Olga Gushchina
All work

Full product

The rep hears the answer while the objection is still being said.

US founder, SaaS platform for AI sales coaching

What it was like

  • The sales script exists, but in a live conversation the rep forgets it.
  • A manager cannot listen to every call: in practice it is 2–3% of recordings.
  • Feedback arrives a day or two later, when the deal has already gone cold.
  • The brief was not "build a bot" but build a product: from an idea to a live platform with a mobile app.

Where it got difficult

  • The prompt has to land while the objection is happening: after 3–5 seconds it is useless. That means audio capture, speaker-separated transcription and generation, all in real time.
  • Calls live in different places: Zoom, Meet, Teams, phone systems, in-person meetings. The product has to hear all of them.
  • It is the US market, where recording without notice can end in a lawsuit: consent has to be part of the flow.

What was built

  • A live AI prompter: the platform follows the conversation, tracks the stage of the script, recognises objections and shows what to say the same second.
  • Four capture channels: a desktop overlay on top of Zoom, Teams and Meet; a meeting bot that joins from the calendar; cloud telephony; a mobile app for in-person meetings.
  • Push notifications: "call started" reaches the phone, prompts arrive even with the app closed.
  • The script builds itself from the company playbook: the platform reads PDF and DOCX and assembles the prompter with stages and objection handling.
  • Post-call review: summary, score, stages reached, deal outcome.
  • Legally clean for the US: a session does not start without confirming the notice, and consent is recorded.
  • Five languages and a real SaaS architecture: companies, roles, subscriptions, own prompts and documents for each.

How it looks

Click a frame to take a closer look

A live call: the transcript with speaker separation on the left, prompts as the call goes on the right.
The desktop overlay on top of Zoom, Meet and Teams: the prompt arrives while the client is still talking.

Result

100%

of calls visible to the manager as transcripts and reviews, instead of a sampled 2–3%

4

capture channels: desktop, meeting bot, telephony, mobile app

5

languages supported by the platform

seconds

prompt latency: the objection is still in the air, the answer is on screen

For the platform's subscribers the money is in conversion: an average US B2B deal is $2,000–5,000, and every extra deal closed per hundred calls adds about 10% to the team's revenue. For the client himself: instead of hiring a development team in the US, a working product built by one contractor.

Similar problem?

Let's go through it in 20–30 minutes: I'll show what's solvable here and do the math in your numbers.

Next one

The bot booked 120 students. Half of them while the team slept.