The U.S.’ leading B2B telecommunication technology provider developed a conversational AI interface, built on Google Cloud stack, to help simplify contact center interactions. The conversational AI was intent-driven and helped end users navigate the traditional interactive voice response phone trees with minimal inputs, minimizing human agent intervention in most cases.
Our client wanted to create a data analytics layer on top of its conversational AI platform to help analyze customer interactions and get insights into customer satisfaction, agent performance, churn rate, and root cause analysis from end-customer feedback. The client wanted real-time analysis of data, such as the number of daily calls received, how many callers opted to speak to an agent, first-time resolution rates, call duration, and the caller sentiment during and after the call.
It turned to Persistent to create this data layer to feed into an insight dashboard that could help gain visibility into crucial caller-side metrics and make strategic business decisions that positively impact net promoter scores and customer retention.
Persistent builds Google Cloud-native data platform
Persistent’s Google Cloud experts helped create a data platform that sourced inputs from live customer interactions through over 30 real-time streaming data pipelines built on Pub/Sub and Dataflow. The platform could process over one billion messages per day in near real time, creating 600 GB of files in BigQuery, which acted as an enterprise data warehouse and was the single source of truth. A Looker Dashboard displayed actionable insights on caller sentiment, agent performance, and service satisfaction.
The platform, built on a microservices architecture, worked with stringent business rules for data features to be classified for analysis. For instance, call logs should have time stamps in the correct format, accurate contact center number, or time of the call etc. We instituted an elimination process for records that did not have this data and grouped them into rejected tables.
We also enabled archival data reprocessing for thin-file cases, providing reliable insights on new events. A strict audit process governed data sourcing, transformation, and validation, with alert policies defined via notification channels for various metrics (CPU, memory, throughput, and data freshness).
We also built an incident reporting mechanism with Service Now/Jira Service Management to report errors in GCP services such as Cloud Scheduler, Cloud Run, Dataflow, and Cloud Function. The platform was developed using end-to-end continuous integration and deployment practices in Cloud Build and Terraform to accommodate changing business requirements faster.
The new data platform delivers real-time insights
Persistent complemented the client’s GCP-hosted conversation AI model with a GCP-native data platform that helps businesses:
- Analyze caller sentiment, agent behavior, and call satisfaction scores in real time.
- Loop caller feedback into service upgrades and business decisions.
- Timely identify and address service issues contributing to a spike in call volumes.
- Visualize billing usage with call volume.
- Improve caller satisfaction and reduce churn rates with improved service delivery.
- Offer personalized assistance to callers based on historical call records.
