Speech analytics, or keeping control of the whole lead journey

What does speech analytics actually give you?

Credit where it is due. Without call review, the only record of a conversation is whatever someone remembers.

• You hear how the rep ran the call: did they find the need or jump straight to price, did they agree a next step or just say goodbye.
• You see what the customer was actually told, not what got typed into the card afterwards.
• You get one shared scale. Ours is a 0—100 score: the outcome sets the band, process quality sets the position inside it.
• Repeating weak spots surface: the same phrasing, the same collapse at the end of the call.
• You get coaching material. A direct quote convinces a rep faster than an opinion does.

In Omnia Lab this is one module, not the whole product.

What can speech analytics never see?

Call review works on what happened. Most leaks on a sales floor are things that did not happen. 93% of deals close by the sixth touch, and half of all leads never get a second call — Velocify, roughly 3.5 million leads.

• A lead with no call and no message at all. No conversation, nothing to analyse — and a person is still waiting.
• A promise nobody kept. The rep says "I'll send the quote tomorrow". The call scores well for a clean close, and the quote never leaves.
• A close with no reason. The deal went to spam in one click, and there may have been no conversation there at all.
• A missing second touch. The customer went quiet and nobody thought about them again.
• A missed call nobody returned. There is no recording, because there was no call.
• A lead stuck on a service account, or on someone who left the company.
• Marketplace system notices like "the user viewed your number". In a report that looks like customer contact, when no contact happened.

Where do the checks sit along the lead path?

Responding within an hour produces 7 times more qualified leads than responding an hour later, and 60 times more than a day later — Harvard Business Review, "The Short Life of Online Sales Leads", 2011, 2,241 companies. So the checks sit along the path, not at month end.

1. 01

   A lead arrives

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

2. 02

   The customer is waiting

   An escalation ladder at 5, 10, 15, 17 and 20 minutes. 5, 15 and 17 go to the shared rep chat, 10 to the rep's private messages, 20 to the head of sales. A reply, a call, or the "No reply needed" button clears all of it at once.

3. 03

   A call happens

   Transcription, a check on whether the call was a real sales lead, and a scored review. New calls always get a review task in the CRM; follow-up calls get one only when the score is below 65. If someone other than the deal owner picked up, the score and the task go to whoever actually spoke.

4. 04

   A rep promises something

   Words like "I'll call back" or "I'll send it" create a task with a deadline. 30 minutes before the deadline: a private ping. 15 minutes overdue: private again, with "Done" and "Postpone" buttons. The shared chat only comes in after an hour overdue. There is no public callout before the deadline at all.

5. 05

   The customer promises something

   "I'll look at it in an hour" becomes a task called "Customer promised to come back", due at their stated 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.

6. 06

   A call is missed

   A missed call becomes a "Call back" task due in 30 minutes. At night Telegram stays silent and the deadline moves to 30 minutes after the start of the working day. A real conversation of 15 seconds or more closes the task — 5 to 8 seconds of ringing does not count.

7. 07

   The customer goes quiet

   Follow-up touches on days 1, 3, 7 and 30, daily at 15:00, in a single private digest with buttons. One list, not a stream of chat messages.

8. 08

   A deal gets 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 goes back into play — once per deal at most, with a decision log.

9. 09

   The day winds down

   At 11:00 and 16:00, a "No next step" block: the customer got a reply, but no open task exists. At 18:00, active deals are reviewed on four axes, led by a "Needs attention today" section: priority from 70 up, 10 deals maximum. In the morning each rep gets a four-block day plan in one message.

How does call review work inside Omnia Lab?

We do review calls. It simply sits inside the wider control loop rather than at the centre of it.

Score

0—100. The outcome sets the band: 90—100, 80—89, 65—79, 40—64, 1—39. Process quality sets the position inside the band.

Who does the maths

Code calculates the score. The model only finds facts and backs them with quotes.

Critical mistakes

They cap the score rather than subtract from it, and only count with a verbatim quote. No quote, detection discarded.

Compensation

A step covered later, or covered by the outcome itself, gets full credit. Asked for the name at the end but the sale happened — not a deduction.

When a task is created

Always for new calls. For follow-up calls, only when the score is below 65, so reps are not buried in reviews.

How the task reads

Written from the rep's point of view, with no mention of AI: what to do now, the score and the logic, what was strong and weak, one recommendation.

Customer profile

Every conversation gets a segment with a confidence level and quotes. Failing to push an objectively unqualified buyer is not counted as a mistake. Politeness is judged for everyone, always.

The distrust rule

A mitigating segment counts only at 75% confidence or higher and with a direct quote. Haggling, "that's expensive", asking for a discount, short replies and silence are never evidence that someone cannot pay.

Phone systems without events

If your PBX does not send call-end events, we poll it ourselves every 2 minutes.

What do teams get wrong when choosing a system?

• Assuming call review is all the oversight a sales team needs. It grades what happened and stays silent about what did not.
• Measuring response time from the first message in the thread. "OK" at 12:00, a question at 14:00, an answer at 14:05 is 5 minutes, not two hours. The false-start filter counts it honestly.
• Letting marketplace system notices into the stats. "The user viewed your number" is not a customer message, so it never starts an SLA clock. It still stays visible in the history.
• Confusing a call score with a deal score. They are different metrics and live in different columns.
• Expecting reps to fill in the CRM for the sake of analytics. Deal details are pulled from the call transcript, and a title is only changed when it is still the default. Anything a human typed stays untouched.
• Switching control on across the entire CRM at once. There is a configurable list of pipelines and stages with no SLA and no digests, and you set it before launch.

What does the system deliberately not do?

• It does not replace your CRM. It runs on top of the CRM you already use — Bitrix24 is the primary integration, and we work with amoCRM too. Reps stay in the interface they already know.
• It does not call anyone out in public before a deadline. Until the time is up, the conversation stays private.
• It does not argue with a deliberate human decision. During a mass pipeline cleanup — more than 30 closes in two hours — reinstatements are muted.
• It never rejects a deal with a prepayment. Not to spam, not to the cold pipeline, ever.
• It does not turn chats into a feed. A new reminder deletes the previous one, and the old one goes only once the new one is delivered.
• It does not defer night escalations to the morning, it cancels them. Overnight leads arrive as one digest at 10:05.
• It does not hide data when something breaks. Fail-open: the lead still gets assigned, the conversation still appears, a filter switches off rather than concealing anything.
• It does not spend your budget quietly. The AI budget is capped daily, with alerts at 50% and 100%, and only deferrable processes get throttled — customer-facing flows never stop.

What changes in the numbers after rollout?

Measured before and after rollout at existing clients, including a premium car dealership.

Median first response time

from 32 minutes to 6.5

Share of deals with a prepayment

from 4—7% to 13%

Revenue

+33% in 4 months, with no extra headcount

Timelines

rollout in 3—5 days, free audit in 3—5 days, technical setup of a new client in 3—4 hours

Flexibility

33 modules with individual switches and 85 settings in the panel, applied in under 60 seconds, no restart

Until you switch to live mode every module sleeps, so your team does not notice the system until you turn it on.