Free audit ↗
Sales OS

An AI sales director that runs every lead from arrival to money

Omnia Lab is a sales operations system on top of your CRM — Bitrix24, amoCRM or any other — an AI sales director for your pipeline. Silence escalates at minutes 5, 10, 15, 17 and 20, every promise a rep makes turns into a dated task, and every close is reviewed before it sticks.

· · Author: Dmitry Marenich

An AI sales director is not a dashboard you read on Monday. It decides the next step after every event in a deal, then makes sure that step actually happens. It routes the lead to whoever is on shift, flags silence at minute five, files a task for the quote your rep promised, catches the missed call, and pulls back a deal that was closed for no good reason. Omnia Lab runs on top of Bitrix24, so you keep your CRM and your team keeps its habits. Reps see normal tasks and reminders; you see a short list of deals worth your attention today.

How is an AI sales director different from sales analytics?

Reporting describes the past. This works inside the present, while the deal is still alive.

  • A dashboard tells you the median first reply is 32 minutes. This system intervenes inside those 32 minutes — at five, at ten, at twenty.
  • Every event recalculates the next step: a new message, a call, a stage change, a close. Not a note in a report, but a task with an owner and a deadline, followed until it's done.
  • Decisions land as ordinary CRM objects: tasks, deadlines, timeline comments, a deal moved back into the pipeline. Nobody has to learn a new place to work.
  • One live signal per thread. A new reminder physically deletes the previous one, so the chat never becomes a feed people stop reading.
  • Call and chat analysis is one input among many, not the product. A signal only matters because an action follows it.
  • It sits on top of your CRM instead of replacing it. Bitrix24 stays Bitrix24 — what changes is what happens between its events: every lead gets a next step with an owner and a deadline.

What happens to a lead from minute one to close?

Real timings from the platform. None of this needs a manager to press anything.

  1. 01

    9:40 — a lead arrives from a classifieds marketplace

    The deal is created and the system reads what's actually in it. Platform service notices like "the user viewed your number" don't start the response clock and don't reach the floor. If the whole lead is nothing but a platform card, it goes to a delayed quarantine. Any real line typed by a human wins, and any failed check is resolved in the client's favour.

  2. 02

    A ping with a Take button

    The lead goes to the priority new-leads chat with a Take button. If nobody takes it, distribution by shift schedule kicks in — only to people working today, and only to people not currently suspended from new leads.

  3. 03

    The clock starts

    Start is deal creation, stop is the first real contact. Only working seconds count, capped at three days.

  4. 04

    Silence: 5 / 10 / 15 / 17 / 20 minutes

    Minutes 5, 15 and 17 go to the sales team chat. Minute 10 is a private message to the rep. Minute 20 reaches the head of sales personally and the leadership chat. Overnight the ladder is cancelled rather than deferred, so nobody wakes up to an avalanche.

  5. 05

    The rep replies and everything stops

    A reply, a call, the "No answer needed" button, or closing the deal all count as a touch and clear the whole chain. A sticker doesn't lift the SLA. A file or photo from the client is a full message; a file from the rep counts as an answer.

  6. 06

    "I'll send the quote by 3"

    Trigger words in the reply create a CRM task with a deadline. Thirty minutes before it: a private ping. Fifteen minutes late: private again, with Done and Postpone buttons. Only after an hour late does anything reach the team chat. There is no public shaming before a deadline, ever.

  7. 07

    A check before every ping

    The AI re-reads the thread right before reminding. If the quote is already sent, the task closes itself and no ping happens. If a deadline has been postponed twice or more, that becomes its own signal in the Important notifications chat.

  8. 08

    "I'll call you tomorrow afternoon"

    The client's promise becomes a task too: "Client promised to come back", due at their stated time plus 30 minutes. "Let me think about it" with no time creates nothing. If the client comes back on their own, the task closes itself, and the reminder suggests a reason to reach out rather than a reproach.

  9. 09

    A missed call

    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 conversation of 15 seconds or more closes it — five to eight seconds of ringing does not.

  10. 10

    The client goes quiet

    Follow-ups on day 1, 3, 7 and 30, delivered daily at 15:00 as one private digest with buttons. One list for the owner of the deal, not ten messages in a group chat.

