Omnia Lab or speech analytics

What does speech analytics do well?

The Russian market splits into three groups: contact-centre platforms built on dictionaries and checklists (CRT/SpeechXplore, 3iTech, imot.io, Speech Analytics), LLM tools aimed at sales teams (SalesAI, Rechka), and modules built into telephony or marketing analytics (Roistat, Mango Office). Their common strengths are real.

• You hear what was actually said, not what someone typed into the CRM afterwards. The mechanics are plain: transcription with operator and customer turns separated, then scoring against parameters — was the script followed, how were objections handled, why did the customer say no.
• Full coverage instead of sampling. SalesAI advertises analysis of 100% of calls and meetings; Speech Analytics auto-scores 100% of dialogues on a five-point scale. A head of sales no longer has to choose which ten recordings to listen to this week.
• Speed. Rechka returns a report within 2 minutes of a call ending and claims 98.7% analysis accuracy; imot.io lands results in statistics and reports 10–15 minutes after the call.
• More than calls. imot.io covers calls, meetings and written conversations. SpeechXplore (CRT) covers calls, chats, assistants, bots, email, social media and in-person conversations. 3i TouchPoint Analytics detects emotion and hidden dissatisfaction from intonation and auto-clusters millions of minutes.
• Plain-language queries. With 3iTech, SpeechXplore and Roistat you can ask the whole communication archive a question in words instead of maintaining a dictionary for every case.
• It is already inside tools you pay for. Roistat ships around 22 dictionaries, scoring on 22+ parameters and a 100-point operator rating inside its marketing analytics. Mango Office sells semantic call analysis as an add-on to its cloud PBX, with 20 free AI topics and a 7-day trial.
• A low price of entry. SalesAI has a free-forever tier: 40 minutes of analysis per day, unlimited users. Rechka has no subscription at all — per-second billing, a one-minute minimum, and a minute package that renews as you spend rather than expiring on a calendar date.
• Onboarding measured in days. SalesAI says "from a few minutes to a couple of days", Rechka says one day, 3iTech says two days for the cloud edition.
• Russian compliance is settled. imot.io, 3i TouchPoint Analytics and SpeechXplore are in the Russian Software Registry; SpeechXplore runs on UZ-1 class protected infrastructure; Speech Analytics is registered with Roskomnadzor as a personal-data operator and keeps its servers in Russia.

If the question is how your team talks to customers, this is the right class of tool. We analyse calls too, but that is one module out of 33, not the whole product.

What does it not see, by design?

This is not a complaint about vendors — it is a property of the class. The input has to be a communication that already happened. Roistat states the mechanics openly: calls are uploaded through the interface, an API, call tracking or a virtual PBX. The object of analysis only exists once contact has occurred and been recorded. Everything that did not happen stays outside.

• A lead nobody called or messaged. No conversation, nothing to analyse — while a person on the other side is waiting.
• A lead with no owner. A deal stuck on a service account or on someone who left the company leaves no trace in call recordings, because there are no calls.
• A promise that was not kept. The rep said "I'll send the quote tomorrow"; the analysis will fairly note a strong close to the conversation, and the quote may never go out.
• A deal closed without a reason. Someone moved it to spam in one click — there may have been no conversation there at all.
• A missing second touch. 93% of deals close by the sixth touch, and half of all leads never get a second call — Velocify, roughly 3.5 m leads. No second call means no recording either.
• A missed call nobody returned. A few seconds of ringing holds no conversation, so there is nothing to score — and the customer is already gone.
• Marketplace service notices such as "the user viewed your number". There is no conversation behind them and so no object of analysis, while the person who looked up the number and could not get through is still waiting.

What limits should you know upfront?

Everything below comes from what the vendors publish about themselves. These are not flaws — they are the conditions under which the tool delivers what it promises.

