Call Transcription & Quality Assurance

Situation

This company (CarFinance247) hosts a call centre that receives over 10000 calls a day through Twilio. At the time of this project, Twilio's price for audio transcription was too high, and it was too complex for them to perform dual-channel transcription which was required by the Quality Assurance team who were to be using it for downstream applications. In addition the Quality Assurance Team were using expensive SaaS tools like EvaluAgent and Voyc.

Task

Setup a service to perform automatic call transcription on both channels on the calls, store these & make them available along with the recordings in a single place for the Quality Assurance Team (QA) to access and view.

Action

  • Host OpenAI's Whisper speech to text model internally with Azure Machine Learning Workspace
  • Setup a Function App in Azure to act as a callback from Twilio
  • Push all calls to a Queue to ensure none are lost
  • Trigger transcriptions on the locally hosted model.
  • Connect with internal APIs to retrieve call meta data (agents, customer ids etc.

Result

This dropped the price of transcription significantly (x5) when compared to twilios cost.

6 months later, we came back to this project to drop the costs again by routing all calls to a third party service (DeepGram) for transcription. Cutting costs again by x5.

Follow On Project

Now that the call transcripts were readily available, we were able to complete two downstream applications:

  • Summarisation of calls & making them visible to sales agents
  • Creation of a dashboard to replace expensive SaaS products

For the dashboard, we created a webapp in Azure, that enabled the following:

  • Searching for calls based on meta data filtering
  • Select & view the transcripts of calls
  • Select the relevant 'Scorecard' for grading the call quality
  • Submitting & downloading reports
  • Sharing links for specific reports
  • Colect data generated by QA

Result of Follow On

  • Two expensive SaaS products replaced saving £17000 a month
  • Streamlined QA's reporting process
  • Started to collect QA data for future automation