top of page

Mass Communication-How Machine Learning is Optimizing Customer Support -B-AIM PICK SELECTS

Consider all the different ways humans have come to rely on machines to expand their abilities. From calculators to cell phones, industrialization to robotics, humans have been able to achieve more than they ever have before, due to adopting machines. Machine learning is the next frontier in using machines to work more efficiently. And it’s particularly helpful in optimizing customer support.

What is machine learning? Machine learning is “a set of techniques that gives computers the ability to learn1 without being explicitly programmed.” And just like other machines that we rely on to expand our human abilities, computers equipped with machine learning allow humans to do more. Machine learning allows computers to sort through data and calculations much faster than a calculator controlled by a human can. It’s like outsourcing our brain to a computer – but for every cycle our brain makes, a machine makes thousands.

Using machine learning requires a lot of data, or examples to “train” the computer, and a lot of computing power to make connections. With the amount of data being generated by humans every day, and the rapidly increasing power of computers, we’re at a special time in history for machine learning. In fact, “it’s the first time in history we have sophisticated computing power on the one side and huge amounts of data to learn with on the other,” says Lukas Rathman in his Medium piece on Machine Learning for the Rest of Us.

How Machine Learning Optimizes CS

Customer support is a prime target for machine learning because we have a lot of unstructured data. (Unstructured data is information that we haven’t sorted or tagged or told the computer what’s important about it). This unstructured data (generated through our everyday conversations with customers) contains a lot of insights that are helpful in understanding what customers are thinking and doing.

Machine learning can help us do more with the information we already collect through our human conversations. There are four main ways that customer support teams can use machine learning.

Tagging Incoming Conversations

One of the primary methods of automating customer support is to route conversations to the right person, quickly. If you can tag a conversation with what it’s about, what language it’s in or the sentiment of the customer, it becomes simple to tailor the response to that case. This might look like increasing the priority or sending the message directly to someone who speaks the right language or has the product knowledge for that question.

Machine learning can accurately identify the right tag for each incoming conversation by using natural language processing (NLP). By looking for patterns in language and text, machines can “read” or understand human conversation. While they don’t understand it exactly like we do, they can uncover meaning and themes in a wide variety of topics. The more historical conversations the machine gets to “read”, the more accurate its tagging will be. Most machine learning tools will also let humans provide feedback on their accuracy by marking the suggested tags as correct or incorrect. This feedback improves the tagging over time.

Uber built a tool called COTA (Customer Obsession Ticket Assistant) that uses machine learning to categorize incoming conversations2. When customers submit a ticket, COTA “reads” the message, identifies what it’s about likely and sends it to the right customer care team along with three potential solutions. The agent can then review the solutions, personalize the right one, and send it to the customer. They’ve seen really promising results already. “By improving agent performance and speeding up ticket resolution times, COTA helps our Customer Obsession team better serve our users, leading to increased customer satisfaction. Moreover, COTA’s ability to expedite ticket resolution saves Uber tens of millions of dollars every year.”

Ocado, an online grocery store in the UK, uses natural language processing to label incoming messages with tags that help them prioritize customer concerns. Using the Google Cloud Natural Language API, “they were able to label messages with tags such as “Feedback” or “Positive.”” Messages that were urgent due to customer sentiment or because they required a quick reply were filtered to the top of the queue. Ocado was able to respond to urgent emails four times faster3, and saved money on headcount in the contact center.

Magoosh, an online tutor service, uses machine learning to help service agents with case classification, tagging and re-routing. I wanted to highlight them here because they’ve reported that their solution is showing a 92% accuracy for tag predictions – which means that the majority of customers are getting sent exactly to the right place immediately after submitting a ticket. It also greatly reduces the amount of classification work support agents need to do when filing a ticket.

Predictive Support and Advice

For frequently asked questions that only require a straightforward response, machine learning can often predict exactly what customers need to hear.

Bank of America takes this one step further with Erica: their helpful assistant that provides advice for customers4. Providing every customer with a personal financial advisor that digs through their financial information to offer advice would be costly and inefficient. But, Bank of America’s customers were consistently happier when they received service that could help them become more financially secure. Enter Erica. She is a machine learning driven assistant that examines trends and surfaces recommendations for the bank’s customers at scale. For example, she might highlight a recurring payment that you need to review or an opportunity to reduce debt.

The bank hopes to use Erica to streamline mobile banking for their customers and provide a more personalized experience to every single one of their customers.

Customer Insights

When you run a bigger customer support team, it can be almost impossible to decide the most important customer gripe to focus on. One agent might only see 2% of the total volume (on a team of 50) so it’s impossible for them to identify trends. Even in smaller teams, humans struggle to make accurate assessments of trends, due to an emotional attachment to certain customers or features. This is where machine learning can help. By analyzing customer conversation data, machine learning can uncover trends that your team didn’t even think to look for – simply because of the speed that machine learning computers can crunch through data.

For example, Air Canada, an airline company, used machine learning to look through thousands of customer conversations5 initiated during online booking alongside the user sessions stored on their server. By examining customer complaints and common errors, they found several common problems customers had when booking tickets. They could also identify the most buggy devices and browsers. This allowed them to prioritize fixes to make booking flights easier for their customers, improving the customer experience, and saving their contact center the labor costs of supporting these customers.

Improve Knowledge Base by Analyzing Search Activity

More frequently, customers help resolve their own questions by using a knowledge base or help center. This is great because it’s often much faster for the customers to get help and it’s more cost effective for the company providing support. Machine learning can help improve knowledge base through two strategies: highlighting gaps in available content and improving the search functionality to display more relevant help.

Codecademy uses Solvvy AI to analyze customer search trends in their6 knowledge base. By identifying common trends, they were able to streamline and restructure their knowledge base documentation. They also found topics that customers were searching for, but not finding answers on (for a variety of reasons). Closing these gaps helped customers find better help faster, and reduced the incoming load on the customer support team.

Freshdesk’s Freddy chatbot uses machine learning to continually improve the solutions it suggests to customers through chat. Freddy relies on NLP to understand what customers are asking, identify solution articles that have helped customers before, and proactively resolve customer inquiries.

Uncovering the Benefits of Machine Learning in Customer Support

It can be hard to find examples of teams using machine learning, because many teams don’t want to give up their secret weapon. It’s also a fairly new technology, so as more companies start incorporating it and the vendors offering these tools mature, we’ll see more use cases come up.

However, the emergence of machine learning as a critical technology for the future of customer support can’t be ignored. It’s only a matter of time before machine learning is a requirement for mature support organizations – just like a help desk.

Watch this video:

Post: Blog2_Post
bottom of page