• Recognition accuracy has a ceiling and depends on audio quality. 3iTech publicly states ASR accuracy of at least 80%, targeting 84% in 2025; speechanalytics.ru quotes 80–95% depending on acoustic models. The conclusions are built on text that itself contains errors.
• Results arrive after the conversation, not during it, and the lag differs by an order of magnitude: Rechka 2 minutes, imot.io 10–15 minutes, Roistat up to 24 hours for transcription in deferred mode. Fine for post-hoc review; not a basis for intervening in a live call.
• Dictionaries and checklists need an owner. Roistat ships around 22 ready dictionaries and expects custom ones for your specifics — in this sub-class, output quality is a function of who maintains them.
• Some platforms are built for a contact centre, not a five-person sales team. 3iTech states a minimum of 25 operators and 50,000 minutes a month for cloud, 50 operators and 100,000 minutes for on-premise, with 2-day SaaS and 3-week on-premise deployment.
• Call recording is personal-data processing under Russia's 152-FZ: notification to Roskomnadzor under art. 22 (1), an automated recording announcement before the substantive part of the dialogue, and an internal policy on audio recording. That is a precondition for any call analysis, ours included.
• Published outcomes are project examples, not guarantees. SpeechXplore attaches a no-guarantee disclaimer to its own figures (up to 60% annual sales growth, 2x QA productivity). Read our numbers the same way: prepayment share moving from 4—7% to 13% and median first response from 32 to 6.5 minutes are before-and-after measurements at specific live clients, not a promise.

What does it cost in Russia?

Publicly listed prices, checked 21 Aug 2026. Not every page states an effective date, so treat this as a basis for estimating, not as an offer.

SalesAI

Free — 0 ₽/mo, 40 minutes of analysis per day, unlimited users. Team — 49,000 ₽/mo (7,000 minutes, up to 20 users). Business — 99,000 ₽/mo (20,000 minutes, aimed at 15–50 reps). Enterprise — 199,000 ₽/mo (50,000 minutes, on-premise). Separately: an AI implementation engineer at 150,000 ₽/mo.

imot.io

From 53,990 ₽/mo for 10,000 minutes to 399,990 ₽/mo for 150,000 minutes. Annual payment takes 25% off, from 40,499 ₽/mo. A load estimate is done before connection.

Rechka (Rechka.ai)

No subscription; minute packages with per-second billing — 11,000 minutes for 60,000 ₽, 140,000 minutes for 600,000 ₽. Roughly 5.5 ₽ per minute with full AI analysis. Integration is a one-off 20,000 ₽, rollout up to 3 working days, free trial on 30 minutes of recordings.

Roistat

0.8 ₽/min in deferred mode, 1.1 ₽/min in real-time mode. Unused minutes are cancelled 180 days after purchase.

Mango Office

1,100 ₽/mo subscription, analysis minutes from 1.25 ₽. 10,000 minutes of AI-topic analysis included each month, then +1 ₽ per recognised minute on top of the plan. 7-day trial access.

3i TouchPoint Analytics

0.54–1.38 ₽/min depending on volume and deployment type. Entry thresholds: cloud from 25 operators and 50,000 minutes a month, on-premise from 50 operators and 100,000 minutes.

Speech Analytics

Pilot projects from 100,000 ₽ depending on call volume and scope; detailed tariffs are not published.

CRT / SpeechXplore

No rouble prices published — tariffs are calculated from communication volumes and the consulting services included.

The billing model deserves more attention than the headline rate. Roistat cancels unused minutes after 180 days; Rechka ties the package to consumption rather than the calendar. With uneven load, the same volume costs very different money.

What does Omnia Lab do between conversations?

Responding within an hour produces 7x more qualified leads than an hour later and 60x more than a day later — Harvard Business Review, "The Short Life of Online Sales Leads", 2011, sample of 2,241 companies. So the checks sit along the whole path of a lead, not only on the conversation.

1. 01

   A lead arrives

   A ping with a "Take it" button lands in the priority chat. If nobody takes it, the lead is distributed by the shift schedule — only to people actually working today. A separate check every minute finds deals stuck on the service account.

2. 02

   The customer is waiting

   An escalation ladder at 5, 10, 15, 17 and 20 minutes: shared managers' chat, the rep's direct messages, then the head of sales. At night the stages are cancelled rather than postponed, so nobody wakes up to an avalanche. At a live client, median first response fell from 32 minutes to 6.5.

3. 03

   A rep promises something

   Words like "I'll call back" or "I'll send it" create a CRM task with a deadline. A private ping 30 minutes before it; another private one 15 minutes past it, with "Done" and "Postpone" buttons; the shared chat only after an hour overdue. Before every ping the AI re-reads the conversation — if the promise is already kept, the task closes itself and no ping goes out.

