Presented by Zendesk
Large language models (LLMs) are really nothing new as they have formed the basis of AI initiatives for many companies during recent years. In the wake of OpenAI unleashing ChatGPT into the wild, layperson awareness has exploded, and with it, a wider understanding of the capabilities of LLMs and generative AI. As a result, business leaders are realizing that generative AI is crucial to stay competitive across industries.
“What’s changing right now is the tools we have available, how fast we can go, and the fact that everyone is aware and impatient to start adding AI to their operations or get left behind,” says Cristina Fonseca, VP of product, AI and machine learning at Zendesk. “AI initiatives are famous for their dismal success rates, but when people got their hands on ChatGPT and understood the potential, they also recognized its power — and began to expect AI systems to just work.”
The first mature AI CX (customer experience) solutions are coming into play now, able to tap a store of customer information to personalize user experiences at every touchpoint. That said, getting personalization right in these one-on-one customer moments has become the most important competitive advantage in both B2B and B2C interactions.
“To create AI for the CX future, businesses cannot lose sight of the importance of trust and customer experience when adopting and scaling new AI technology,” she explains. “Building this trust and creating the best customer experience will require personalization and unique customer solutions.”
And over the next handful of years, trust will be the biggest challenge, she adds.
The challenge of trust at every level
“Everyone is excited about the possibility of AI for customer interactions,” Fonseca says. “But if you ask companies if they are ready to have AI automate their customer interactions, the answer will most likely be ‘not yet,’ because we still don’t fully trust the technology and know there are still some risks we need to consider as we implement an automation strategy.”
That includes everything from the potential for ChatGPT models to hallucinate and provide wrong predictions to the growing number of possible cybersecurity risks as more and more data flows between systems. These risks can directly damage brand reputation, all of which make companies understandably wary of going all-in, despite the benefits.
“To build trust, we don’t suggest our customers should automate everything from day one — it doesn’t make any sense, and they’re not prepared for that,” she explains. “Our approach has been first to get the customer’s insights into what’s coming in, build systems able to understand the customer very well and give our customers control over what should be automated and what should not be automated, going step by step.”
That means analyzing and labeling service requests to understand what each entails, whether it’s a request for a refund, a change to an order, problems with payment or a need for information, along with the sentiment and language of that customer. Once the brand knows this, they have insight into what should be automated versus what needs to be escalated to an agent right away and prioritized, with agents having the final say. There are great productivity benefits, but the AI is still only making recommendations, while the humans make final decisions.
Another way to build trust is making the confidence level of AI predictions fully transparent. Every time the AI recommends a reply to an agent, it can offer its level of confidence and whether it has the correct data to offer a good answer. In this way, the algorithms continue to improve with a feedback loop, and the agents continue to build trust in the AI’s responses.
Personalization – or making customers feel seen
Zendesk’s recent report on CX Trends for 2023 found that what consumers want most from their customer experience is personalized service. To Fonseca, that means customers want to be understood in a way that solves their issues faster.
“We talk a lot about personalization in the CX context, but we should be talking about: what do customers really want? What’s the best way to provide good service? This links back to the question around what should be automated,” she explains. “In regards to AI, that’s the fundamental question, understanding the best course of action for every single type of request we can handle. And, of course, leveraging different data points about a customer to make better decisions.”
The more the CX ecosystem is integrated with other systems, from an internal CRM to an order management system and so on, the deeper the level of automation and the better a customer is served. At this point in the CX world, that level of integration poses a considerable challenge, taking a great deal of effort to set up. However, there’s still a sizeable opportunity to eliminate much of the repetitive manual work that still exists, that makes companies inefficient and makes experiences less seamless. Customers often feel like a company doesn’t see the urgency or importance in their request, which can sometimes get lost in the queue of tickets during periods of high volume.
“Even before we look at one-to-one personalization, we need to remember 80% of consequences come from just 20% of our inputs, and if we tackle those very obvious issues that are impacting customer service directly first, we can unlock a lot of gains,” she says. “Maybe this is not the visionary thing leaders want to hear, but at a time where everyone is just trying to understand how to implement AI, I suggest that companies go step by step, change how customers feel cared for — and then we can tackle the issues of data source integration and direct personalization.”
The Liberty London case study
Liberty London, the luxury British retailer, is a good example of the way even just basic automation can change the way customers perceive a company. One of their first use cases was to integrate AI to handle a high volume of service calls related to their extraordinarily sought-after, and quite pricey, annual advent calendar. (Liberty partners with premium beauty brands to curate a collection of products exclusive to the advent calendar.) While the calendar traditionally generated a huge number of sales, it also historically generated a hard-to-manage wave of follow-up questions that Liberty agents struggled to keep up with.
AI distinguished between regular service calls and those related to the advent calendar. Then by determining which calls needed to be prioritized, and automating low-level service calls, managers could better allocate work, agents felt less overwhelmed and out of control, and were able to manage their days more efficiently. More satisfied employees, doing a better job for the customers they served, meant more efficient operations. As a result, the company’s customer satisfaction score (CSAT) rose almost immediately.
“In the things that really mattered, Liberty London was able to be quick and deliver timely support. Sometimes for luxury brands that’s not the expectation — customers really went the extra mile,” Fonseca explains. “This is a great example of leveraging technology to be efficient and deliver an outstanding experience.”
Automating any intelligence that is usually done manually – for example, agents commonly are required to label incoming requests for root cause analysis and reporting purposes – saves an extraordinary time, she adds. Some Zendesk customers have been able to eliminate 80 to 90% percent of the manual work associated with these tasks – and agents are freed from boring administrative work to do far more interesting, fulfilling tasks.
“Double digit automation, even when you’re just addressing the lowest hanging fruit, is usually a no-brainer for most companies,” Fonseca says. “And then because you get powerful insights, you can design the strategy that makes sense for you.”
Dig deeper: Learn more here about how AI is transforming the customer experience.
Sponsored articles are content produced by a company that is either paying for the post or has a business relationship with VentureBeat, and they’re always clearly marked. For more information, contact firstname.lastname@example.org.