  11. 11

    The rep closes it as junk

    Every close to spam or lost is reviewed against the full history: messages, comments, the card, the transcript of the recent call. The verdict is stored as its own record, not a line in a comment. An unjustified close goes back into the pipeline — once per deal, with a decision log.

  12. 12

    Nothing disappears quietly

    If a quarantined client writes or calls again, they come back automatically. A deal with a prepayment is never junked and never moved to the cold pipeline.

What does the system do every day, hour by hour?

All times are Moscow time. This runs every day, with nobody asking for it.

Every minute
New deals stuck on a service account are found and distributed
Every 2 minutes
Phone system polling for PBXs that don't push call-ended events
Every 30 minutes
Lost-message recovery and two-way sync of the "replied" status
10:00
Deals of departed reps reassigned: up to 25 deals, early stages first, manual assignments left alone
10:05
Overnight leads delivered as one digest instead of a burst
Hourly, 10:00 to 21:00
Unanswered-client summary with per-client buttons, up to 15 of them
11:00 and 16:00
"No next step" block: deals that got a reply but have no open task
15:00
Follow-up digest for silent clients on day 1, 3, 7 and 30
18:00
Daily review of active deals across four axes: rep, client, deal, process
20:05
Team chat and Important notifications cleared; messages written by people are untouched
20:50
Evening summary for management: promises made, promises missed
20:59
Confirmed spam moved to cold quarantine, with a report
Monday 10:10
Weekly analytics: personal reviews where every conclusion rests on a quote
More than 3 hours of silence
No inbound client messages during working hours triggers an alert — usually a broken integration

What do the owner and head of sales actually see?

You don't scroll the pipeline. You get a short list and the reason each deal is on it.

  • "Needs attention today" opens the report: priority scored 0 to 100, threshold 70, at most 10 deals. Not every problem — the ones worth walking over to someone's desk about today.
  • The rep's score is 0.6 × conversation quality + 0.4 × process. Rep score and deal score are separate metrics in separate columns and never get blended.
  • The 20:50 evening summary shows who promised what to clients and what slipped. The Done button works straight from Telegram and closes the task in the CRM. Only the task owner or a whitelisted admin can press it.
  • Conversion is shown two ways — actual and cohort — and you can drill from any number down to the individual deals. Alongside it: response time, funnel, junk-rejection, at-risk deals, win/loss, and the promises register.
  • It reads its own copy of the data instead of querying your CRM live: refreshed hourly, fully reconciled once a day, 90 days deep. Your portal doesn't slow down because someone opened analytics.
  • A prepayment is always counted in the month it happened, no matter what happens later; if the deal then falls through, that shows up as its own separate metric. Anti-gaming rule: a deal must sit in a stage at least 20 minutes to count.
  • "Reviewed by the head of sales" is how a risk gets closed honestly: five reasons across two classes. "Work has been done" holds until a new risk appears; "I disagree with the system" holds until the risk worsens, 14 days maximum. The mark is reversible and everything is audited.
  • Each rep gets a morning plan as one private message: deals needing attention, promises due to clients today, tasks for the day, and recommendations with buttons.

What does the system deliberately refuse to do?

Half the value of an AI sales director is where it stays quiet. Automation that nags and exaggerates gets switched off within a month.

  • It never shames anyone publicly before a deadline. The first two reminders about a promise are private; the team chat only enters after an hour of delay.
  • It doesn't argue with a deliberate human decision. During a mass pipeline cleanup — more than 30 closes in two hours — returns are suppressed.
  • It never treats haggling, "that's expensive", a discount request, short replies or silence as proof the client can't pay. A mitigating client segment counts only at 75% confidence or higher and only with a direct quote.
  • It doesn't penalise a rep for not pushing a genuinely unqualified buyer. Politeness, on the other hand, is scored for everyone, every time.
  • It cancels overnight escalations instead of stacking them for the morning. No reminder lives longer than 48 hours.
  • It never hides a waiting human being from the summary because of a deal stage. The governing rule is fail-open: any failure resolves in the client's favour — the lead still gets routed, the conversation still shows up, filters switch off rather than conceal data.
  • It doesn't mention AI in anything a rep reads. A call-review task is written through their eyes: what to do now, the score and the logic behind it, what was strong, what was weak, one recommendation.
  • It never overwrites human work. Card comments are appended, never replaced, and a deal name is only changed while it's still the default one.
  • It won't ping or spend money analysing a deal whose owner isn't in the rep directory. That directory is the source of truth.

