100% of calls audited by private AI, up from 2-3% manual sampling
The client brand and internal business parameters are anonymized under a signed Non-Disclosure Agreement (NDA). Architecture scope, rollout timeline, and verified business KPIs reflect the delivered Indi Vision engagement.
Full call oversight at a fraction of SaaS telephony cost
Every call is checked without hiring more QA staff, and audio processing costs more than 10x less than per-minute SaaS.
- Coverage moved from 2-3% manual sampling to 100% automated scoring
- Each sales and QA supervisor gets 15+ hours a week back
- NDA risk closed: recordings and EDRPOU data never leave the client perimeter
Every rep scored with evidence, every objection turned into practice
Sales leaders see a 0-100 score per call with quotes, churn flags in real time, and a voice trainer built from real calls.
- Intent routing: support calls are not penalized on a sales rubric
- requires_manager_action flag when churn or conflict risk crosses the threshold
- Separate profiles for 7 B2B segments and for new hires vs senior reps
Speech analytics on the existing PBX, inside your perimeter
No number migration, no external call-tracking vendor: audio is processed in the client contour and lands in Dynamics 365.
- PBX-agnostic ingestion from IP-PBX archives and SIP recorders
- Gemini Flash Audio diarization tuned for uk-UA and bilingual speech
- Structured JSON per call into Dynamics 365, workload data into Power BI
Challenge
QA supervisors could listen to only 2-3% of thousands of weekly sales and support calls. NDA rules excluded public SaaS speech add-ons: the calls carry tax IDs, contract pricing and decision-maker contacts.
Architecture AI Cloud 365 (Managed IT), Hybrid Cloud & M365 Migration
Hunter AI runs inside the client's isolated perimeter: Gemini Flash Audio transcribes every call from the existing IP-PBX, a context-aware AI Supervisor scores it, and results flow into Dynamics 365 and Power BI.
Verified Outcome 100%, 2-3%
100% of calls scored with verbatim evidence, 15+ supervisor hours saved per week, and real objections now feed a voice role-play trainer.
Architecture & Implementation Details
Private conversation intelligence for a Ukrainian B2B information market leader
Cloud telephony add-ons had already shown what Conversation Intelligence gives a sales team: automatic transcription, call summaries, mood detection and next-step prompts. For a Tier-1 Ukrainian B2B IT, SaaS and corporate information provider with a large contact center and enterprise account teams, sending tens of thousands of confidential client calls into a multi-tenant SaaS phone cloud was not an option, both for security and for unit economics.
What was broken
- 2-3% QA coverage. With thousands of inbound and outbound calls a week, team leads and QA supervisors could audit only a small sample. Lost deals, unhandled objections and early churn signals went unnoticed.
- No third-party audio retention. Enterprise calls contain decision-maker contacts, company registration numbers (EDRPOU), contract pricing and compliance context that cannot sit in a public call-tracking cloud.
- Real Ukrainian speech. Generic cloud widgets distort live Ukrainian business speech (
uk-UA), mixed Ukrainian-Russian phrasing (surzhyk) and B2B product terminology. - Seven segments, one checklist. Scoring a 45-second license-activation call with the same rubric as a 25-minute enterprise discovery call produces false penalties, and managers stop trusting the scores.
Hunter AI architecture: six modules
Batch and streaming ingestion from the existing IP-PBX archives through the multimodal Gemini Flash Audio API. Speaker diarization (Operator / Client), normalization of Ukrainian and bilingual speech, acoustic loop suppression and language-standard checks.
The call type is classified before grading. Short service calls are judged on resolution and courtesy, not on a sales rubric. Sales calls get a 0-100 scorecard across greeting, discovery, value argument, objection handling and next-step commitment, with a verbatim quote behind every penalty and every best practice.
Separate department and segment profiles for Enterprise, Mid-Market, SMB, Self-Employed, Micro and Public Sector pipelines, with different thresholds for new hires and senior account executives.
Tone is scored for both client and manager (Positive, Neutral, Skeptical, Negative, Conflict). Recurring price and competitor objections are clustered, and a requires_manager_action flag fires when churn or conflict risk crosses the threshold.
Every call produces a structured JSON payload for Dynamics 365 / enterprise CRM: summary, deal-closing recommendations, company name, decision-maker role, tax ID, email and next steps. Power BI shows the workload: clean talk-time utilization, 08:00-18:00 hourly load and call duration distribution.
Top-scoring real dialogues (85+ points) and real failure patterns (<= 40 points) are exported into an AI voice role-play trainer (Gemini Live Audio + AI Call Judge). Reps rehearse hard objections by voice before live deals, and training scores are tracked against pipeline results.
Public SaaS AI telephony vs private Hunter AI
| Dimension | Public cloud AI phone add-ons | Hunter AI (private NDA contour) |
|---|---|---|
| Data sovereignty | Audio and transcripts processed and stored in the vendor's multi-tenant cloud | Isolated private contour under the client's IAM and NDA control |
| Telephony lock-in | Phone numbers and SIP trunks must move to a specific SaaS VoIP provider | PBX-agnostic: connects to the existing IP-PBX, SIP recorders and CRM |
| Unit economics at scale | Per-minute and per-seat markup that grows linearly with talk time | Direct AI compute cost (~EUR 15 per 160 hours of call audio) |
| Funnel and role context | One generic checklist per department | Intent routing, 7 B2B/B2G segment standards and quote evidence |
| Team enablement | Static text tips in the call log | Real objections exported into a live voice role-play trainer |
What changed
Need speech analytics that stays inside your own perimeter?
Indi Vision deploys private AI Cloud 365 conversation intelligence on top of your existing IP-PBX, Dynamics 365 CRM and Google Gemini Enterprise, under NDA and without moving you to a public SaaS phone cloud.