How AI finds blown deals

What a blown deal is

A blown deal is not the client's refusal but the moment the salesperson let them go. The client asked about instalments — the salesperson did not answer and moved on to something else. The client wrote in the evening — the reply came only in the morning, and the person had already bought elsewhere. From the outside it looks like an ordinary refusal, but in fact the money was on the table.

The problem is that such losses are not visible in CRM reports. The deal is simply marked “refused” or hangs in the pipeline without movement. The reason sits inside the conversation — in the intonation, in the pause before the answer, in the question the salesperson did not ask. Without going through the call or the chat itself, that reason cannot be found.

Why blown deals are hard to find by hand

The head of a sales department physically cannot listen to every call and read every chat. They have time to check five or ten conversations a week selectively; the rest goes past. If a salesperson has a hundred conversations a month, the manager sees at best a tenth of the picture.

The situation is complicated by the fact that enquiries do not arrive only during working hours. In the live GetGate project 27% of enquiries come in the evening and at weekends — they used to wait until morning. While the salesperson is away, the client either waits or goes to whoever replied faster. Manual checking hardly catches such losses, because the review happens after the fact and selectively.

It is harder still with chats. A text conversation can be skimmed quickly, but an important detail is easy to miss — for example, that the client asked about deadlines three times and the salesperson never answered directly. It is hard for a person to hold a ten-point checklist in their head for every conversation. AI does the same work by one and the same algorithm, without tiring and without selectivity.

How AI listens to calls and reads chats

AI goes through every call and every chat, not a sample. It turns speech into text, picks out the client's questions and the salesperson's answers, notes pauses and repeated enquiries. In chats it sees the whole conversation history, not just the last message.

Then each conversation is checked against a checklist: did the salesperson say hello, did they find out the need, did they name the price and the next step, did they offer an alternative when there was an objection. This happens for every enquiry without exception. If the salesperson missed the client's question about delivery times, the system records it just as an attentive listener would have.

This kind of review is described in the section of quality control. There is a second effect hidden here: AI can be put in not only as a listener but as a participant in the conversation. We checked on real correspondence how far the assistant's replies differ from a live salesperson's, and in 80% of cases in a blind comparison the assistant's reply was no worse. That means the checklist used to assess salespeople applies to the robot itself as well.

The checklist: by which signs AI finds a blown deal

The first sign is a long answer to a hot question. If the client asked about price or availability and the salesperson answered an hour later, that is already a potential blown deal, even if it is not formally closed. The second sign is a question without an answer. The client asked about instalments, a guarantee or deadlines, and in the chat that question simply got lost among other topics.

The third sign is the absence of a next step. The conversation ended without a date for a call, without an invoice sent, without an agreement about a meeting. The deal hangs not because the client changed their mind but because nobody set what happens next. Such cases are looked at in more detail in the material on to build the sales department, where the next step for every lead is a task in its own right.

The fourth sign is the tone of the conversation. Irritation, an indifferent answer, a formal phrase instead of a live dialogue — AI reads all of this from the intonation in a call and from the wording in a chat. The fifth sign is a discrepancy between what the salesperson wrote in the CRM and what was actually said to the client. Sometimes the deal card says “the client is thinking” although in the conversation the client said outright “I'll take it”.

Salespeople's scoring: how it is calculated

Every conversation gets a score against the checklist, and the scores of all conversations add up to the salesperson's score for the period. This is not the manager's subjective opinion but the sum of specific points: said hello, found out the need, named the price, closed on a next step. The score shows not a general “good or bad” but exactly which stage fails most often.

If a salesperson consistently scores low on “offered an alternative when there was an objection”, that is a specific point to go through at the team meeting. The manager does not have to guess where the problem is — the checklist has already named it. The same score helps to compare shifts, channels and periods: where deals fall apart more often — in chats or in calls, in the evening or during the day.

Without AI such a review would cost the manager hours of listening and subjective notes on paper. With automatic scoring this work runs in the background and the result appears as a ready report. What remains is to decide what to do about the gaps found — train the salesperson, change the script or rethink how enquiries are distributed.

What happens after a blown deal is found

Finding the loss is only half the job. Next you have to work out whether it can be put right now, while the client has not gone for good. If a deal hung without an answer, the manager sees it in the report and can return the lead to work before it goes cold.

If the problem is systemic — for example, evening enquiries regularly go unanswered until morning — that is a reason to connect a about the sales robot, which replies to the client immediately at any hour. Then the 27% of enquiries that arrive in the evening and at weekends will not wait until morning but get an answer at once. That does not remove every blown deal, but it removes the most frequent and the most galling one — losing a client because of a banal pause in replying.

In the live GetGate project introducing such automation freed up about an hour of a salesperson's working time a day. That time goes not on routine replies but on going through complex cases and on calling the clients worth pushing personally. Quality control and automated replies work together: one finds the problem, the other removes its most frequent cause.

How quality control is linked to the CRM and analytics

Going through calls and chats makes no sense without somewhere for the results to flow. All the data about deals, conversations and statuses is stored in the CRM, and that is where AI finds the context for its assessment — the client's history, previous enquiries, the pipeline stage. Setting up such a system is described in the section CRM implementation.

After that the data on blown deals and salespeople's scores become part of the general sales picture. In sales analytics you see not only plan against actual on revenue but also at which pipeline stage enquiries are lost most often. If the breakdown by salesperson shows that one person consistently scores low precisely at the closing stage, that is visible at once rather than a quarter later.

Such a link turns the review of individual conversations into a constant process rather than a one-off check. The manager does not hunt for the problem by hand once a month — they see it in the report every day, alongside the revenue and conversion figures. This is the case where quality control stops being a separate task and becomes part of ordinary work with the pipeline.

Where to start

The first step is not buying technology but reviewing the current sales department: how many calls and chats there are a month, where exactly enquiries are lost, whether there is already a checklist for assessing salespeople. Without this review it is hard to understand what exactly AI should look for in a particular company's conversations — selling furniture and selling consultancy have different points of failure.

Next the communication channels are connected — calls, Telegram, WhatsApp, chats from the website — and the checklist is set up for the client's real process rather than a universal template. Gradually it becomes clear which salespeople cope and which pipeline stages need going through at team meetings. More about what this process looks like as a whole can be read in the blog or on the page of quality control.

A blown deal is rarely visible straight away — it hides in the pause before an answer or in a question nobody responded to. AI does not replace the conversation with the client, but it shows exactly where that conversation broke down. That turns quality control from a rare spot check into constant work on every deal.