Research is clear: customer service is one of the biggest drivers of customer loyalty. In fact, 78 percent of U.S. consumers say customer service is important to loyalty, according to Netomi’s State of Customer Service 2020 report.
Increasingly, customers expect support that is fast, personal, and effective. To deliver the experience that customers expect, companies are adopting AI to provide immediate resolutions that bring customer delight and business value. But in the race to automation, there are seven common pitfalls that companies should avoid to ensure higher customer satisfaction and a successful AI program.
AI is powering the relationship economy
While over 60 percent of U.S. consumers say speed is critical to a great customer service experience and nearly half expect convenience, these expectations are not being met. In fact, more than 50 percent report not seeing any improvement in customer service over the last year, and 23 percent even claim that customer service has grown slightly or significantly worse. With customer service being more influential than ever before, companies are turning to AI to deliver against rising expectations.
Over the last few years, AI capabilities have matured. AI-powered solutions are no longer limited to chatbots with rigid decision trees that limit the user to interacting solely with keywords or buttons. Modern conversational AI leverages natural language understanding and deep reinforcement learning to enable users to engage as if they were interacting with a human.
Virtual agents can now respond within seconds to a variety of customer needs without human intervention. AI is also helping agents work more efficiently and enabling them to focus on high-touch and advanced work. As a result, we’re seeing rapid adoption: according to Gartner, enterprise use of AI tripled in 2019.
AI in the workforce: pitfalls to avoid to ensure high CSAT
While the benefits of AI in customer service aren’t under question, there are strategies companies can implement to help to ensure that AI improves CSAT, agent satisfaction, and overall business value.
Here are seven tips for avoiding common pitfalls when using AI in customer service:
1. Use AI for the right reasons
Prioritize the user experience over delegating more use cases to AI. It’s about the quality of the customer experience, over the quantity of use cases an AI is tasked to manage.
Not every customer query should be automated, especially those that are critical or high-risk. Instead, leverage AI to automatically respond to queries that are high-volume, have low-medium business risk, and have low-to-medium exception management. Examples include order status and refund policies for a retailer, order modifications and cancellation requests for a subscription company, and baggage policies and upgrade requests for an airline.
Companies should determine the ideal use cases based on an analysis of historical tickets. In fact, most companies find that the same 5-7 scenarios account for over 50 percent of all tickets.
In addition to automating the right use cases, companies should also dictate if specific customers should immediately be routed to a human agent. For instance, some companies want to ensure their most loyal and valuable customers always have VIP support from human agents.
2. Let AI help customers help themselves
Most customers prefer self-service for low-risk issues. This requires companies to give AI the authority to help solve these customer concerns.
For instance, virtual assistants can help customers help themselves by directing them to relevant knowledge base articles. Or, they can solve issues within the conversational interface by integrating with business systems like CRM and E-commerce platforms. This allows a company, for instance, to provide the exact status of an individual’s order within the thread.
And if a customer needs to speak with a real person, virtual assistants can always connect that customer to an agent, too.
3. Start with the right channels for your business
Companies often look first to Web chat or messaging platforms to launch a virtual agent. But email is often a better channel to start with. That’s because email provides a unique, real-world training environment. There’s no expectation for consumers to receive an immediate resolution on email. This allows an AI agent to work behind the scenes, recommending a response to a human agent to review, edit, or approve. AI optimizes performance based on the human agent’s actions.
As a good rule of thumb, the rollout of AI on different channels should follow along with how quickly customers expect a response. Following email, companies should look to asynchronous messaging on Facebook Messenger, WhatsApp or Twitter DM, Web/mobile chat, voice channels like Alexa or Google Assistant, and lastly phone.
4. Delegate work between humans and machines
Don’t look at AI as a means to reduce headcount. AI should augment work and automate specific tasks, not replace humans agents. AI can take over the mundane work your agents dread, and enable them to spend more time on projects that require empathy, creativity, and complex problem-solving.
In addition to reserving specific scenarios for human assistance, ensure a virtual assistant knows when to escalate a conversation to a human agent. If a virtual agent can’t understand what a person is saying by the second attempt, or if the person has grown increasingly frustrated, real-time sentiment analysis can signal when a human needs to get involved.
5. Prioritize training
Take advantage of your historical data to launch a more accurate AI. The user experience is better when customers aren’t limited to asking a question in any single way or a finite amount of keywords. However, this also requires sufficient training in order for the AI agent to understand the various ways a person might ask a question.
For example, Comcast found that its customers asked the simple question, “I want to see my bill,” in 7,500 unique word and phrase combinations. You’re not going to be able to anticipate every way a person might ask a question, but the more data that you use during training, the better probability AI will correctly classify a person’s intent.
6. Continuously optimize
Training AI is a continuous process. Optimizing AI over time is just as important as the initial training—you shouldn’t train an AI agent initially and walk away. AI algorithms need to be continuously tweaked and changed.
Reinforcement learning centers on telling an AI when it understood and responded correctly, as well as when it misclassified a person’s intent. This reinforces good behavior and flags wrong behavior, enabling algorithms to adjust for increasingly better performance.
7. Establish the right success metrics
Many companies measure AI like other technical systems, but AI requires other metrics to compare against for success. Conversational AI within a support organization is performing human work, and it should be measured like a human as a result. Look at how AI is impacting CSAT and resolution time, how quickly it’s learning and improving performance over time, as well as how it impacts agent productivity.
According to research by McKinsey, companies that prioritize customer experience can increase revenue by up to 15 percent and customer satisfaction by 20 percent. And many companies are adopting AI to deliver a better customer experience. Avoiding these common pitfalls can help you provide AI-powered experiences that drive satisfaction and, ultimately, business growth.
The original article was published here on February 26, 2020.