MoySklad
Odoo
amoCRM
Tizimlashtirish
Marketing
AI agentlar
BUline
MoySklad
Odoo
amoCRM
Систематизация
Маркетинг
Маркетинг: обзор
Стратегия и контент
Стратегия и планКонтент и SMM
Рекламные каналы
Google AdsYandex DirectРеклама в TelegramInstagram и FacebookМаркетплейсы
Органический рост
SEO
Измерение и аналитика
Аналитика и KPI
AI-агенты
BUline
Home/AI agents/Call analysis
AI agents · Calls

Every call reviewed and scored

A manager usually listens to only a small share of calls. The AI agent transcribes every recording, scores it against the script and surfaces only the problem conversations for review.

VoIP telephonyUzbek and RussianScript complianceCRM summary
Goal

Calls are recorded, but no one listens to them

VoIP telephony records every conversation, but managers have no time to listen. As a result, a sales rep’s mistakes repeat for months. The call analysis agent transcribes the recording, separates the speakers and scores the conversation against a checklist. The summary and score are written to the deal in the CRM.

  • Greeting, needs discovery, offer, next step: were they there?
  • Customer objections and how they were handled
  • Promised deadlines and terms
  • Customer sentiment and signs of a complaint
—
—
—
Workflow

From recording to CRM summary

Analysis starts automatically when the call ends.

01Trigger: the call endsVoIP telephony

The recording file and call details (number, sales rep, duration) arrive in the CRM.

02Speech is converted to textSpeech-to-text (STT)

The audio is transcribed into Uzbek or Russian text, and speakers are separated.

03The agent scoresChecklist

The text is checked against the company script, with evidence for each item: a quote from the transcript.

04The CRM is updatedDeal record

A short summary, the score, the next step and objections are added as a note on the deal. A task is created for the agreed next step.

05A human reviewsQuality control specialist

Calls with a low score, a complaint or “red flag” words go to the manager in a separate list.

Technology

Uzbek speech recognition: where it stands in 2026

Public services support Uzbek, but quality varies sharply with the phone line and the dialect.

Yandex SpeechKit

  • Uzbek was added in 2023 (Yandex Cloud blog)
  • In the documentation, uz-UZ is Latin script
  • Supports automatic language detection per sentence

ElevenLabs Scribe

  • Uzbek is on the list of supported languages
  • Speaker separation and timestamps
  • The vendor’s accuracy figures have not been independently verified

Fine-tuned Whisper

  • Open models fine-tuned for Uzbek are available
  • Can run on your own server, so audio never leaves your premises
  • Quality depends on the training data; testing is a must

Dialects and mixed speech are still a challenge

Regional dialects, Uzbek and Russian words mixed within a sentence, noisy or low-quality phone lines all increase errors. That is why, before choosing a service, we manually measure the word error rate on 50–100 of your recordings and pick the option with the fewest errors.

More on limitations
KPIs

How results are measured

Technical quality and business results are measured separately.

MetricCalculationType
Word error rate (WER)Incorrect words in a manually checked sampleTechnical
Score agreementCalls where the AI score matched the reviewer’s scoreTechnical
Analysis coverageCalls analyzed / all callsProcess
Script adherenceAverage checklist score by sales repBusiness
Feedback speedTime from the call to feedback for the sales repBusiness
Preparation

Required data and rules

Beyond the technical connection, legal and organizational groundwork is needed.

Data

  • VoIP telephony that provides recordings via API
  • A written sales script and checklist
  • Calls scored by a reviewer, for testing

Rules

  • Notifying customers that the call is recorded
  • Retention period for recordings and transcripts
  • Who can see transcripts, by role

People

  • The checklist owner is the head of sales
  • Weekly calibration: comparing AI and human scores
  • How results are discussed with sales reps
FAQ

Frequently asked questions

Can we penalize a sales rep based on the AI score?

We don’t recommend it. The AI score is a tool for selecting calls and for coaching. Disciplinary decisions are made only on recordings a person has listened to.

How does it connect to amoCRM?

The telephony integration attaches the recording to the deal record. The agent picks up the recording, analyzes it and writes the summary into a note and fields.

What else can be analyzed besides calls?

The same approach applies to Telegram and Instagram chats: response speed, script adherence, unanswered inquiries. A daily report goes to the manager on Telegram.

Where are the analysis results shown?

Each call summary is on the deal record, and overall metrics are in a weekly table by sales rep. A manager clicks a low score and immediately hears the relevant part.

Where is the audio processed?

It depends on the service chosen: a cloud service or a model on your own server. This is agreed at the start of the project together with personal data requirements.

Free diagnostic

We identify which work can be handed to AI

We review your processes and recommend a specific task, KPI and plan for your first agent.