Speech analytics: what it is and why sales teams need it

Speech analytics is the automated analysis of conversations: what the customer asked, how the salesperson answered, what they agreed on and where a difficulty arose. In a sales team, it helps identify mistakes in communication and analyse why deals stall. To make a decision, the head of sales also needs the terms of the sale, the communication history and the actions taken by employees: one remark cannot show who could actually have changed the outcome.

A customer needs a launch on Friday. The salesperson says the earliest date is in ten working days. The buyer declines, and the call review includes a criticism: the salesperson did not address the objection about timing. This is a training example, not a RENTROP case study; below, we will work through it to decide what to do.

The business owner was looking for a tool that would explain why capable employees lose buyers. The answer was familiar: the salespeople need to sell better. But before sending the team for training, it is worth finding out: did the company have a way to fulfil this order by Friday?

I believe good analytics should help us see both a salesperson’s mistake and a situation where they honestly told the buyer the limits of what the company could do. Otherwise, you will pay for training in how to promise things the business does not do.

How speech analytics works

The analysis is based on the content of the conversation. The recording is transcribed, then the relevant events are identified in the text: a question about timing, an objection, an offer, an agreement on the next contact. The resulting finding is checked against the original recording: a number transcribed incorrectly or a missed negative changes the meaning of the answer.

Searching for words serves a narrow purpose. It helps collect conversations that mention delivery, price or a competitor. Analysing meaning answers the next question: what exactly was happening at this point in the conversation. The buyer refused to wait, asked about the date or agreed to reschedule — the word “timing” will appear in all three cases.

Next, the conversation needs to be linked to the deal. If a suitable option has already been sent in writing, the absence of another offer in the call means one thing. If no options were offered at all, it means another. So when choosing a system, it is worth finding out which sources it reads and whether it can link conversations to the right customer record.

Analytics differs from transcription in that it provides a finding on a specific question. It differs from reviewing calls by listening to them: software does the initial analysis. The head of sales sets the focus of the analysis and decides whether there is enough evidence to act. Having a detailed transcript of the conversation does not do this work for them.

What speech analytics does for sales teams

It helps to start with a question people are currently debating. For example, salespeople attribute refusals to long lead times, the head of sales blames weak negotiation skills, and the business owner hears both versions. This debate calls for different types of information.

What the buyer needs. Whether they gave a date that must be met, explained the reason for it and discussed other options. Without this, it is easy to mistake a preference for a firm requirement.
What the salesperson did. Whether they clarified the deadline, offered an available alternative, requested approval or ended the conversation straight away. This reveals the specific skill to work through with the employee.
What the company could offer. What lead time was confirmed at that point, whether a faster option existed and who had authority to approve it. The head of sales adds this information from the company’s rules and the deal history.
What happened after the call. Whether the promised estimate arrived, whether the person responsible for approval replied and whether the salesperson got back to the buyer. Some sales are lost after a perfectly competent conversation.

This analysis helps you choose what to do: train the employee, change the approval process or review the terms of the offer. It also shows when the selected reason for a refusal in the CRM system is too general. A similar problem with the word “expensive” is discussed in the article what to find out before lowering the price: the same recorded refusal hides different reasons for the buyer’s decision.

Speech analytics example: twenty refusals over lead times

Let’s continue the training example. The head of sales took twenty closed deals with the reason “unsuitable lead time” and reconstructed the options available at the time of each conversation. The numbers illustrate a method of analysis, not market statistics or a client’s results.

Six deals: a faster option existed, the salesperson had authority to offer it but did not. Here, the head of sales examines the missed opportunity and teaches the employee to recognise when that option is appropriate.
Four deals: the salesperson requested a shorter lead time, but approval came after the date by which the buyer expected an answer. Here, the deadline for a decision and the way the request is passed on need to change; otherwise, a more persuasive conversation will end in the same wait.
Seven deals: the company could not physically fulfil the order by the required date, there was no suitable alternative, and the customer refused a later date. The decision rests with the owner: meet this demand in another way or deliberately choose not to.
Three deals: the recording cuts off, and the available data does not include the messages about approval. The head of sales reconstructs the information. These cases cannot be assigned to one of the reasons in advance.

The result was 6 + 4 + 7 + 3 = 20. There is a direct basis for training salespeople in six deals. Another four require work on approvals. In seven, the company’s offer itself needs to be discussed. For three, the conclusion is deferred until the data is available.

If everyone is sent to a session on handling objections, seven orders that cannot be fulfilled will not become feasible. But the salespeople will get a clear message: they will be criticised for an honest answer about timing. The next salesperson will start promising a shorter lead time before it is confirmed, and the conflict will move from sales to fulfilment.

