Customer-service AI, chatbots, and related tools are increasingly adept today at handling basic tasks, such as answering mortgage loan queries or FAQs. But for a variety of reasons, many customers still like to haggle with human beings. Nearly half (46%) of U.S. consumers prefer to speak with a live agent even if a chatbot could save them 10 minutes, according to Usabilla.
To make them happy, human agents are getting help from some of the same technologies that benefit customers. An emerging class of software tools powered by machine learning and artificial intelligence is helping agents respond more quickly and more knowledgeably to more complex problems. Because of these new tools, interactions with callers can be more relaxed because agents aren’t struggling to find answers. Callers can spend less time on the phone and they generally leave more satisfied.
Adoption of AI-powered customer service is still in the early stages, but the technology shows promise. According to a study by Genesys and MIT Technology Review, just 20% of customer-experience executives say they’re using AI to assist agents with customer calls. Of the respondents who implemented AI assistance for service reps, 34% found significant improvements in agent satisfaction.
AI-based agent tools work in a variety of ways. Some use machine learning algorithms to scan knowledge libraries, purchase histories and even the content of phone calls. Others learn by analyzing a caller’s language and tone, and then alerting agents to signs of frustration or anger.
Still other applications, like customer-service software provided by Talkdesk, use machine learning to transcribe voice conversations in real time, creating an instant and searchable record of the call. The technology also identifies important words and phrases in the conversation and then uses them to call up relevant background while the agent and customer are talking.
AI agent assistants crunch all this data into high-level insights about a customer’s relationship with the company and provide agents with step-by-step instructions to address the customer’s problems. “It provides the agent with the next best action based on what the consumer is saying,” says Jason O’Brien, conversational design strategist at LivePerson, which offers an agent help product called Maven. “It expedites what the agent can do.”
Learning on the job
The models are also trained on data from previous interactions with other customers who experienced the same problem, and can guide agents more quickly to proven outcomes. “If the bulk of the people come back two days after they’ve bought a product and have two questions, that helps push the conversation early to the right topic,” says Sander van der Kraan, co-founder of consulting firm Conversed.ai.
Here’s how it works: A customer calls the help desk—again—with a persistent computer problem. An AI monitoring the call transcribes the conversation in real time and displays summaries of previous calls on the agent’s screen. It also notes that the customer was frequently angry.
Simultaneously, the software searches summaries of past calls and the company’s technical library for similar problems. The AI then suggests a couple of options for technical fixes. If those don’t work, the AI, which recognized the customer’s frustration and the repeated problem, then authorizes a repair or replacement for the customer’s equipment.
Customer satisfaction and problem resolution are two metrics companies use to assess the health of their workflows. Another is speed, and not just in consumer interactions. By putting detailed, complex information at agents’ fingertips, an AI-based “coach” can help shorten the time it takes to train new personnel. That’s significant, because 51% of customer service agents spend three weeks or more in training, according to McKinsey research. New agents frequently fail to successfully answer queries because they don’t know the company or its products well enough.
As yet, there isn’t a lot of data about the bottom-line impact of AI agent-assist tools. Common metrics are the first-contact resolution rate, or the percentage of problems that are solved with the first call, and average call handling time. In theory, agents with the right information at their fingertips should be able to solve customer issues more quickly.
Janelle Dieken, senior vice president at Genesys, predicts that the tools will continue to get better and faster at recommending solutions. For instance, the software can already tell an agent whether a customer is more or less likely to make a purchase during the conversation. What if an AI could suggest a customized special offer when its sentiment-analysis engine detects that the customer is losing interest?
“There’s a lot of opportunities to make a difference with the agent experience as well as the customer’s experience.”