AI in sales helps analyse calls and written conversations, find delays in deals and mistakes by salespeople, and compare conversations with data in the CRM system. For daily use, you need access to the right sources, your team's rules and a head of sales who decides what to do with the findings. You should start with one task where you can measure the benefit and the time the team spends.
A business owner wanted to stop listening to calls in the evenings. He set up analysis: now, in the morning, he has a list of issues; by lunchtime, explanations from the head of sales; and in the evening, an argument with a salesperson over yet another conversation. The software listens to the recordings, but the owner's day is still taken up with sales. This is what an implementation trap looks like. We will examine it through an illustrative example with fictional data.
The first explanation seems obvious: a poor choice of AI tool. But suppose the issues are valid. A buyer really did receive no answer; a salesperson really did interrupt; in another conversation, they really did fail to establish what the customer needed. The more accurate the analysis, the longer the list. The team now has more reasons to step in than it has time to deal with.
I believe that buying AI should start with deciding how the head of sales will use the time it frees up. If they continue to review every issue individually, the queue of reviews will eat up the time saved on listening. Below are the tasks worth giving to the software and ways to organise the work so this does not happen.
AI for sales: what work to give it
People use the same term to look for different things: an assistant for working on a single text, a tool for analysing conversations and a system that regularly reads the team's data. It is easier to choose between them by considering the result an employee needs to receive.
- Analyse what was said in a conversation. Identify the buyer's question, the salesperson's answer and where the answer did not address the question. For example, the customer asks about equipment compatibility, but the salesperson keeps listing the configurations. The head of sales needs this extract for training, along with an explanation of the mistake and a suitable answer.
- Find a deal that needs action. Compare the deal record with the communication: who was promised a reply, where a question remains unanswered and which buyer it is time to contact again. A useful result is a specific deal with a reason to act. A list of all records created a long time ago does not do this job.
- Prepare material for a conversation with a salesperson. Collect examples of their work so the head of sales can discuss an observable action. Instead of describing them as “inattentive”, show customer questions the employee did not answer and examples of how the conversation should have continued.
- Link communication to sales figures. Look at response times, stages, salespeople and revenue by source. This helps you choose an area to examine: where there are many enquiries but few lead to payment. Then look for the cause in specific deals.
These are different tasks. If an employee needs a one-off analysis of a few anonymised texts, preparing the material for the chosen tool is enough to start with. If they need a daily list of deals requiring attention, uploading data by hand becomes an ongoing job. When choosing a service, it then matters whether it receives up-to-date data from your CRM and communication channels.
You should not give AI everything labelled as analysis. A CRM filter can display records with an empty field. Working out whether a salesperson answered a buyer's question requires analysing meaning. There is no reason to buy a second tool for a filter you already have.
A sample instruction for analysing a conversation
For a first trial, the head of sales can use the following wording. This is an instruction for analysis, not a promise that every AI tool will produce the same result.
Analyse the attached conversation together with the deal history. State what the customer wants to achieve and which question remains unresolved. Show the extract on which your conclusion is based. State separately whether this deal needs action now and which part of the conversation should be discussed with the salesperson during training. Apply only the attached team rules. If there is not enough information to reach a conclusion, state what is missing. Do not attribute intentions to the buyer that they have not expressed.
Along with the instruction, you need a recording or transcript, the related deal record and the applicable company requirements. The head of sales checks the conclusion against the original material before speaking to the employee. The article what AI sales analysis reveals covers data errors and false alarms in detail. Here, we will go further: suppose the findings have already been confirmed — how should the work be allocated?
AI in sales: the company decides what is urgent
An issue in a conversation and a reason to step in immediately are different things. A salesperson interrupted a customer, then apologised, established what the customer needed and agreed on the purchase. This is useful for training. In another deal, the salesperson is polite, but a promised reply is late and the buyer is choosing a supplier today. This calls for action on a specific order.
If both cases are sent as identical alerts, the head of sales will start with whichever catches their eye first. The order of the rows will determine what gets done first. So the owner needs to approve the criteria for urgency before daily notifications are switched on.
A practical way to separate the cases is:
- Act on the deal now. A commitment has not been met, or there is an open question on which the buyer's next decision depends. The head of sales assigns the salesperson an action and a deadline.
- Review repeated mistakes during training. One salesperson repeats a mistake in completed conversations, but there are no commitments to customers in these cases that require urgent action. The head of sales brings them together into one session to review the skill.
- Get a decision from outside the department. The salesperson has replied, but the company has not decided whether to accept non-standard terms. The head of sales sets out for the owner what needs to be decided, the acceptable options and the deadline for a reply.
The head of sales separates the cases. Having analysis does not in itself mean that the service allocates tasks to people or controls how they spend their time.
A large amount in a deal record should not automatically make an issue urgent either. If the buyer has already agreed a pause, a demand to call immediately will break that agreement. Priority should be based on what is happening in the sale: what the company needs to do and when it must act to still influence the customer's choice.
