The shared contact centre of a large public transport group handles requests from several hundred thousand beneficiaries: human resources questions, payroll, staff benefits, administrative situations. First contact resolution had plateaued, and nobody could say precisely why.
Rather than audit a sample, the analysis covered the entire flow over two quarters, which came to forty-five thousand conversations and around two thousand hours of audio.
The method
The recordings are first renamed by technical identifier and anonymised before any processing, then transcribed automatically. Each call is then classified against a strict framework: two-level reason, segment, resolution status, satisfaction, repeat contact, friction point. No free text, only categories that can be compared and added up.
A composite model combining severity and volume then ranks the fifteen largest problem groups, with automatic detection of satisfaction anomalies and excess repeat contacts. For each area, systematic sampling of summaries and targeted reading of transcripts make it possible to reconstruct the real customer journey, with every conclusion carrying an explicit confidence level.
The conclusions
The analysis overturned the initial diagnosis. The problem was not the quality of the advisers but a set of systemic causes: case statuses the customer cannot see, fragmented portals, fragile authentication. Five root causes account for most of the volume.
The audit delivered a prioritised roadmap separating quick wins from structural work, an assessment that twenty-three per cent of the flow could be automated, and, just as important, a list of sensitive flows explicitly ruled out of any automation. It also produced a glossary of the language customers actually use, built from their own words and directly usable to equip a future self-service channel.
For reasons of confidentiality, the end client is not named. The figures quoted are those of the engagement.