Can AI chatbots 'with a brain' replace customer service reps?

Frustrated by protracted waits to get a customer service agent on the phone?

Help may be on the way in the form of conversational chatbots. These computer programs, powered by the technology behind ChatGPT and other artificial-intelligence platforms, can interact with customers by voice or text.

They are designed to reduce — or even eliminate — the need for a human operator.  The question is: Can they become smart enough to handle the job? 

Unlike ChatGPT — whose gee-whiz tricks (for example: write a diner menu in the form of a Shakespearean sonnet) are grounded in massive “large-language model” data sets with general knowledge and no specific goal — business chatbots are focused on a small, but thorny, problem: cracking the code of customer interaction for individual companies. 

Frank Schneider, vice president and artificial intelligence evangelist at Melville-based Verint Systems Inc., a supplier of AI-powered chatbots and call-center software to 85% of the Fortune 100, said the need is great because human agents are simply swamped.

Big consumer brands and their contact centers  "are so incredibly overwhelmed,” he said. “That’s what we call the engagement-capacity gap. There are just not enough human agents.”

Those issues came into sharp relief in December, when extreme weather forced Southwest Airlines to cancel more than 16,700 flights around the country, leading to call-center chaos for passengers. On Long Island, electric utility PSEG last month reported that customers waited more than nine minutes for a call-center agent in 2022 versus 33 seconds in 2021.

In response, Verint and other technology vendors are racing to deploy intelligent virtual assistants, or IVAs, a term used to differentiate them from dim-witted chatbots whose responses are strictly choreographed. Some IVAs are accessed via text when you enter a company's website, popping up to ask "Can I help you?" while others interact by voice  when you call the customer service number. Their improving language skills and intelligence allow them to share the load with human agents — and maybe even close the deal themselves.

Seeing economic benefits

The economic incentive for businesses is crystal clear. Schneider said that on a per-call basis, an interaction with a human call-center agent runs about $10-$15, while an encounter with a voice-enabled chatbot or IVA costs from about 50 cents to $2.50. 

The median hourly pay of the nearly 3 million customer service representatives nationwide stood at $17.75 as of May 2021, according to the U.S. Bureau of Labor Statistics. Almost 400,000 job openings are projected each year, but the overall customer service workforce is forecast to contract by about 4% from 2021 to 2031. Meanwhile, some companies try to trim costs by using low-wage overseas call centers to serve U.S. markets.

An emerging hybrid model will keep those humans in the loop, Schneider said, with smart chatbots and customer service agents sharing data to speed resolution of consumer issues.

Live customer-service agents can be particularly useful when it comes to dealing with hair-trigger human tempers, a skill that chatbots have yet to master.

“Our patience threshold is incredibly low,” Schneider said, particularly when humans think they are texting or talking to a robot.

A survey commissioned by Verint of more than 1,000 consumers in January 2022 found that 40% felt that chatbots worked well for simple issues like retrieving account information, but a majority were frustrated by having to re-explain a situation to a human agent after starting to interact with a chatbot. The survey also found that:

• 32% said they were rarely or never understood by a chatbot;
• 30.8% said they always or often abandon their efforts to resolve an issue when interacting with a chatbot;
• 54.5% said they always or often must request a human agent after speaking to a chatbot;
• 41.7% said chatbots lack context for their unique situation.

A separate December 2022 survey by the Pew Research Center found that only about a third of the U.S. adults questioned were even aware that AI powers some chatbots. 

Nevertheless, advances in artificial intelligence such as Verint’s patented “speech to intent” technology — which lets voice-enabled IVAs reduce delays  by combining contextual cues from previous call-center interactions with detailed knowledge about a company’s products and services — are enabling more complex transactions, Schneider said.

Listening to customers

Verint has been  acquiring small companies to bolster its AI portfolio, including London-based Conversocial Ltd., in August 2021, and Speakeasy AI, based in Old Greenwich, Connecticut, where Schneider served as CEO, in August 2022. 

A decade ago, companies sought to build a computerized model of the answers to the hundreds of requests a customer might make, creating a static decision tree: press 1 for service, 2 for sales.

“Now, the market has moved to: Let’s get AI in front of the customer listening,” Schneider said. “The way for these engines to become experts is to actually learn from the customer.”

AI, however, needs to be fed data at high volume to learn. Schneider said that Verint's AI systems analyzed 9 billion contact center calls over about a year to improve their responses.

The trick is to combine an understanding of customers’ needs with knowledge of internal company policies, said Tinglong Dai, professor of operations management and business analytics at Johns Hopkins University Carey Business School.

