In a conventional contact centre, quality is assessed by listening manually to a tiny fraction of the calls, under two per cent. The remaining ninety-eight per cent is dead data. There is no way to spot the weak signals in it, the recurring irritants, or the customers about to leave.
The platform built for MyQM turns that ratio round. Every recorded call is transcribed, classified, scored and made usable.
The architecture
The chain is event-driven and decoupled from end to end. Recordings dropped daily by the telephony provider are detected, deduplicated, then pushed onto a processing queue.
Transcription separates the adviser’s voice from the customer’s acoustically and produces structured output. A fallback path takes over on the calls the main model rejects, and a deterministic post-processing step removes the familiar faults of speech recognition, repetition loops in particular.
A second pass, constrained by a strict schema, extracts for each call the reason under a two-level taxonomy, the customer segment, the resolution status, satisfaction, repeat contact, friction points and a score for compliance with the client’s procedure. The reference frameworks are specific to each client project.
From score to action
Data is only worth something if it changes something. The platform provides dashboards of trends and critical combinations, by campaign, segment and product. Each adviser receives a report comparing them with the project averages and ranking their areas for improvement. Finally, a conversational assistant with tools of its own answers managers’ questions by citing the calls it relies on, and never puts forward an answer without a source.
The platform is multi-client and multi-campaign, with role management and a data retention period configurable per client. Quality assessment stops being a survey and becomes a complete measuring instrument.