Call-center speech analytics with AI scoring
Call-center conversation quality was checked by listening to random samples: it was unclear at which step of the script agents were losing clients and which agents were underperforming.
Telephony has no call topic — it has to be inferred from the CRM; the external service accepts one dealership per job, and the partner reworked the checklist along the way (11 → 17 criteria).
- Collecting recorded calls from Bitrix24 telephony, filtered by topic via the linked lead's business line
- Submitting to an external AI scoring service split by dealership, status polling and idempotent result storage
- Merging several calls with one client into a single score — at the partner's request, behind a separate flag
- A 17-criterion checklist: answer, weight and total score for each call
- A dashboard: calls, agents, criteria sorted by failure rate, a call card with the recording, CSV export
The pipeline and dashboard are ready: a supervisor sees which script criteria fail most often and for which agents. Regular runs in production start in September 2026.
Related services and solutions
Bitrix24 D7: boxed development and stabilization
I work not only with interface settings, but at the level of the box, the code and internal APIs.
Business process automation
When employees transfer data manually, reconcile spreadsheets, lose requests or repeat the same actions — I turn the routine into code.