The $400 billion opportunity for AI in customer service

Not all AI-powered customer service chatbots are created equal—or created well.

Take AVA, the AI-infused customer support bot that AirAsia introduced in 2019. AVA racked up nearly as many customer complaints as case resolutions, forcing AirAsia CEO Tony Fernandes to admit earlier this year that AVA was Southeast Asia’s “most hated AI chatbot.”

AVA, of course, is not the only sad sack in the chatbot universe. (She’s been replaced by a second-generation AI bot, Bo.) Of customers who turned to a support bot from December 2022 to February 2023, only 25% said they would willingly use them again to solve a problem, according to Gartner.

That could soon change thanks to the meteoric rise of generative AI, which promises to make bots’ chat more human by contextualizing customer requests and synthesizing natural-sounding language. Generative AI will generate $404 billion annually in increased productivity and reduced costs for global businesses, according to McKinsey.

With enterprise spending on generative AI projected to hit $1.3 trillion by 2032, many companies have their eye on the customer experience (CX) prize. In retail alone, 63% of companies say they’re exploring how to use generative AI to improve their customer service, according to Capgemini, while Gartner research shows that nearly 40% of companies across all industries plan to make CX the primary focus of their generative AI investment.

Where should companies invest in generative AI strategies to get the biggest payoffs with the lowest risks? Here are four use cases where customer service experts say generative AI can improve experiences for agents and customers.

“Copilot” for live interactions

The pain of mediocre complaint handling—such as ineffectual chatbots, endless wait times to speak to a human, and inexperienced agents—could lose companies future business, risking $887 billion in future revenue, according to the 2023 National Customer Rage Survey.

But generative AI has the potential to alleviate this pain by providing a digital assistant—increasingly called “AI copilots”—for human call center workers, says Maeve Condell, customer success lead for Ultimate, an AI-powered virtual agent platform.

“As a ‘copilot’ for call center workers, a generative AI-trained assistant can help them quickly access information or suggest replies by linking to the customer knowledge base,” says Condell. “We can pass auto-generated replies as an internal note to the ticket, where the agent can quickly review, rework if needed, and send.”

With little risk, an AI copilot can raise human contact center workers’ game by giving them access to actionable solutions quicker and even suggesting language to communicate that information more effectively—a boon for less experienced workers, in particular.

A recent Stanford study shows that contact center agents with access to a copilot saw a 14% boost in productivity, with new or low-skilled workers showing the largest gains. Generative AI levels the playing field, the study’s authors conclude, decreasing inequality in productivity and helping lower-skilled workers significantly.

Creating and training chatbots

Today’s chatbots often rely on identification of keywords that lets them pluck an FAQ from a knowledge base and plop it into a window, asking antiseptically, “Does that solve your problem?” It’s annoyingly hit-or-miss, and the communication style is flatly impersonal.

While it may currently be too risky to let a generative AI bot directly interact with customers without a human in the loop, their synthetic language capabilities can be used to glow up existing customer service bots. Think of it as “My Fair Chatbot.”

“Using generative AI as a part of the chatbot creation process is one of the most promising use cases at present, and certainly the least risky,” says Benedikt Schönhense, co-founder and head of data science at Springbok AI.

Schönhense suggests that companies use generative AI to paraphrase questions a customer might ask and even generate sample conversations, automating large parts of the training. Further, they can use generative AI to test an existing chatbot by simulating user inputs with prompts from a human tester, with different levels of granularity.

Best of all, the generative AI training process can infuse whatever style and tone of communication the company wants to project, from warm yet authoritative to low-key casual, depending on the customer base.

Tracking interactions over the life of a support ticket

Everybody knows the mind-numbing agony of explaining the same issue to multiple agents over repeated calls. “Personally, I’ve been in support loops where I’ve had the exact same conversation with three or more people to get something resolved,” Schönhense notes.

But generative AI’s capacity to synthesize and summarize is a true superpower that companies can and should deploy in customer service, removing this pain point for both the customer and the support worker.

This is especially helpful when that communication occurs over multiple channels, such as combinations of phone, email, web, app, and social media interactions.

“For a customer support agent to understand what has happened with a customer who is at the breaking point, now you may need to read a five- or six-long email chain or a support ticket with lengthy notes from five or six different interactions,” says Vijay Vittal, product innovation lead with LocoBuzz, an AI customer support platform. “With generative AI, case summaries can be automatically generated to fit in five or six sentences to get the agent up to speed and get the customer off the ledge.”

Training and onboarding new call center workers

The attrition rate of call center employees is brutal; in 2022, the average turnover rate was 38%. And most call center agents—55% of them, according to a Salesforce survey—say the training they receive is insufficient to provide quality customer support.

Much like generative AI can be a great tool to train chatbots, it can also be used to train call center workers with simulated conversations, both familiarizing them with the types of issues they’re being asked to handle and preparing them to use generative AI as a copilot.

“This is absolutely a low-risk, high-reward use case,” says Condell. “Using this tool can get new or low-experience workers up to speed much faster, and it is offline and a safe, low-stress way to train before direct customer interaction.”

Condell adds that one Ultimate customer with complex internal processes says that they envision using generative AI to cut the time it takes to train a new support agent in half.

Today, the human touch is essential for ensuring not only the safe deployment of this technology, but also extracting the greatest ROI from it. Keep a human in the loop: As companies look to mine gold from generative AI, consider it the golden rule.

This article was originally published on The Works