Euskaltel – Call transcription
Euskaltel
Sector: ICT
Business Case
The challenge is to improve offer personalisation, reduce customer churn and simplify commercial complexity in customer service. How can we enrich customer knowledge to feed sales and retention models, knowing that the interaction needs to be analysed and automated?
Objectives
Enrich sales models with accurate data. Identify and reduce the risk of customer churn. Simplify the sale of multiple services. Automate database updates without human intervention.
Use case
The solution consists of transcribing and analysing customer service calls to extract valuable information. This is achieved through audio preprocessing and the use of machine learning models (Whisper and PyAnnote). Transcripts are stored and analysed to improve sales and retention models. The implementation is hybrid, combining Google Cloud storage and BigQuery for data analysis.
Infrastructure
Hybrid (Cloud and On Premise).
Technology
Automatic or deep learning Text Mining Voice recognition
Data
Time series, audio recordings of calls. Private datasets
Resources
Internal Staff: Data Science: 100% Business: 50% Business Product Owners 50% Cloud Architects: 20% Infrastructure: Whisper Google Cloud / Azure
Difficulties and learning
Idealisation: Speaker detection (How many agents? Who says what?). | Call splitting with multiple agents, key words Business buzzwords (e.g: Trademarks) | Word boosting
KPIs (business impact and metrics of the model)
KPI enabler: Understanding of the business Quick actions to lower CHURN Training
Funding
Collaborators, Partners
In-house Project
