Wise Insights

Why keyword-based customer service automation is no longer enough

illustration1.png

As a customer service leader, your goal is to provide an excellent customer experience in spite of limited resources. Automated response and classification technologies have the power to analyze huge amounts of data and simultaneously release customer service agents from the burden of manually sorting tickets and typing out the same redundant responses. However, it’s essential to stay in control of that automation in order to maintain the highest level of service at every interaction.

The actions of many automated technologies and some machine learning (ML) applications are often based on keyword triggers—that is, filtering through customer communications for specific words, classifying them appropriately, and then initiating particular actions based on those keywords. This method of automation makes staying in control a challenge because things continually change. With keyword-based triggers, your are constantly faced with:

  • Figuring out appropriate keywords. You need to be intimately familiar with the user experience and know the exact words people will associate with their issues.

  • A world that is changing, along with its language. Words and word usage change constantly. People may use slang or abbreviated terms that a system doesn’t recognize.

  • Keywords becoming more complex over time. An increasingly complex set of keywords is likely required to trigger the same actions, triage process, or even more nuanced results, depending on the issue.

iStock_92459167_XLARGE.jpg

Automation Should Not be Left to Keywords Alone

Let’s imagine that your company initiates a trigger around the word “password.” An agent or simple automated application can easily search for the word “password” in customer emails and response forms, and then trigger a response to send out a password reset form when it finds it.

Another example is the keyword “urgent.” Tickets with that word can be classified as critical and routed into an urgent queue in order to flag the need for a more rapid response.

However, launching one standard action based on a single keyword is no longer enough.

Perhaps one group of people is writing in to say that they forgot their passwords and need instructions for resetting them. At the same time, however, another group of people might be writing to alert you that their passwords have been stolen and that their bank accounts are in jeopardy.

In this case, there must be separate classification procedures and action triggers for the customers who simply need a password reset versus the ones who need more immediate and personal assistance recovering their security.

One solution might be to use a second keyword in indicate a second trigger. For instance, searching for two keywords used together, like “reset” and “password” to trigger one action, and “stolen” and “password” to trigger another.

Still, customers won’t always use those exact words to describe their problems. Instead of saying “someone stole my password,” a customer might say any of the following: “my account has been compromised;” “someone stole my login;” “I’ve been a victim of password theft.” People may even abbreviate the word password to “pass,” adding even more of a challenge in assigning the right response.

Download our Pinterest Case Study to learn how Pinterest achieved 87% first-touch close rates and 250% improvement on CSAT.

The point is that a keyword system is not only challenging to setup and maintain, but that automating responses or classification based on keywords alone can be too broad and ultimately ineffective. The goal is to organize responses in a way that actually fixes the customer’s problem. In the case of an urgent concern regarding a stolen password or identity theft, responding with a simple password reset solution definitely won’t cut it. In this case, replying quickly with a personal tone and solution that honors the customer’s negative experience is crucial.

Sophisticated Machine Learning Elevates Response

Sophisticated machine learning applications are able to use natural language processing technology to infer the customer’s intent, the same way a human would. Instead of looking only at one or two keywords, the machine can analyze the entire paragraph by looking for historical patterns, catching word combinations, and reflecting on the words within the context of the message to detect nuance and tone—just as a human does.

Machine learning also creates a feedback loop based on what the customer agent ends up doing in response to the customer’s complaint, as well as what the customer ends up doing based on the agent’s response. This is how the machine learns, and thus improves its response curation accuracy as the business changes and grows.

In fact, the most sophisticated machine learning technologies can keep learning, reading an entire message and inferring the problem that needs solving with greater precision over time. Unlike a human team which may experience turnover and require frequent training to keep people up to speed, the machine never needs to start fresh. Instead, it can act with the consistency and accuracy of your best employee—one you may eventually trust to respond to sensitive or complicated issues all on its own.

How Wise.io can help

The advantage of Wise intelligence over other automated response tools is sophistication. Wise’s machine learning applications are designed to learn and grow based on response history, context, agent/customer interactions, learning from past behaviors, and constantly progressing toward improved customer satisfaction.

For more information about how machine learning can impact your customers’ overall experience, download our Essential Guide to Automating Customer Service.

The Essential Guide to Automating Customer Services, Wise.io

Topics: Machine Learning, Customer Success, Customer Support