Business Objective
Our client is one of the largest P&C insurance providers in the US receiving tens of millions of calls annually into their call centers. Less than 1% of the calls are manually reviewed by the client’s Quality Assurance team. The existing models predict a limited number of call intents. The client wanted to
- Assess the quality of interactions with customers across multiple dimensions and automate compliance measurement using ML models
- Identify customer call intents and opportunities to refine customer interactions for improved call center operations productivity and customer experience
Challenges
- Transcription and diarization errors
- Unavailability of Baseline metrics for several key Business KPIs
- Unavailability of PII data to verify customers’ information and policy
details from transcripts - Insufficient labeled data for some use cases
Solution Methodology
- Leveraged call transcript, keyword corpus data, and compliance rules.
- Leveraged call transcripts, call data, SOPs/FAQs etc. Pre-processed the data using sentence segmentation, lemmatization, etc.
- Leveraged call transcript data and converted client SOPs into scenario database.
- Built a Detection engine to extract call intent, state, and product.
- Generated chunks (combination of conversations) from preprocessed call transcripts.
- The final model provided sentiment prediction for the call chunks which could be summarized at the call level
Business Impact
- Automated authentication compliance measurement of interactions with
customers - Identified customer call intent at the initial stage leading to faster and
accurate resolution - Identified inappropriate transfer of calls
- Automated evaluation of customer sentiment throughout the call
