An X-ray of calls: finding where the sales team loses deals

An X-ray of calls: finding where the sales team loses deals

A manager listens to three calls out of three hundred and draws conclusions from a sample he chose himself. A full sweep finds something else: where exactly the conversation breaks and which objections are said verbatim. Below: what gets checked, what gets found and what makes up the fee.

A manager listens to three calls out of three hundred and draws conclusions from a sample he chose himself. A full sweep finds something else: where exactly the conversation breaks and which objections are said verbatim. Below: what gets checked, what gets found and what makes up the fee.

There are leads, conversion is falling, the cause is unknown

The owner sees the figures from above: this many enquiries, this many sales, conversion lower than it was. He then asks the head of the department what is happening and hears: "clients have got worse", "the market is down", "competitors are undercutting us".

There is no way to check that. The call recordings sit in the system, there are hundreds of them, and there is nobody to listen: listening to ten conversations takes a working day, and ten conversations prove nothing.

So decisions are made blind. They rewrite the whole script, introduce incentives, hire a trainer. Sometimes it helps, more often it does not — because they are fixing the wrong place.

And the place is specific, and it is almost always the same one: deals collapse at one particular step, with most of the team, for one reason.

Why the usual approach does not find the cause

The root cause: the audit is done on a sample and the problem is statistical.

A trainer or a manager listens to a few conversations and draws conclusions. The sample is random and tiny. If the problem occurs in a third of calls, ten recordings may not show it at all — or show it where it is not.

The misconception: "we need an experienced expert, he will hear it". He will — in the conversations he listens to. But he physically will not listen to three hundred calls, and it is at three hundred that the pattern becomes visible.

A second misconception: "we need to rewrite the script". The script is usually not to blame. What is to blame is that it is not followed in one specific place — and until you know which, the new script will be ignored in exactly the same way.

And a systemic layer. A conversation is a sequence of steps, and the loss happens at a specific one. Contact established, need identified, price named, objection handled, next step agreed. Count how many conversations reach each step and the breakpoint is immediately visible.

How the pipeline works

The work runs on an array of recordings rather than on impressions.

Step 1. Full transcription

Every conversation for the period — not a sample. Two hundred, three hundred, five hundred recordings turned into text. That used to be impossible in labour terms and now takes hours.

What matters: it is the full sweep that gives statistics. A sample gives an opinion.

Step 2. Tagging by step

Every conversation is broken into stages: which one it reached, at which it broke. An assistant from the "Sales" category tags the recordings against a single scheme.

The output is a funnel inside the conversation: what percentage of calls reach need identification, how many reach price, how many reach an agreed next step.

That is usually where the failure appears. For example: ninety percent reach the price and twenty percent an agreed next step. That means the managers name a figure and let the person go away "to think".

Step 3. Analysing the breakpoint

Then only the conversations that broke at the identified point are taken and analysed in detail: in what words it happens, what the client says, what the manager answers.

Here specifics replace a diagnosis. Not "they handle objections badly" but "when the client mentions price, in seven cases out of ten the manager starts making excuses and offers a discount unprompted".

Step 4. What the client says

A separate pass through the clients' lines. The assistant collects a bank of objections verbatim: how people actually phrase their doubts.

This is the most underrated part of the audit. A business usually assumes clients object on price, and the recordings reveal that what is said more often is "it is not clear what exactly is included" — that is, the problem is in the product description rather than in selling.

Step 5. Manager profiles

A comparison on one scheme: whose step sags where. This is not about punishment — it is about the fact that different people need different training. One is short on diagnosis, another on closing, a third never gets to a next step at all.

Step 6. Correction and a simulator

For the identified breakpoint it is not the whole script that gets rewritten but the relevant fragment. Then a simulator: the assistant plays a client with real objections from the bank, and the manager practises the conversation in writing or by voice.

A mistake on the simulator costs nothing. A mistake with a live client costs the deal.

What the business gets

The last point matters more than the rest. A one-off audit gives a one-off improvement. A built-in review gives continuously accumulating experience: every new conversation becomes data rather than air.

The four breakpoints that occur most often

The picture repeats across different departments, and it helps to know what to look for.

Diagnosis skipped. The manager goes straight to the pitch without finding out what hurts. He then sells the wrong thing, and objections appear out of nowhere.

Price named without justification. The sum was said on its own, tied neither to a result nor to what the current situation costs the client. From there the conversation turns into haggling.

An objection taken as a refusal. "Too expensive" and "I will think about it" are heard as the end of the conversation, when they are a request for clarification. No clarifying question follows.

No next step agreed. The most common and most galling loss. The conversation went well and ended in nothing: no date, no proposal sent, no agreement. The client "will think about it" and does not come back.

The fourth usually gives the largest gain for the least effort: it is fixed with one sentence at the end of a conversation.

What a call audit costs

Work like this costs from $2 400. An honest caveat: that is the stated cost of the scenario by our methodology, not market statistics — I have not measured what such audits fetch on the market on average.

What makes it up: the volume of the recording array, the number of managers, the depth of analysis, whether a simulator is needed and whether a regular process is being set up. The cost is calculated from those parameters rather than taken from a price list.

What to compare it against. Not the price of a training course but the price of the error: what a third of deals collapsing at the same point costs the business. Usually the question of price closes after the first calculation.