“Customer service is a highly structured business,” he said. “It’s ripe for automation.”

That's why systems like OpenAI's ChatGPT won't be adopted by businesses, Schneider said. "It's tuned and modeled on the world's information and not specific granular elements that matter to a business or a customer. You're not going to be able to get a lot done with it."

At the same time, Verint and some competitors are using ChatGPT technology to rapidly distill the transcripts of interactions with customers into a summary that can be filed in a category.

'We're in a race'

1-800-Flowers.com Inc. grew from a flower shop to a publicly traded corporation with more than $2 billion in annual revenue by being an early adopter of toll-free telephone sales and later e-commerce.

The Jericho-based company adds temporary and outsourced customer service representatives for seasonal business peaks like Valentine's Day. In recent years, the company has been trying AI systems, including IBM’s Watson and, more recently, voice and text chatbots from Aigo.ai.

James McCann, 1-800-Flowers' executive chairman and founder, sees the merger of AI and voice recognition as a powerful path to customer engagement. 

“We’re using AI to anticipate what the customer wants,” he said. “We’re in a race to get there.” 

Many large companies are frustrated with chatbot performance “because the system really doesn’t have a brain,” Aigo.ai chief executive and chief scientist Peter Voss said last year  on The Just Word podcast. 

1-800-Flowers’ approach is to make its chatbot a digital concierge “that knows your preferences” and will have “perfect recall of what your situation is,” he said. "What we're introducing is a chatbot with a brain."

Unlike ChatGPT, that brain will require the agility to navigate a complex conversation with a human and a focus on specific business goals, Schneider said. "The reality is, when you talk to a chatbot or an intelligent virtual assistant or a digital concierge at Holiday Inn Express … you want to ask: 'Can I book a room with points plus cash?' You don't want to ask, 'Can you write a poem as if you're Emily Dickinson but have it be about the Eagles in the Super Bowl?' "

More detailed human-chatbot interactions also can work to reduce friction, reducing the time of a call even if the customer eventually gets transferred to a human agent, he said. 

Turning contacts into sales

Though brands that require armies of customer service agents have led the march toward virtual agents, smaller businesses are getting in the game with monthly plans that can run $200 a month or less.

For instance, Melville marketing and public relations firm The EGC Group has launched a chatbot and marketing analytics spinoff called Raydeus. 

Its chatbots are used by clients including optical chains, dermatology practices, franchisees of Edible Arrangements and the Mayo Clinic's diet website, said EGC president and Raydeus founder Nicole Penn. 

Client companies have the option of using human agents in concert with the Raydeus chatbot or giving the chatbots a warehouse of answers to likely questions that are assisted by artificial intelligence, Penn said.

"It helps your website convert [customer contacts to sales] dramatically," she said. "It's a conversion aid. … It hasn't hit Main Street America, but it will."

Schneider said the simplicity of installing some new  IVA systems also is helping to push adoption down to small business.

"That reality is already here," he said. "These kits can empower ... the pizza shop owner  to [do this] on their own. They don't need a consultant from Accenture."   

Mega-tech companies like Amazon.com Inc. and Alphabet Inc., the parent of Google, also are marketing digital assistants to business.

For instance, Google Business Messages is a chatbot that connects to customers through a company's website, Google Search results or Google Maps.  

  When a user clicks on a messaging button, they start a conversation with a chatbot or a human agent. 

Johns Hopkins professor Dai sees a "big-time revolution" as increasingly sophisticated AI chatbots handle routine requests that can steal time from everyone from doctors to small business owners. 

"It's going to fundamentally change the way we think about chatbots," he said.

Health care, legal services, accounting and even education are some of the fields where he sees potential for smart chatbots. 

"I love teaching," Dai said. "But what I don't love is too much customer service" in addressing administrative or basic questions from students that could be handled by the digital equivalent of a graduate student assistant. 

For instance, in January, Northwell Health announced the launch of an "artificial intelligence-driven" chatbot for pregnant women designed to educate and identify urgent issues such as preeclampsia, a condition involving high blood pressure. . The chatbot was developed with Conversa Health, a unit of Boston-based Amwell, a provider of digital health care systems.

McCann also sees the day when chatbots inhabit devices all around us, even in the lunch room.

“Why can’t I walk up to a vending machine … swipe my phone and say, ‘Snickers bar?'”

By Ken Schachter

kenneth.schachter@newsday.comkschach

Ken Schachter covers corporate news, including technology and aerospace, and other business topics for Newsday. He has also worked at The Miami Herald and The Jerusalem Post.