What results do live deployments show?

Deals with a prepayment
from 4—7% to 13%
Median first response to a client
from 32 minutes to 6.5
Revenue
+33% in four months, with no extra headcount
Rollout
3—5 days
Free audit before you commit
3—5 days

Measured before and after rollout at working clients, including a premium car dealership. We don't name clients. For outside context: Harvard Business Review ("The Short Life of Online Sales Leads", 2011, 2,241 companies) found that responding within an hour yields 7 times more qualified leads than an hour later, and 60 times more than a day later.

How does Omnia Lab fit on top of Bitrix24 and your stack?

  • Bitrix24 is the primary CRM: deal events, stage robots, tasks, timeline comments, custom fields, pipeline and stage moves. amoCRM is also on our integration list — we implement and configure it.
  • It connects your messengers and open channels, your phone system, Telegram, Google Sheets for the call log and shift schedule, and classifieds marketplaces.
  • Standing up a new client takes 3—4 hours with the quick-launch wizard. Until you flip it to live, every module sleeps and messages nobody.
  • 33 modules, each with its own switch: any function can be turned off without touching the rest. 85 settings in the panel apply in under 60 seconds, no restart.
  • Stages, trigger words, VIP criteria, conversation rules and the ignore list are configured for you. Some pipelines and stages simply shouldn't have an SLA, and that's a setting.
  • The daily AI budget defaults to $15, with alerts at 50% and 100%. When it runs out, only deferrable processes pause — client-facing flows never do.
  • A backup AI provider fails over automatically, and two transcription services cover each other. The reason for any failure is visible in the Errors section.
  • Branding: your company name large, "powered by Omnia Lab" small. To your team it reads as an internal company tool.

Частые вопросы

Does this replace my head of sales?
No. It removes the manual policing: nobody has to hunt for overdue promises, verify that a close was justified, or remember who owes what to which client. What's left for a human are decisions about people and about deals that need judgement. At 18:00 the system reviews the day's active deals and hands over a report that opens with at most ten deals scored 70 and above — not the whole pipeline.
Do we have to switch CRMs?
No. Omnia Lab runs on top of Bitrix24 rather than replacing it — it files tasks, writes timeline comments, fills card fields and moves stages. amoCRM is on the integration list too. Your reps keep the interface they already know, plus a few buttons in Telegram.
How is this different from conversation intelligence?
Conversation intelligence analyses calls and hands you a report afterwards. Here, call and chat analysis is just one input alongside CRM events, missed calls and the shift schedule. The point is what follows the signal: a task with a deadline, an escalation, a deal pulled back into the pipeline.
Won't my reps drown in notifications?
One live reminder per thread — a new one physically deletes the old. Overnight leads arrive as a single digest at 10:05, silent-client follow-ups as a single list at 15:00. Promise reminders start private and only reach the team chat after an hour overdue. At 20:05 the team chat is cleared entirely.
What if the system is wrong and reopens a deal that was fine to close?
A deal can be returned at most once, and every decision is logged. Closes by whitelisted admins aren't reviewed at all, and during a mass cleanup — over 30 closes in two hours — returns are suppressed. The head of sales can settle a risk with a "Reviewed" mark and one of five stated reasons, and that mark can be undone.
How long until we see something?
Technical setup takes 3—4 hours; the full rollout runs 3—5 days. Response time usually moves first: at one live deployment the median first reply dropped from 32 minutes to 6.5. We start with a free 3—5 day audit so you see your own numbers before committing to anything.

Read next

Start with a free audit

In 3—5 days we look at your inbound leads, response times and closed deals, and show you exactly where the money leaks. It's free and there's no obligation — even if you stop there, the numbers are yours to keep. Write to us and we'll set up a half-hour call.

Получить бесплатный аудит ↗