August, 2023
The voice of the customer is a crucial parameter in gauging the quality of customer experience. Analyzing service center interactions can help identify gaps in services that can be plugged to deliver a minimum-friction experience.
To gauge the quality of its customer experience, a US networking and digital communication provider floated a three-question survey after every service center interaction. In 2020, it developed an in-house model based on machine-learning (ML) to analyze post-interaction customer sentiment via survey responses. To arrive at an overall score, the model used 25 factors to assess the customer’s mood at the start, during, and toward the end of the interaction. The model was built to help service engineers identify ways to improve customer sentiment. However, after three years, the model was degrading owing to an outdated training dataset, incompatibility with a new single-question feedback survey, and inability to map customer queries with newly implemented solutions or processes. In its current state, the model was operating at 55% accuracy and low explainability levels, did not provide actionable insights to improve customer sentiment, and fell short of identifying areas for improvement based on customer feedback.
The client turned to Persistent to upgrade its in-house ML model to accurately reflect customer sentiment, map service center engineers’ impact on customer sentiment, and offer root cause analysis for problems flagged in these interactions.
Persistent uses Google Cloud to upgrade the client’s in-house ML-based sentiment analyzer
Persistent assembled a team of data engineers and ML architects to study the client’s model and identify opportunities for increasing accuracy and explainability. We re-engineered the existing model to fit the client’s new feedback system.
We leveraged Google Cloud Platform’s (GCP) advanced AI solutions to accelerate model training, iterate quickly and frequently with varied data sets, and enable multi-channel support between customers and agents. GCP’s business solutions for contact center transformation helped us to use natural language processing to identify call drivers and sentiment.
With GCP, we enabled automated, near-real-time ingestion of customer interactions from service centers. We updated the underlying business logic to extract actionable insights by mapping these interactions to the client’s new processes and solutions. We introduced self-learning capabilities that ensured the model stayed relevant with acceptable levels of prediction accuracy. We also helped the client monitor the customer service engineer’s performance based on their interactions on previously open tickets, delivering granular insights into individual impact.
Accurate predictions based on customer sentiments help enhance customer experience
Our ML experts and process engineers improved the accuracy of the client’s ML-based sentiment analyzer by 10%. The model now:
- Accurately predicts customer sentiment based on interactions and feedback mechanisms
- Offers possible resolves to improve customer experience with root cause analysis
- Monitors service engineers’ efficiency with the delta in customer sentiments
- Enables service request engineers to prioritize tickets based on the severity
- Drives customer-focused decision-making that helps align solutions to customer needs
- Delivers improved CSAT scores, indicating enhanced customer perception
Technology Used
- Analytics
- data science development
- ML Ops
- Vertex AI