4. 04

   The customer promises something

   "I'll look at it in an hour" becomes a task called "Customer promised to come back", due at the customer's own time plus 30 minutes. If they come back on their own, the task closes itself. "I'll think about it", with no timeframe, creates nothing.

5. 05

   A call is missed

   A "Call back" task due in 30 minutes. At night Telegram stays quiet and the deadline moves to the start of the working day plus 30 minutes. A real call longer than 15 seconds closes the task automatically.

6. 06

   A conversation happens

   Transcription, a target-or-not check, then analysis with a 0–100 score. Every new call gets a review task; a follow-up call gets one only if the score is below 65. If someone other than the deal owner picked up, the score and the task go to whoever actually spoke.

7. 07

   The customer goes quiet

   Follow-ups at 1, 3, 7 and 30 days, delivered as a single private digest with buttons at 15:00 rather than as a stream of reminders.

8. 08

   A deal is closed

   Every close to spam or lost is reviewed against the full history: messages, comments, the card, the transcript of the recent call. An unjustified close is returned to work — once per deal, with a decision log. During a bulk clean-up of the pipeline (more than 30 closes in 2 hours) returns are muted: the system does not argue with a deliberate human decision.

How do the two compare, point by point?

The criterion on the left, how each side handles it on the right.

What goes in

Speech analytics: a call recording or a written conversation — Roistat describes it as uploading calls via the interface, API, call tracking or a virtual PBX. Omnia Lab: deal events — lead created, owner assigned, customer waiting, promise made, call missed, deal closed.

When results appear

Speech analytics: after the conversation — Rechka in 2 minutes, imot.io in 10–15, Roistat up to 24 hours in deferred mode. Omnia Lab: the first reminder about an unanswered customer fires at minute 5; call analysis follows the call.

A lead with no conversation

Speech analytics: never enters the loop, there is nothing to analyse. Omnia Lab: a "Take it" ping, then distribution by shift schedule on timeout, with its own SLA chain.

A rep's promise

Speech analytics: the words stay in the transcript, and whether they reach the score depends on the dictionary or checklist someone set up for that case. Omnia Lab: creates a task with a deadline and follows it to closure — 30 minutes before, privately at 15 minutes overdue, shared chat after an hour.

Closing a deal

Speech analytics: reviews the conversation, if there was one. Omnia Lab: reviews the close itself against the whole history and returns an unjustified one to work, once per deal.

Scoring the rep

Speech analytics: an operator rating — 100-point at Roistat, five-point at Speech Analytics. Omnia Lab: a 0–100 score where the outcome sets the band and process quality sets the position inside it; the code does the arithmetic, the AI only finds facts with quotes.

Where the output lands

Speech analytics: reports, dashboards, XLS/CSV exports; Speech Analytics also fires triggers — Telegram, email, SMS or a CRM task within 5 minutes of the call. Omnia Lab: a CRM task with a deadline, auto-filled card fields, and Telegram buttons — "Done", "Postpone", "No reply needed".

Team size

Speech analytics: a wide spread — 3iTech states a minimum of 25 operators and 50,000 minutes a month, while SalesAI offers a free tier with unlimited users. Omnia Lab: 3–4 hours to deploy a new client, 3—5 days to roll out, 33 modules each with its own switch.

Legal groundwork

Identical for both: call recording is personal-data processing under 152-FZ, requiring Roskomnadzor notification, an automated recording announcement before the substantive part of the dialogue, and an internal audio-recording policy.

Which one should you pick?

A short decision guide, no pitch.

• You need wall-to-wall quality control of conversations in a contact centre with serious minute volumes — buy dedicated speech analytics: 3iTech, CRT, imot.io. That is their core job and they do it better than any generalist.
• You want a quick, no-rollout look at how your reps talk — SalesAI Free gives 40 minutes of analysis a day at no cost, and Rechka offers a trial on 30 minutes of recordings.
• Speech analytics is already inside something you pay for (Mango Office, Roistat) — start there. It is already funded, and that is an honest first step.
• The problem is not how people talk but that leads sit unanswered, promises go unkept and deals get closed without a reason — that is control of the lead's path, and speech analytics does not reach it by design.
• You are not sure where you are losing money — start with a measurement, not a purchase. It will tell you whether the loss is in the conversations or in the gaps between them.