AI and the future of tech support
As machine learning and chatbots evolve, AI is driving improved efficiency and experience
June 24, 2020 | By: Ram Palaniappan
Anyone who has worked in tech support knows the relatively simple, repetitive requests end up eating up most of analysts’ time and mental energy. While those types of IT support requests are usually easy to solve, there tends to be a lot of them, which adds up to a big drain on the tech support team’s capacity.
Enter AI and machine learning. Although often solely thought of as the cutting edge of technology, artificial intelligence and machine learning can make the most sense for solving day-to-day problems—even those that require communicating with people.
Gaining efficiencies in tech support with AI and machine-learning chatbots
A few years ago, one of our high-tech manufacturing clients found their tech support staff was responding to a high volume of L1 support calls from users about whether particular applications were down, or to reset passwords or provisioning access to new applications. The client wanted to explore solutions through our innovation center of excellence to see whether there was a more efficient way to handle routine requests without impacting user satisfaction.
In response, we developed TEKsystems.sAIge, an AI-based solution that helps end users solve tasks without intervention from the tech support team, take action to fix the problem, or know when to hand the end user off to a technician when the problem wasn’t so easy to solve.
For example, if a user types, “I can’t access the lead management system,” the machine understands the intent of the message and passes on the context to an AI engine to analyze. If the machine is down, the chatbot communicates the reason to the user—or restarts the services through automation scripts that are predesigned to execute tasks.
Scaling the solution for better user satisfaction
After providing our initial proof of concept, our team incorporated new tech support tasks into the chatbot so it could provide support for different types of tasks.
As the system worked on a problem, it asked the end user whether it was solved to their satisfaction. After two or three iterations without solving the issue, the chatbot handed the problem off to the human tech support personnel. Because the chatbot captured the conversation with the end user, the tech support person who received the request had context about the issue—and what actions had been tried to resolve it. This saved time and allowed the conversation to immediately move to more sophisticated support.
AI and machine learning for tech support use cases
There are a multitude of situations that could be helped by integrating an AI tool to respond to users, assess problems and share data with people on the support team. Here are a few ways we’ve helped clients resolve tech support problems with AI and machine-learning solutions.
- Front-line COVID screening support for healthcare provider: The COVID-19 pandemic created multiple challenges and stress to front-line workers. We worked with multiple healthcare providers and deployed virtual front-line screening agents to direct questions in the right direction at the same time and scale as the front-line call center team through AI- and natural language processing-based conversational assistant platform. In four days, we deployed a coronavirus chatbot assessment tool, TEKsystems.sAIge, that leads users through critical initial screenings for COVID-19 to help people determine if they should seek medical attention. The tool is easy to update as CDC recommendations and knowledge about the virus evolve. By helping to manage this significant demand of calls, the chatbot assessment has given valuable time back to healthcare providers. Read more about the COVID-19 assessment chatbot.
- Multilingual global tech support: A manufacturer needed to refine their service desk support model to ensure seamless operations for users outside of the U.S. In smaller countries without a dedicated desktop resource, the client looked to a live chat application. We implemented a multilingual, live chat tool to support 57 countries with 141 languages. End users would receive 24/7 support with instant responses instead of having to wait multiple days. Our client has created over 2,000 knowledge articles, enabling a 79% resolution rate and self-service knowledge for end users, and reduced their speed to answer by 50% and total cost of ownership by 21%. Read more about multilingual chat.
- Accelerating service for insurance claims through auto adjudication: In a competitive market, customer retention is critical—and the key to retention is quality service. TEKsystems.sAIge is helping insurance companies accelerate claims processing through an AI-based engine. One large healthcare provider has increased the auto adjudication of claims processing from 40% to 73% for new claims. This has significantly reduced the manual processing of approximately 400,000 claims per month by increasing the automatic adjudication process—allowing this client to meet the higher volume of demand during the COVID-19 pandemic without increasing bandwidth.
Higher-level machine learning with data analytics and insights
Beyond reaping first-contact efficiencies, you can program an AI application with data analytics and insights to spot patterns that will help your team perform root cause analysis to prevent them. For instance, it collects data and identifies patterns about recurring problems, such as many similar requests, a disproportionate number of requests coming from the same location, or the requests more frequently require human intervention to find a solution. This can help your team fix underlying problems and prevent lots of service requests.
Senior leader, innovator and technologist Ram Palaniappan brings broad experience in big data, BI, mobility and cloud solutions. Ram has worked with Oracle and Deloitte and is a featured thought leader in the space, including contributing to articles in major IT publications, such as Information Week, and serving as a featured speaker at multiple forums and conferences.
Portions of this article appear in an earlier version.