Login
Register
Read how Telekom Malaysia leveraged machine learning and TM Forum's AI Maturity Model to develop a diagnostic and advisory solution for faster resolution of customer problems.
Telekom Malaysia leverages machine learning to slash contact center queue times
How important is great customer service? In a recent survey, bad customer service was the 'number 1' reason why 39% of people canceled a contract. In fact, just a single negative experience—such as having to sit on hold—was enough for 18% to churn. Telekom Malaysia Berhad shows how operators can fix the problems that frustrate customers, including slashing contact center queue times by up to one third and increasing service desk troubleshooting accuracy to 98%. In the process, Telekom Malaysia is saving over RM780,000 (about US$190,000) each quarter through fewer truck rolls, and greater efficiency and productivity by contact center and service desk staff. Telekom Malaysia didn’t have to look far for help. It partnered with Telekom Malaysia Research and Development (TM R&D), an innovation hub of the Telekom Malaysia group, on Intelligent Network Diagnostic & Expert Advisory System for Service Desk (IDEAS+). Using technologies and practices such as machine learning and DevOps, the IDEAS+ platform supports Telekom Malaysia’s unifi and pre-unifi fixed broadband and mobile services for consumers and enterprises. “It provides an effective way to diagnose problems, offers simplified ‘next best action advisory’ to all front liners and also recommends second-level escalation to the respective units, if necessary,” says Mohd Khairi Mohd Yunus, Senior Experience Architect, Telekom Malaysia Berhad. “This tool really helps all front liners solve unifi problems faster and more efficiently."
Machine learning helps agents work smarter and faster The IDEAS+ project launched in August 2018 with four objectives: The machine learning in IDEAS+ uses two pattern-recognition algorithms. One is k-nearest neighbors, which uses existing, known variables to understand a new, unknown variable. The other is naïve Bayes, which creates a hypothesis or assumption about a variable based on information it already has. For example, knowing that most people don’t like cold rain, naïve Bayes could look at the current weather and assume that most will stay indoors. These algorithms help IDEAS+ quickly analyze the myriad variables that Telekom Malaysia agents face each day as they field customer inquiries. “IDEAS+ has more than 500 classifications,” Khairi says. “It federates the data, metadata and multimedia from various Telekom Malaysia data sources, which come from passive and active network equipment, network performance and alarm systems, customers, billing and customer applications.” The human brain uses experience to do similar analysis, but IDEAS+ can do it much faster. Now agents are presented with the information they need to resolve a problem faster than if they had to do the bulk of that troubleshooting analysis on their own. This streamlining improves customer satisfaction by reducing the time they spend on hold and speaking with an agent, and by ensuring that the agent selects the right solution, such as dispatching a technician trained for that particular problem.
Hold times and diagnostic times plummet The initial IDEAS+ rollout in March 2019 covered 41 service desk agents. By August, it was expanded to 522 agents, along with 82 TMPoint retail outlets.
“In 2018, AHT averaged 516 seconds, with 240 seconds used by the agents to troubleshoot, diagnose the issues and provide resolution,” Khairi says. “IDEAS+ takes 30-40 seconds to pull data from multiple sources, and another 1-2 seconds for its machine learning algorithm to analyze that data and complete its diagnostics. So we were able to reduce those 240 seconds into an average of 35.70 seconds.” “Based on observation from April to July 2019, 42% of agents completed the same job in less than 20 seconds. Only 2% required more than 120 seconds.” The IDEAS+ platform received a software upgrade on December 3, 2019, further improving resolution times. “Over 61% took less than 40 seconds to solve complaints, and none above 80 seconds,” Khairi says.
The platform’s machine learning algorithm played a key role by making the troubleshooting process less reliant on each agent’s experience. So in addition to making the process faster, machine learning also improved its accuracy, which improves customer satisfaction and Telekom Malaysia’s bottom line.
Here’s how: Before IDEAS+, the service desk had a troubleshooting accuracy rate of less than 60%. When agents were wrong, they dispatched technicians who, over 95% of the time, didn’t have the right expertise to fix that type of problem. “By using IDEAS+, we are able to improve the diagnostic accuracy to 97.59%,” Khairi says. By virtually eliminating the waste of truck rolls with the wrong technicians, and by maximizing each agent’s productivity through machine learning, IDEAS+ is currently saving over RM780,000 (around US$190,000) each quarter . TM expects additional savings from being able to reduce man hours by 75% as each agent can now handle more customers. Machine learning also helps new agents get up to speed quickly by providing diagnostics that used to require extensive hands-on experience.
To successfully implement IDEAS+, Telekom Malaysia’s development and operations teams had to be able to collaborate effectively. “We endorse the scrum-Agile and DevOps framework,” Dr. Zainuridah Yusof, Technical Product Owner, TM R&D says. “The operations engineers who deploy and operate IDEAS+ have no Agile practices and don’t work at the same pace as the Agile developers. The developers create software or migrate it off as ready-to-go to the operations team. The developers will make sure everything is working as it should. “The operations team, however, sees new algorithms and software as the potential for risk and can slow down the deployment process. When there are any communication challenges between the development and operations teams arise, it will slow down during the deployment processes.
“This makes both employer and employee flexible, trusted and willing to learn new skills,” Dr. Zainuridah says. “The employer needs to educate themselves on the Agile framework and DevOps philosophy. Besides that, TM R&D provides DevOps with suitable tools such as Jira, Gitlab and Confluence to achieve every goal.”
A variety of TM Forum assets were used to develop IDEAS+, including the AI Maturity Model, which serves as a guide to how the industry uses AI to realize business value. Aligned with the TM Forum Digital Maturity Model and developed through collaboration between leading players across the telecoms industry, it helps CSPs harness AI’s potential by identifying gaps across six key dimensions of their business: TM R&D also drew on some of the Forum’s best practices and guidebooks to help it fully leverage the AI Maturity Model, including: “TM R&D believes that TM Forum assets enable interoperability and collaboration for a common framework,” Ir. Dr. Abdul Aziz Abdul Rahman, Head of Unit for Data Science, TM R&D says. “This framework allowed us to develop a future-proof platform so that we can easily integrate any external system and support new requirements as they emerge.”
