Getty Images/iStockphoto
Early adopters of agentic AI for customer service offer advice
Two successful early adopters of Salesforce's agentic AI have molded it to customer service business needs.
Separating the hype from reality for agentic AI in customer service can be challenging right now, as the hype-to-reality ratio heavily favors the hype.
Early adopters, however, can offer insights on how to make sound investments in AI technology and use AI to improve service. That process begins, they said, with analyzing processes and assigning measurable KPIs in your service workflows that AI could improve -- and then seeking out the technology to make it happen.
But first, the hype: Respondents to the 7th Salesforce State of Service survey of 6,500 customer service professionals released this week, indicated that they aspire to have AI handle half of service cases in two years. Yet, so far, Salesforce reports single-digit percentages of Agentforce adoption among its customers, and not all of them are using it for customer service.
The AI hype cycle created a "top-down" push from the C-suite on customer service teams to adopt AI agents when Agentforce first came out, said Kishan Chetan, executive vice president and general manager of Salesforce Service Cloud.
It started with a tendency to buy it now and figure it out later. A year later, Chetan said, customers are getting pragmatic.
"I think now people are taking a much more deliberate view -- [looking at] where to have more customer-facing AI so that they can have more and more self-service conversations," Chetan said. "[Also] really having the right context when you move from agents to humans, because I think more and more customers are realizing that humans are needed for handling complex cases, or for bringing empathy that's needed. Having the right transfer becomes very important."
It takes thoughtful planning -- and not just the CEO dreaming about how cost-effective the savings will look on the spreadsheet -- to accomplish such objectives. Two Agentforce early adopters shared their strategies for deploying agentic AI in customer service, as well as their advice for laying the groundwork for success.
Have a business goal in mind
Assigning an AI agent the right processes is key, as is measuring the job the agent does. AI doesn't understand how your customers want to interact with your company; you do, and that's why it's essential to choose carefully what to automate, said Mollie Bodensteiner, senior vice president of operations for Engine, a travel-booking service for small businesses that claims one million customers and $2 billion annual revenue.
Initially, Engine selected a broad range of tasks related to flight, rental car and hotel reservations to automate with its agent, named Eva. That quickly led to a narrowing to just cancellations, because the original scope had overwhelmed the agent with choices and resulted in a lot of escalations. Since Engine decided to focus on cancellations, customer satisfaction (CSAT) scores have generally improved because agents are able to handle many requests instantly.
Currently, Engine customers generate approximately 600,000 service cases annually. Eva has handled 12,000 cases so far and tackles approximately 30% of incoming cases, which can range from 300 to 400 per day. Engine has aggressive growth targets, so more are likely to come as the company scales up.
That said, while some cancellations can be handled automatically, others require a human agent to make a phone call to a supplier. But that was all factored into Engine's original decision to tackle cancellations with the agent. Outlier cases where Eva's automation can't handle a cancellation due to supplier rules or procedural obstacles are quickly escalated to humans. That goes back to knowing the customer.
"We're fortunate in that our customers don't really care how it gets solved, as long as it gets solved quickly and correctly," Bodensteiner said.
Simplyhealth, a private U.K. insurer that supplements the National Health System, focused on automating healthcare reimbursements to customers after they pay cash for appointments. That is the "moment of truth" for the company's AI agent, and Simplyhealth itself, said Dan Eddie, director of customer service.
"That 50 pounds, 200 pounds, 150 pounds, I don't want that coming back a week later when I've got, you know, food to shop for or bills to pay," Eddie said. "I want it back in a day, two days, please."
Automating the process with an AI agent -- trained to flag anomalies that could indicate fraud -- has enabled Simplyhealth to eliminate 160,000 calls per year from the contact center and increase claim satisfaction to 99%, Eddie said.
Simplyhealth also has an automated frequently asked questions (FAQs) section, which addresses questions customers pose through the website's "Contact Us" form. Questions that previously required human answers -- 2,000 per week -- during business hours are now addressed by an AI agent through email 24 hours a day.
The most interesting measure for Simplyhealth might be increased compensation for contact center agents as they take on more complex customer service work that the AI agents can't handle.
"They've had a 35% pay rise over the last three years," Eddie said. "That will continue to rise."
Making AI fit in the workflow
A critical consideration for agentic AI in customer service is when not to use it. It was clear to Engine from the start which travel cancellations required human intervention, a factor also taken into account in Eva's setup.
"When you think about customer experience, you probably don't want to talk to somebody if you just need to cancel a hotel," Bodensteiner said. "But if your flight gets canceled and you're sitting in O'Hare [the massive Chicago airport] and you need to get rebooked, you probably want to talk to somebody, real time, immediately. That requires a level of empathy versus potentially going through an agent. That's what we're really putting our human agents on -- more impactful, human-empathy-led cases they need to support."
Engine treats its AI agent like an employee. In fact, she's held to higher standards than humans, Bodensteiner said, during performance reviews and manual quality-assurance analyses. Engine applies the rigor in part because it's still figuring out where the agent "spins out" and doesn't escalate a case right away. Bodensteiner's team wants to improve on that particular performance metric.
Reviewing service logs from the AI agent, just like it does with humans, Engine discovered a pleasant surprise: Eva is multilingual, having spoken in both French and Spanish to customers.
"It's always fun to read the logs and see when they thank Eva," Bodensteiner said.
In the future, Bodensteiner plans to develop its AI service agent into a more comprehensive travel concierge, capable of delivering more personalized search results to customers. For example, AI that can show hotels based on a user's preferences, such as amenities or vendors with which they are accumulating rewards -- all within company budget policies. The company is also piloting agentic service over the voice channel.
Simplyhealth's Eddie said that governance is key to success in deploying agentic AI for customer service. The company's AI Forum, co-chaired by Eddie and chief technology officer Tim Gough, has streamlined several AI usage policies across the company into one. It also oversees the rollout of new AI tools; for example, the email FAQ was initially limited to a few people and a limited number of questions, as the team monitored the tool's performance.
"The responsibility was on us, so data governance was the starting point," Eddie said. "The gradual acceleration was based on the data saying, 'This is working, and it's working well.' It's all part of the journey we've been on."
Don Fluckinger is a senior news writer for Informa TechTarget. He covers customer experience, digital experience management and end-user computing. Got a tip? Email him.