Why this stopped being expensive

An audit like this always existed. It cost as much as a separate project, because it ran into person-hours: somebody had to listen to hundreds of recordings and tag them against a single scheme.

Exactly one thing changed: processing a whole array became cheap. Transcribing three hundred conversations and tagging them by step takes hours instead of weeks, and the tagging is better — because the scheme is applied identically to every recording, with no fatigue and no personal sympathies towards staff.

Two consequences follow. The audit became available to companies that never used to commission it. And it stopped being one-off: something that takes hours can be done monthly, which means accumulating a trend rather than a single snapshot.

What a call audit needs

The work is assembled from ready tools rather than written from scratch.

AI assistants — over 200 across different categories, including a separate "Sales" block: conversation analysis, an objection bank, scripts, simulators with client simulation.

Automation templates — so that analysing new recordings runs by itself rather than being launched by hand each time.

Your own assistants for the company — when tagging is needed against your sales scheme rather than a general one.

The last two live on the higher plans: automation templates and creating your own assistants from Creator, advanced automations Full. A one-off audit can be done on a lower plan; a permanent process needs the higher ones.

Registration is free and opens three days of Basic access — enough to run a dozen of your own recordings through and see what an audit actually shows. From there the decision is made on the result rather than the description.

Do one thing today: take the twenty most recent recordings and count one figure by hand — in how many of them the conversation ended with a specific agreement on a next step rather than the words "we will think about it". If it is fewer than half, you already know where the money is being lost.

Not building from scratch: a ready system

Everything described above can be assembled by hand. But this system is already assembled and can be taken whole.

The advanced n8n template archive holds “Sales team control via chats and calls” — the same system we at AI Skills sell to external clients.

What it does

It listens to the telephony itself, reads chats, calls and the client's CRM, transcribes voice and delivers a verdict on each manager. Not "reasonably good overall" but specifically: who ignored seventy percent of price objections today and who needs sending on a training course.

Five scenarios inside:

Collecting calls with automatic transcription and attribution to a manager. Collecting messengers — a full log of correspondence with no junk and no auto-replies. Staff synchronisation — the manager database updates itself. Daily reports with consolidated arithmetic. A weekly review of the team's work.

What is needed to sell it

The template opens in your workflow fully configured. All you do is enter your own logins and passwords per the instructions — and the system can be sold as a ready service. Or refine individual nodes for a specific client: adjust the prompt, the system instructions, the assessment criteria.

That is a fundamentally different entry economy: you are not spending a month assembling, you are selling something that already works.

Who has access

The advanced n8n template archive opens with the Full plan (or for a one-off $9, or 1000 XP); regular templates are included in Creator, Make, N8N and Full. In total the section holds over 3 000 ready scenarios, including RAG agent templates.

Why n8n rather than Make. Large businesses fear leaks and bills from cloud platforms where every action is metered. n8n deploys on the client's own server through Docker — and the automations become unlimited for a fixed $5–10 a month. For a client with a large call volume that is the decisive argument.

How this pipeline is built in the programme

The same pipeline is assembled from the catalogue assistants. It breaks down step by step:

The X-ray of calls. Recordings of real conversations are run through a transcriber and the text is given to the Sales analyst bot from the catalogue. The bot performs a merciless analysis: where the manager pushed too hard, where he failed to identify the pain, which red flags led to the deal collapsing.

Armour against refusals. The holes found go to an objection-handling assistant — for example the Inside Sales Representative: it turns refusals into repeat contacts and writes scripts for your market rather than general ones.

Automating the monitoring. Then the logic of an agent that listens to calls in real time itself and sends the manager a notification if a salesperson breaks the script. The agent itself is built in Make or N8N modules.

The client receives not a table of errors but a monitoring system and new scripts that go straight into use.

What to transcribe with. The Gen AI section, "File → text" mode: speech transcription with subtitles, recognition from an image, audio analysis. Wallet payment from $5, the price visible before launch.

The legal frame, without which the audit is not done. Recording conversations between staff and clients is processing of personal data. Consent is obtained in advance and in writing: from managers in their employment contract, from a client by a sentence at the start of the call.

The junior version of the same service

The monitoring system covers a whole sales department. But approaching a client with it straight away is hard: the fee is large and there is no trust yet.

There is a simpler version people start with.

A messenger conversation analyst — the Make template “Analysing managers' calls in Telegram”. The bot catches any message in a working chat. Text — the model takes it apart into script errors and client pains. Voice — it is transcribed, and a summary arrives of the form "the client doubts the price, the manager did not address the discount".

Why this is a good entry point. It is implemented in a day, costs $360–1 200, and in the first week the manager sees something he has never seen: what happens inside conversations.

And then expansion. Once the client has seen the value on messengers, the conversation about the full system with calls, CRM and daily reports goes very differently.

The phrasing for the sale: you are not selling a bot — you are selling control over profit.

Both solutions are ready templates in the section: the first on n8n, the second on Make. Both are advanced: the Full plan, or a one-off $9, or 1000 XP.

What to take on the platform for this task

What to read next

Four scripts and two simulators — what is done with the objections found.

AI for salespeople — the same skill for yourself rather than to sell.

Transcription and file analysis — the technical side of working with recordings.

Auditing procedures — if the problem is the process rather than the people.

n8n and Make automation templates — the full guide to automations.

Also on AI SKILLS