The case for and against AI chatbots
AI chatbots may be able to tell a shopper how much to feed their dog, what size pants to order or when their shipment is arriving. But a future where all customer service inquiries are seamlessly handled by computers may be a ways off.
E-commerce retailers are rapidly looking at customer service as one area where they can apply generative AI. And chatbots are among the most promising of applications, according to 39% of CEOs and 33% of management teams surveyed in a March 2024 report from First Insight. Among consumers, the appetite is there. IBM found in its 2024 Consumer Study survey that around eight in 10 consumers want AI to help them get service, answers or resolve issues while shopping.
But anyone who has used a chatbot knows the technology is still maturing. Conversations can be frustrating if the chatbot doesn’t understand the person’s inquiry, or repeats unhelpful answers. The IBM survey found that just one-third of shoppers who used chatbots and virtual assistants were satisfied with the experience. And nearly 20% of shoppers said they were so disappointed they would never use them again.
Here’s a breakdown of what retailers should consider as they experiment with AI-powered customer service tools on their websites.
For: Fast, always-on customer service
One of the best aspects of dealing with a chatbot is how quickly it can answer. “You’re not waiting for an agent to help you,” said Amit Jhawar, CEO of the SMS and email marketing platform Attentive.
In cases where someone has a simple question, a bot can do a better job than a human, Jhawar said. “If I want to reset my password or have a quick question about shipping, a chatbot is a great first line of defense. It never gets tired. It knows all the answers and the use cases and can get through the process with low variability, high consistency and immediate responses.”
Against: Challenges handling idiosyncratic or complex experiences
For all the simple explanations that an AI chatbot can ace, brands may be wise to have human customer service agents respond to higher inquiries. Attentive, which uses AI for SMS and email chatbots, has a Concierge service that kicks complex questions to human agents.
“You have to be aware of AI’s limitations,” said Kyle Landry, founder of the Delavie, a skin care brand. “Simple things, like ‘What’s my tracking number?’ or ‘How many reward points do I have?’ a chatbot can handle. The touch point between the consumer and AI chatbot is maybe one or two sentences.”
At Delavie, simple email inquiries are routed to an automated chatbot to answer. An in-house staffer developed a tool that scans keywords from the email to route routine questions to a chatbot. But questions that are more complex — like concerns about product use or inquires about how Delavie products mix with other skin-care products — are answered by customer representatives.
“You can’t assume AI will be able to answer that. Instead, you get this vicious loop that we as consumers get frustrated with,” he said.
Jhawar from Attentive said many AI models work by having certain pre-programmed buckets of information that are triggered by the customer’s inquiry — whether it’s about a product, shipping, an account or sizing. Based on which bucket the inquiry falls into, the bot will have a series of potential responses lined up in its model to get the answer. “You need a model where a chatbot is taking 60%, 70%, 80% of the questions. But if it doesn’t have confidence that it can answer, it kicks out to a human,” he said.
For: A cost-effective way to grow customer service operations
Like many buzzy tech solutions, AI is frequently seen as a way to reduce costs in retail. Landry said that AI chatbots cost about one-eighth of employing a human service agent. In turn, companies are going to change up their staffing: Gartner analyst Uma Challa predicts that generative AI will lead to a 20%-30% reduction in customer service and support agents by 2026.
Jhawar from Attentive also said that AI-powered chatbot services are effective because of how easily they can scale based on needs. During Black Friday, for example, a brand couldn’t necessarily justify hiring dozens of new staffers just to answer customer questions related to the holiday. The AI chatbot, however, can handle an influx of customer inquiries.
“The amount of people you can process vastly exceeds what a human can do,” he said.
Against: Accuracy and efficacy concerns
Ultimately, a chatbot is only as effective as the inputs it is fed. Otherwise, the service runs the risk of “hallucinations,” or giving the wrong information. Roger Williams, head of loyalty for relationship marketing firm Marigold, said chatbots are not necessarily performing as well as some might expect.
Even if a chatbot does provide an accurate answer, the language, vocabulary and general tone used can make or break the customer experience. Jhawar from Attentive said brand voice is one of the key components it focuses on when creating its AI-powered services. Marketers using its services for text or email can pre-select the messages the bot will use. “Some of the [large language models] not trained on your data aren’t going to sound authentic,” he said.
Consider how some image-generating AI tools have issues with creating hands and fingers. Part of the problem, as covered in The New Yorker, is that the model may not have enough close-up source images of hands and fingers to nail down the replication.
Still, many experts watching the space expect the technology to improve.
Williams from Marigold argued that “some customer service centers are almost as frustrating as some of the nascent chatbots.” He believes that “those chatbots will soon surpass that average human support agent.”
“Once it does that, it’s game over and people will want more bot,” he said.