Traditionally large organizations have leveraged proprietary software like CRM systems to automate customer interaction. However, CRM solutions have not been sufficient enough to achieve the desired customer delight due to their limited diagnostic capability. This was also realized in the case of Airtel — a large telecom organization, where a customer advisor, an app, and a bot were needed to understand the live customer experience and troubleshoot customer issues.
For example, to resolve a customer query regarding slow internet experience, the specific network parameters observed at the customer location have to be captured. Accordingly, the diagnosis is performed. Airtel has a call center team of nearly 15,000 advisors, across various channels, and a uniform, consistent, and precise communication between every advisor and the customer was required. Additionally, the backend systems like network interfaces and routers have high latency as compared to current web response time standards, due to limited technology and infrastructure support. This resulted in increased AHT (average handling time) per consumer interaction resulting in a bad customer experience. Airtel also needed a comprehensive analytics engine to auto-generate a collated report of daily call center actions tagged under various topics.
Airtel’s Decision Tree platform has been developed by the digital transformation team using the latest open source technologies. The platform effectively predicts and diagnoses customer queries and resolutions by capturing relevant information across various lines of business and channels. These include channels like email centers, social centers, call centers, retail stores for businesses like prepaid, postpaid, DTH, broadband, B2B customers, and many more. Airtel has a large ecosystem of its own backend systems. Decision Tree provides a single API abstraction to its clients like a mobile app, chatbot, IVR (automated interactive voice response), and an advisor’s web view. The platform also offers a single view after collating and analyzing the data from multiple backend systems.
Broadband Prediction Flow
The Decision Tree platform has been used for handling nearly 96% of the broadband customer queries landing at the advisor desk. The successful predicted resolution has been observed in nearly 60% of cases for broadband customers.
The platform comprises of key components as shown below.
Decision Tree Components
a) Workflow Flow Engine
The platform has been able to support dynamic changes to its business decision execution via a self-developed workflow orchestration engine. The workflows for various diagnosis paths can be easily configured by the product teams using a web interface. The engine persists the information over an open-source database Mongo DB. The engine then optimally executes the workflow using the advanced graph traversal techniques. The development team has designed the protocol and schema standards. The system resources like memory, CPU, and IO (both disk and network) are consumed optimally when compared to available open-source and enterprise workflow engines available in the industry. The workflow engine is deployed as an spring-based microservice that reads the workflow definition from MongoDB and maintains the workflow instance state in a Redis Database.
b) Rule Engine
The platform leverages the benefits of static and dynamic evaluation of scripts to achieve the performance benefits of required statically compiled code and the flexibility of the runtime compilation of the dynamically configured scripts. The Rule engine loads the knowledge base from a database into the memory, instead of a typical file-based .drl files. The product teams can configure the business rules according to business requirements on the fly. For instance, Decision Tree has been able to cater to requirements for Airtel’s Thanks program easily. The engine provides a mechanism to let the backend system upgrade and migrate to a newer stack, while the core logic remains decoupled from legacy backend systems. The rule engine makes use of open sources framework like Drools to evaluate the rules.
c) Integration and Transformation Engine
The platform lets the engineering and product teams configure the integration with multiple systems with the least boilerplate code overhead. The platform makes use of advanced JSON transformation framework like JOLT. The typical effort spent by development teams to integrate and transform the response from new backend APIs has reduced significantly. This enables the product team to focus on the required data transformations across various systems. The engine uses a pluggable caching framework to act as ephemeral storage. One of the major capabilities of the Decision Tree platform is to fetch the information from the critical systems proactively, cache it, and analyze it for insights using a configured workflow graph. Moreover, in case a southbound systems like CRM observe high latency at their end, the inbuilt Hystrix circuit breaker pattern lets the integration engine to block the calls to CRM, till it recovers. The platform adheres to PaaS standards.
d) Search and Analytics Engine
Searches and suggestions for an exact user query scenario from hundreds of business cases have been quick as search-oriented databases like “Elastic Search” are used by the platform. The configurable analytics dashboard built over ELK and reactive extensions help the business teams to measure the efficiency of the advisors.
The usage of Decision Tree has helped Airtel to reduce the AHT. The time spent on training new advisors has also reduced drastically. Due to its configurable core framework, the Decision Tree application has also found its usage as a “serve to sell” platform that analyzes customer segments and customer preferences. An overall reduction in costs from approximately 70 crore per month to 40 crores per month has been observed at 121 voice centers in the past 2 years. Some of the business metrics post-DT launch have been shown below.
As a multi-tenant product, the Decision Tree has found application in diverse non-telco systems like Airtel Payment Banks as well as human resource bot. The Airtel engineering team is currently endeavoring to develop self-healing systems built on top of Decision Tree capabilities.
The Team :
Amit Parashar, Samarth Sinha, Anirudh Sharma, Shekhar Govind, Saurabh Agrawal, Arjit Gupta, Mohit Bhunwalia