text, puts them into a businesslike format, compares shifts and reminds staff to submit them. It is useful where the team leader manually collects employees’ messages every evening; you still need to define what counts as completed work.
Mistake 3. Expecting sales analysis to get the customer to pay on its own
I would only consider a sales review useful when the salesperson knows whom to reply to and what to do next. AI can find a mistake in a conversation, but the sales team takes the deal through to payment.
You may have an excellent analysis of a message exchange: the buyer asked for a quote, the salesperson sent a catalogue, the conversation went quiet. If all the salesperson gets afterwards is a comment on the quality of their communication, the buyer is still waiting for a quote. The company has learnt to describe a lost sale in detail, but has not yet done anything about it.
That is why, when choosing a tool, you should ask to see the result on your own task. To find stalled deals, you need a specific customer and a question that went unanswered. To train a salesperson, you need an extract from the conversation and an explanation of where they went wrong. General advice to improve communication does not give you a way to start work.
Distinguish between the data sources too. A single message cannot tell you how a conversation ended in another channel. If the customer replied by phone but the service only sees the chat, you will have to double-check any conclusion that the buyer has gone silent. Before connecting the tool, find out which calls, messages and records will be included in the analysis.
When you need to see every day where money is stuck, AI ROP — a service that increases sales by 20–30 % from the first day — analyses calls and messages alongside the deal record, shows whom to reply to, which deals to take through to payment and which to recover. More about how AI ROP works in a sales team.
We look at sales work in detail in a separate article: what to ask AI to do in a sales team. Here, there is a simpler test for choosing a tool: after seeing the analysis, it should be clear what work an employee can start without piecing together the customer’s entire history again.
Mistake 4. Judging a candidate by how articulate they are
Candidates should be compared on how they solve a practical work task, so it makes sense to build an AI interview around one. Ask what the person would do with an enquiry from your customer: an account of their strengths will not show you that.
For example, you need a salesperson who knows how to clarify an enquiry before preparing a quote. Give the candidate an incomplete customer enquiry and ask them to continue the conversation. See what extra information they ask for. Confident speech alone will not tell you whether the person will pass an order on for a quote when half the terms are unknown.
Before the interview, the team leader writes down the actions they want to see. For one team, it is important to establish product compatibility first; for another, it is deadlines and the scope of work. AI can compare the answers with these requirements and highlight questionable points. But if there were no requirements, it will be unclear what a high score is for.
It is even more useful to link selection with induction. A question the candidate stumbled over becomes the first topic of training after they are hired. Instead of the same introductory lecture for everyone, the new starter gets an exercise on a specific gap in their knowledge. The head of sales then checks whether they were able to apply the rule at work.
If you need this sequence, RENTROP has a separate AI recruitment service, Robot Recruiter: a page for the vacancy with a chat adviser, an interview following a script for the role with workplace scenarios and assessment of the answers, and training to help the new starter settle in.
Let’s return to the cost of training salespeople. Once the owner’s explanation is saved, it can be used to prepare a task for both a candidate and a new employee. A person still makes the hiring decision, but each conversation no longer has to start from scratch.
Mistake 5. Starting with a large rollout instead of one recurring question
You can test the first benefit of AI on one explanation you already give your staff. The trial will result in a short exercise that the next person can use to work things out and show that they understand the rule.
Open yesterday’s messages with the team. Find a question you hear regularly: what to clarify with the buyer, when to come back with an answer, who to pass an unusual request to. Take your answer and check whether it includes the reason for the decision. Simply telling someone to call the supplier is not enough: a new starter needs to understand what information to get and what to tell the customer afterwards.
If the reason is still in your head, add it to the answer in plain words. Then give the anonymised material to AI with this task:
Using this explanation, prepare a short exercise for a new employee. First describe the situation, then ask a question about the next action. Put the correct answer and its explanation separately. Add a second version where there is not enough information to make a decision: the employee must notice this and say what needs clarifying. Take all rules from the original explanation. If it does not give you enough to put together an answer, ask me a question first.
The head of sales checks both versions against your decision. They then give the employee only the situations and questions, without the answers. The employee must name the action and explain why. If they guess the action but do not understand the reason, the exercise has not yet replaced the explanation and needs more detail. If, in the second version, they start inventing the missing conditions, that is exactly what you work through.
Save the accepted version where new starters already get their materials and record the date of the rule. Next time, the head of sales will give them this exercise first and look at the answer. Any specific point that remains unclear will come to you.
That gives the company its first piece of training that will outlast an employee’s departure. You can give it to the next person tomorrow morning.