How Machine Learning Can Add Value to Customer Service Automation

This article was originally published on CustomerThink.com as part of my monthly advisor column on February 9, 2017. Click here to read the original.

It wasn’t long ago that we hosted a webinar for our clients that focused on emerging messaging channels for customer service. The shiny new object in the room by far was SMS chatbots. This piqued the interest of many of the attendees with one client actually moving forward with a pilot.

Using SMS as a support channel was cool, but the real goal of the pilot was to implement chatbots and see how many customer interactions could be automated. All of the wind went out of the sails when we realized that thousands of interactions would be required to train the bot and our client’s volume was somewhere in the low hundreds. The pilot was abandoned shortly thereafter and the search for other technology that increase customer service efficiency continued.

Is AI the New Shiny Object?

Much is being written and spoken about artificial intelligence and the seemingly imminent impact it will have on the future of the contact center. When companies are spending somewhere between ten and twenty percent of their annual revenue on service and support, is it any wonder that executives are jumping at the opportunities to add new tools to their technology stack?

The reality is that there’s already a vast customer service marketplace teeming with solutions that are using technology to improve efficiency, agent experience, and customer experience simultaneously. And chatbots are just the tip of the iceberg.

There are a myriad of technologies hitting customer service from a variety of angles. Before you cut straight to automating customer interactions and run the risk of either damaging your customer experience, spending time and money on a tool that goes unused, or both, let’s look at a couple areas where new technologies are solving real problems and saving real money.

Improved Self-Help with Natural Language Processing

Knowledge bases always start out with the best of intentions. They become repositories for storing anything and everything customers might ever want to know about our service or product. Articles get written and added on a whim by the support team, and with each article that’s added, the likelihood of ongoing updates and edits moves further and further down the priority list.

Thanks to The Effortless Experience by CEB, we know that self-help is a support channel. We also know that if customers are unable to resolve their problems in self-help, the effort of channel switching has a serious impact on loyalty.

Good news: New tools exist like Nanorep, Inbenta, and Solvvy that allow customers to search a knowledge base in their own language, with the help of natural language processing instead of traditional keyword searching. These tools eliminate the guesswork out of building the knowledge base by tracking the questions customers most frequently ask. This gives companies an instant priority list of the content that needs to be added and updated.

Another benefit that these tools offer is the ability to place the knowledge base search in front of email, chat, other messaging channels so customers can ask their question before contacting support directly. If their problem is solved, they can mark it as such, and it gets tracked as a contact that was deflected — further proving the ROI of the tool. If it’s not solved, the customer is then seamlessly routed to support.

We’ve also seen significant benefit when agents are given the ability to quickly locate answers to the questions that customers ask. Rather than being dependent on the information provided in training or the help of their supervisor, they can pull the most accurate answers from a central location. A robust self-help tool reduces agent training time and their overall time to full proficiency.

Tailoring Canned Responses

Macros (AKA Canned Responses) typically get a bad rap from customers — and for good reason. Having read tens of thousands of customer survey comments, it’s apparent that customers can smell a macro a mile away, and they are quick to cry foul. Why? Well the most obvious reason is that macros are often used as an attempt at a one-size-fits-all solution in a many-size world.

If a response to a customer isn’t specifically tailored to their issue(s), we run the risk of not resolving it on the first interaction. And I’m not just talking about the step-by-step problem-solving type of knowledge that must be conveyed. I’m also referring to the part of the interaction where the customer’s dog died, or their vacation was ruined, or they lost millions in business. You’re going to have a difficult time writing macros for all of these situations.

While I’d be hard pressed to find a company that doesn’t use macros, some use them better than others. To get the most out of your macros, it’s better to view them like templates or snippets. This give agents a launching point when messaging customers. Training efforts should then morph from a traditional information dump to helping agents specifically tailor responses to each situation.

Along with the message itself, macros also allow agents to automatically edit other information about the case like the status and the department with one click. My favorite field is the one that notes the subject matter or issue type of the case. When we can tie a certain issue type to a customer’s satisfaction or dissatisfaction, it becomes a valuable insight to improving the customer experience.

There are a number of technologies that improve the use of macros including companies like Wise.io and Digital Genius. With the use of machine learning, these systems can accurately interpret what the customer’s message is about, smartly route cases to the appropriate agent or department, and suggest the most relevant response to the agent. This saves significant time spent either searching for the appropriate response or freehanding one from scratch. The cool thing about machine learning is that it learns from your agents over time and becomes more confident and accurate in the response suggestions.

How to Get Real Value from Machine Learning

So it might seem that the next logical step with machine learning is to hand over the keys to the kingdom and just let the computers respond — after all, they are 99% sure they know the correct answer, right? Wrong! This feels a bit like that time when I was in Italy and encountered my first pasta vending machine. Also wrong!

As self-help continues to improve, the customers who continue to contact support are going to be the ones who need both accurate responses AND human responses. With this in mind, I offer some suggestions on how to proceed with our customer service operation.

1. Focus on self-help
   Make sure your customers can find the information they’re searching for in the simplest, quickest, and most accurate way possible. This requires a search tool that features natural language processing. Be sure to train your agents to use it as well to maximize the benefit.
2. Use macros wisely
   Make sure your macro responses are accurate, centralized, and in a style that jives with your brand voice. Then employ a tool that uses machine learning to help agents locate the appropriate response more efficiently.
3. Train your agents to tailor macro responses
   The real value that humans provide, and will continue to provide, is the ability to connect with customers meaningfully. Whether it’s verbally or in writing via the numerous written support channels you offer, teach them to send responses that are emotionally intelligent and aligned with your brand voice.
4. Devote resources to these tools
   Like any tool, these too can become expensive toys that sit on the shelf if we’re not careful. When it comes to managing the knowledge base of articles and macros, this is a great opportunity for your best agents to grow in their careers and touch a whole lot more customers in the process.

In the contact center, technologies that augment traditionally manual and human processes represent a huge opportunity. When you look at benefits like reduced training time, faster answers to customers, and less time spent on manual tasks like categorizing cases, you can begin to see some real value.

I do caution us to pump the brakes on installing that all-inclusive customer service vending machine until we have the perfect case for doing so. There are many opportunities where technology can help deflect customer contacts, reduce agent handle time, and get customers faster support right now. But if we rush to eliminate handle time altogether, and therefore the agents and their human connection, I fear we also run the risk of either having no impact on efficiency, or worse, eliminating customers. Choose wisely!