Customers know they have to do something with AI. But what exactly? Techzine held a roundtable discussion with experts from EvolveIP, Freshworks, ServiceNow and Salesforce to clarify.
We recently discussed that AI is not a “silver bullet” for improving customer service. There, however, we focused above all on the challenges organizations face in leveraging AI. Now we turn to another issue: the real benefit of AI adoption for customer service, after many pain points have been resolved.
AI just makes personal
“How personal is customer service when you’re on hold for 23 minutes?” wonders Joran Klarenbeek, Business Development Manager at EvolveIP. Younger customers often don’t even want to call at all, but rather chat. In addition, the vast majority actually have fairly simple questions, he observes. What if those could get the right answer faster? He cites an example of a tire-changing company. A local branch is not concerned with sales when one is called, because they are busy enough. But tools can make it clear to sales that an interested customer should be called back, Klarenbeek says.
Tip: Customer service in 2024: AI is not a silver bullet
Jeff de Graef, Customer Service Workflow Leader at ServiceNow, argues that this is how you can cultivate customer loyalty. That means the interaction is not just about problems, such as a failure or undelivered product. Contact centers will transform from inbound to outbound thanks to AI, according to De Graef. What does that mean concretely? “Think about offering predictive maintenance via external input of data such as signals from a sensor, or a targeted marketing campaign based on your customer’s interests.” That way, organizations get to speak to the customer in a personalized way that ultimately benefits both parties without having to react to incidents all the time, he thinks. “There’s at least as much value there as there is with inbound,” he says.
“The complexity of customer service is increasing,” said Jan Verbrugghe, Salesforce’s Senior Director of Solution Engineering. “The days when you could satisfy a customer with a yes or no answer are over.”
Lubert Meuter, Senior Lead Solution Engineer at Freshworks, for example, sees that customer service and sales departments need to collaborate more. Because AI gives more time to do other tasks, Meuter says hybrid functions between sales and support are emerging. In other words, this goes beyond the traditional duties of an employee. “People who are positive in customer contact can, for example, buy something extra,” he says. Such an upsell is also “many times” more effective if you do it proactively, Verbrugghe argues. That way, support teams suddenly act as sales departments.
A new interlocutor
Because customer service is overloaded, it soon turns to “deflecting,” says Meuter. In other words, brushing off the customer as quickly as possible. “I think it’s a terrible word,” says Verbrugghe. “Because it actually means ‘I don’t care’. Instead, you should want to help your customers move forward.” And on top of that, Meuter adds, “With AI, you have to reverse this process of deflection. You lose information when you do that.”
De Graef: “I have not often had good experiences with chatbots, but things can be done much better. Think of seamlessly switching to a human employee as well as personalizing the communication based on a customer profile.”
Things do indeed need to get more personal, Meuter adds. “We often talk about the term ‘personalized,’ but its definition depends on the context. Ultimately, the customer has to feel like the experience is about them.” That is currently the main focus point for Freshworks to realize. To that end, AI helps provide service employees with the “most likely right answer,” Meuter said. That can be through a better chatbot than before, the experts indicate, but also with a voice agent, for example.
“We should also not think of a conversation with a voice agent as a phone call,” Verbrugghe said. He suspects that the movement toward voice agents is not only due to Gen Z adoption, but also older generations. This is because of the applications themselves. For example, Verbrugghe describes an example of a mechanic who’s working on a car chassis, thick gloves and all, but can still speak to a voice bot to take action. In short: thanks to multimodality, the usability of such a solution is simply greater than with the classic, less attractive chatbot.
And further, Verbrugghe continues, “A customer service agent doesn’t even necessarily have to speak the customer’s language.” In fact, AI can provide real-time translation, he says. That way you achieve the scalability of a chatbot, but you don’t have to have staff in every language area. That doesn’t even have to be through interpersonal contact, as it happens, as talking chatbots are also an option. Verbrugghe believes this will ensure better customer results. “Sometimes you can explain something better verbally than in text, and that also goes for younger generations,” he says.
AI is less imposing than it seems
An important step toward AI adoption is that it is an orderly process. Verbrugghe notes that fewer and fewer companies are building large language models (LLMs). “Ever more companies are instead working on getting the correct information in as grounding. That data underpins the questions posed to LLMs.” One advantage: “Even unstructured data can be structured more easily these days,” he says. “Think of having AI analyze knowledge bases, SharePoint documents or videos.”
Meuter agrees that models are not necessarily the main stumbling block. He argues that good end results depend primarily on prompting, not the models themselves, and that vendors can stand out in this area. Verbrugghe mentions the importance of the metadata you bring to the model, for example, to tell an AI model how long someone has been a customer of the company. A successful implementation based on such data can lessen the workload for organizations and make them use the solution more. Like a real employee, an AI solution must stay informed about the specific customer.
In addition, De Graef calls for organizations to properly educate themselves before adopting AI. “The technology part is important and is already potent, but you additionally need a strong partner to learn, and grow as an organization in this domain. This is both on the process as well as on the people side. “
“If certain processes become tricky to discuss, then it becomes difficult to provide the right solution,” Klarenbeek said. In short, people are just as important to include in AI adoption as (meta)data.
In the long run: the adoption of AI agents
Where 2023 was the year of GenAI, 2024 appears to mark the transition to AI agents. A number of customers are already adopting agents within Salesforce through Agentforce. “We have only been able to do that by thinking very strictly about what the AI agent can and cannot do,” Verbrugghe says. “You can’t just solve any problem rightaway. Just like you don’t give a new employee the most complex tasks from the word go, you don’t do that with an agent either.” Salesforce’s ambition is extensive, though: by next year, thousands of customers may be using agents every day.
Klarenbeek adds, “Step by step, agents will take on more and more tasks.” De Graef talks about the many ways AI will be applied. “Generating knowledge base articles or ‘next best actions’ are technologically not that difficult to implement,” he says.
“In the demos you see now with AI agents, humans are cut out of the process,” Meuter says. “Ultimately, you have to turn it around: people will dare to let go of certain processes and leave AI to perform them. The problem is really about how you apply guardrails, so you don’t lose your business data, for example. That process demands structure. It’s the only way to ensure an AI agent doesn’t generate answers based on all your data.”
Not everyone will adopt AI equally quickly, Meuter points out. “It will depend on the industry in question. If you’re working with an organization where an email is legally binding, you have to ask yourself if you want to automate that through AI as a supplier.” Regulations do help guide the way, Verbrugghe points out. “At the European level, rulemakers are looking at the proportionality of AI. If your goal is something that won’t have many negative effects, you can go quicker. However, misinform a kidney patient about medical information and you have a problem.”
Still, even in such scenarios, AI can be applied in this process: “You can use text-to-speech to stop having to call the customer in person just to provide basic information. That’s a limited use case.” Therefore, it does take away work without making the AI deployment too broad. Decisions-makers must think very clearly about what can be automated and what cannot, the experts conclude.
Klarenbeek sees room in this for virtually every industry, for example, for automatically rescheduling appointments. Thus there are opportunities for all kinds of industries to adopt AI now, albeit at their own pace, together with partners and without wanting to do too much themselves.