Thirty issues: how much work is that for the head of sales?
Let's continue the illustrative example. All the numbers below are invented to calculate the workload; they are not a client's results or a standard for the team.
In a day, the analysis identified 30 confirmed issues. The company decided to discuss each one separately. Preparation and a brief review of one issue take four minutes. That makes 30 × 4 = 120 minutes of the head of sales' working day. This does not yet include the salespeople's actions after these conversations.
Let's look at what makes up the list:
- 24 issues relate to conversations that have already ended. Three employees each repeat one mistake, with eight instances per employee. There are no urgent commitments to buyers in these cases.
- Another six issues relate to six open deals where the buyer is waiting for a specific answer. These need six separate actions.
For the first 24 cases, the head of sales prepares three review sessions — one with each employee. In each, they choose a representative conversation, explain the required action and have the salesperson apply it to another example. Eight repetitions show that the mistake is persistent; eight separate conversations about it are not needed.
Let's allow ten minutes for each such review and five minutes for a decision on each urgent deal in our illustrative calculation. That gives us 3 × 10 + 6 × 5 = 60 minutes. This is a proposed time budget that needs to be measured during the trial. If reviewing the skill takes longer, the actual duration goes into the calculation.
The difference comes from grouping repeated mistakes. The number of issues stayed the same, while the number of separate decisions fell from thirty to nine. At the same time, a separate action was set for each of the six waiting buyers.
This is a boundary that must not be lost for the sake of a convenient report: the same mistake in six active deals requires work with all six customers. A general lesson for the salesperson does not replace answers to buyers. Training on a recurring cause can be combined; outstanding commitments must remain linked to individual customers by name.
How to choose where to start using AI
Start with work that already eats up the head of sales' time and recurs often enough. For example, preparing a review of one salesperson's conversations or finding customers whose promised reply from the team is overdue. This will let you compare the old method with the new one on one task.
Before launch, the owner approves three conditions:
- The result needed: a list of deals to act on or material for training. One initial launch — one main task.
- The resources allocated: the head of sales' time for reviews, the administrator's time for data, and the salespeople's time for subsequent actions.
- The grounds for continuing: the relevant cases are found, the reasons for the conclusions are available, and all the work, including preparation and corrections, fits within the time allocated.
You cannot count only the minutes the software saves on listening. You need to add the time spent gathering material, reading the result, checking disputed points and passing on instructions. If an hour of listening is replaced by twenty minutes of reading and fifty minutes of preparing files, the work takes the head of sales longer. This is clear without calculating future revenue.
There is another reasonable outcome: it takes the same amount of time, but the head of sales notices enquiries that were previously overlooked. In that case, the owner is deliberately buying a more complete review. The write-up of the trial should describe the benefit in exactly those terms so it can be compared with the costs.
A ready-to-use brief for five working days
Five days is the proposed length of a limited trial after access has been set up, not a promised timeframe for sales growth. You can pass the following text to the head of sales without the owner having to listen to calls personally.
Over five working days, compare preparing conversation reviews using the old method and using AI. Before starting, choose comparable groups by sales stage and employees, record what is in each group and the usual time spent on the work. Count the whole process: preparing data, reading, checking and reviewing with the salesperson. Group repetitions of one mistake in completed conversations; keep commitments to waiting customers separate. By the fifth day, bring one page covering the time spent using both methods, cases found and missed, decisions made and what needed my involvement. Recommend continuing, changing the area or stopping the trial — with a reason.
The groups are defined in advance so that one does not contain only short conversations and the other difficult negotiations. Keep the number of recordings, their length and the sales stage alongside the result. Comparing different types of tasks will show a difference in workload instead of the tool's benefit.
Within this period, it is enough to establish whether the new method is suitable for the chosen work. Changes in revenue require separate observation: buyers make decisions in their own time, and sales also depend on what the team does after the analysis. The fact that a payment follows a suggestion is not enough to attribute the entire deal amount to that suggestion.
When there are more useful reviews than time for them
After the trial, the owner needs to make a choice. If most of the time goes into gathering calls and written conversations, address how the data is obtained. If it goes into similar conversations with salespeople, change how training is organised. If it goes into replies to buyers, redistribute the salespeople's current work. Installing a more powerful model will not give employees more hours in the day.
For daily analysis, when information has to be collected from different places, RENTROP offers AI ROP — AI for the sales team: it analyses calls and written conversations together with the deal record, shows who needs a reply, which deals to take through to payment and which to bring back. The service only reads data; the head of sales remains responsible for priorities, instructions and actions.
The owner's evening from the start of the article should not stretch out as the list of issues grows. A recurring mistake adds material to one training session, while every buyer waiting for an answer still requires a separate action. Then the size of the report stops dictating the length of the working day.