The owner faces a very specific choice here. Is it worth changing workloads, buying stock or introducing a separate offer for urgent orders? The head of sales must bring the number of these enquiries, the buyers’ requirements and a calculation of fulfilment costs. Keeping the current lead times is also a valid decision — then declining an order known to be impossible to fulfil is no longer considered a salesperson’s mistake.

How to assess the findings of a speech analytics system

A convincingly worded criticism still needs evidence. For each significant finding, ask the head of sales to establish three things: what the employee knew, what they had authority to do and how much time remained before the buyer’s decision. This sets a consistent rule for assessing people.

The sequence of events is especially important. If a slot opened up in the schedule on Tuesday, you cannot expect the salesperson to have offered it on Monday. If the owner alone approved a shorter lead time after the customer declined, the salesperson did not have that approval during the conversation. The current state of the record does not always reflect yesterday’s options.

Here is a completed sample record for the head of sales’ manual review of one of the four deals in the training example. This is a format for internal analysis, not a promise about the interface of a particular service.

Request: launch by Friday; the buyer will choose a supplier by 16:00 on Monday.
Conversation: at 10:00, the salesperson stated the standard lead time and promised to ask about a faster option.
Action: the request was sent to the person responsible for fulfilment at 10:15; the salesperson has no authority to confirm a shorter lead time independently.
Company’s response: approval was received on Tuesday at 12:00. The buyer had already chosen another provider.
Head of sales’ decision: agree with the head of fulfilment who will take up an urgent request before the deadline for replying to the customer. The rules should specify the response time and the person to pass a delayed request to.

This record is enough to discuss the delay in approval. It provides no grounds for accusing the salesperson of handling objections poorly. If the company also requires the buyer to be warned about a delay, that is a separate action: whether it is required and whether it was done are examined separately, without confusing it with authority to promise a launch.

How to introduce speech analytics: the owner’s instructions

The owner does not need to label conversations themselves. Their job is to define which decision the pilot must inform, who is responsible for it and the amount of resources allocated to it. Below is a ready-to-use set of instructions; the ten working days are the chosen timeframe for the trial, not a promise about what introducing the system will achieve.

Within ten working days, find out which refusals over lead times we can reduce through actions by the sales team. The head of sales is responsible. Take the twenty most recent completed deals with this reason, without selecting convenient examples. Before starting, agree the spending limit for the trial with me and set aside time for the employees whose help will be needed.

By the third working day, report whether the recordings, messages, approval history and rules in force at the time of the conversation are available. For missing information, name the source and the date it will be obtained. If the key data is unavailable, suggest restoring its collection first.

By the tenth day, bring a breakdown of the reasons with supporting evidence: an excerpt from the conversation, an event in the deal, an available course of action. List cases with insufficient data separately. For each confirmed reason, state what you will change yourself and what decision you need from me.

To continue the work, propose one observable result: which action by salespeople or those responsible for approval should change, where this will be visible and who will monitor it. Also show how much time the analysis takes after the automated findings are received.

Twenty deals provide material for an initial decision on how to adjust the work. They do not prove the share of each reason across all deals. If they include only one urgent request, extend the selection period specifically for this reason before changing the company’s offer.

There are three reasonable decisions after the trial. Continue — when the findings are supported by sources and the head of sales can act. Adjust the setup — when the information is available but the rules for analysis mix up different situations. Sort out the data first — when the content of conversations and the history of decisions are not included in the analysis. There is no point in making revenue over these ten days the only measure: those refusals have already happened, and the changes will need to be applied to new sales.

How to choose speech analytics for your sales team

Ask to see an analysis of your own conversations. Include a confirmed mistake by a salesperson, a justified refusal of an order that cannot be fulfilled and a case with incomplete data. This is a demonstration task: the head of sales prepares it, and the owner receives a conclusion on whether the tool is suitable.

When comparing offers, you need answers to five questions:

Where do recordings and messages come from, and how are they linked to the deal?
Which of your company’s rules are used in the analysis, and who updates them when the terms change?
Can you open the evidence behind a criticism and check it against the original conversation?
How does the system report that it is missing a recording or another part of the history?
What will the head of sales receive for working with the employee, and which actions will remain manual?

Pay particular attention to the finding on a justified refusal. If every disagreement from a buyer becomes a criticism of the salesperson, the head of sales will spend time overturning unfounded accusations. The pilot should show whether the chosen tool can distinguish between situations that matter to your sales.

When conversations have to be collected separately from deals, AI ROP from RENTROP handles this task: speech analytics analyses calls and messages together with the deal record, takes the client’s rules into account and shows mistakes with a quote and a suggested response. AI ROP only reads data; the head of sales assigns tasks to employees and makes decisions based on the findings.

The buyer from the start of the article will still not get a launch on Friday if the company can only launch in ten days. The value of the analysis is that the owner stops expecting the impossible from the salesperson and chooses whether the business itself will handle such orders. And the head of sales can then hold the salesperson accountable for missing an available option — there is a basis for that demand.