This article is grounded in the latest scientific research on AI in customer service. Research insights are from institutions including Johns Hopkins University, California State University, Erasmus University, and the University of Maryland.
Your vice president of customer experience (CX) pushed to automate the front line of support. An AI bot now answers all tickets. It was supposed to be faster and raise your Net Promoter Score (NPS). Instead, customer satisfaction has been dropping.
This story is playing out across organizations everywhere. The research is clear about why: It’s not because AI doesn't belong in customer service; it does. It’s because AI use in most organizations leaves a lot to be desired.
“A poor AI implementation can do more harm to customer relationships than it will benefit your customer service productivity,” says Joseph Taylor, a professor at California State University.
The latest scientific research reveals three approaches where AI genuinely helps.
Research shows that AI improves customer experience most when it's used to absorb tedious work, such as retrieving a customer’s history, surfacing relevant details, or drafting a first-pass response.
A field experiment on 2.56 million chats at Alibaba found that AI-generated suggestions cut chat duration by 4.2%. More importantly, customer satisfaction rose by 5.3%.
Research from ServiceNow found that service reps currently spend only 45% of their time supporting customers. The rest is spent bogged down in administrative work. This opens huge opportunities to reallocate that time to better use.
Using AI to free human agents from mundane tasks allows them to focus on customer interactions, the part of the job where they can make the greatest impact on customer satisfaction.
“Once routine work is automated, what remains for human employees is the harder, higher-value work that’s more complex, more ambiguous, and less scriptable than before,” explains Evgeny Kagan, associate professor at Johns Hopkins University.
Map repetitive work such as retrieving histories, summarizing calls, and searching knowledge bases so that AI can perform these tasks at the right moment. Every minute employees aren't navigating systems is a minute they can devote to listening to customers.
“Continuously test the effectiveness of your AI in addressing your most common issues,” Taylor advises. “Regularly quantify success through feedback from both customers and service reps.”
A study from Stanford University and MIT of 5,172 customer service reps found that AI's biggest contribution was distributing existing knowledge. Specifically, AI learned how top performers handled common problems and provided those responses to everyone else. This sharing of knowledge increased productivity by 14% on average.
“Customer service employees now require more advanced training to handle complex requests,” Kagan explains. “AI can help there, acting as a copilot and helping accelerate training.”
This is another great opportunity for organizations. A ServiceNow survey of 34,000 executives, service reps, and clients found that only 39% of service reps feel they have the tools and training needed to do excellent work.
“Your best reps’ judgment shouldn't stay locked in their heads. It should run through your entire operation,” says Eric Bensley, vice president of customer relationship management product marketing at ServiceNow.
“Work with your team to identify where AI can handle the volume and the patterns, freeing them to do what humans do best: build relationships and make judgment calls. That's what lifts customer satisfaction,” he adds.
Identify what your top performers do differently and build AI systems that share those responses with everyone else to help raise everyone’s productivity.
AI can create value behind the scenes but become a liability the moment it tries to stand in for a human relationship. Across six experiments, researchers tested what happens when customers are told their service came from an AI bot rather than a person. Satisfaction and likelihood to recommend dropped by 8.5%. The “AI” label alone changed how people felt about the same interaction.
The pattern holds across contexts. Studies show AI performs well on simple, routine requests (a login issue, a status check) and struggles when things turn emotional, ambiguous, or personal. Research from ServiceNow backs this up: 50% of customers cite a lack of empathy as their top frustration with AI-driven service.
This happens even when AI is used in the background. “AI-generated guidance can be helpful yet rigid, often lacking the judgment and empathy of human agents,” says P.K. Kannan, a professor at the University of Maryland.
“To succeed, organizations must design flexible systems where AI serves as an input rather than a strict protocol, empowering employees to use AI for speed and their judgment to remain in control during complex interactions,” he adds.
Identify which simple tasks AI can handle alone and which emotionally charged, context-dependent, and complex tasks it should route to humans. The goal is to match the appropriate support to the moment.
“If a case is ambiguous, emotionally charged, high value, or requires judgment, employees should be able to override/adapt the AI recommendation and/or provide human support,” Kannan says.
AI succeeds when it works in the background together with humans.
“No customer seeks out a chatbot when they need help. One of the reasons people seek humans is for that personal touch in a key moment of truth,” says Brian Solis, vice president and head of global innovation at ServiceNow.
“It's not because AI can’t help customers. It’s because the experience customers have getting help is rooted in service automation and cost-takeout principles versus CX principles—i.e., delivering an experience customers value along the way,” he continues. “Design service experiences as customer experiences.”
When organizations implement AI thoughtfully, they can drive greater customer and employee satisfaction and realize valuable outcomes beyond efficiency and cost savings.
Find out how ServiceNow can help you put AI to work for customer service.
