In 2023, an explosion of breakthroughs in AI and large language models (LLMs) ignited new advances across industries. For customer service in particular, AI is offering companies the opportunity to tackle longtime customer service challenges — like how to connect with customers in a wide variety of languages — and create richer, more personalized customer experiences that would’ve been unimaginable just a few years ago.

Customer service has come a long way over the years, from call centers to real-time chat features and AI agents. The industry, however, is nowhere near the finish line, and both companies and customers are bullish on the role AI can play.

A 2023 survey of customer service professionals conducted by Hubspot revealed that 78% believe AI helps them spend more time on the more important parts of their roles, and 52.4% of American consumers said they’re optimistic about the future of customer service thanks to AI, according to a recent survey from Infobip . Specifically, they anticipate the integration of AI will eliminate wait times, allow for 24/7 support, and reduce the reliance on time-consuming phone calls.

As this year of impressive innovation comes to a close, it’s the perfect time to think about some of the current customer service challenges and opportunities facing both businesses and their customers. Here’s a look at some of the most significant ones — and the role AI can play.

Personalizing customer interactions

One thing humans in customer service do really well is make small adaptations based on customer context, according to Grant Oyston, Product Marketing Manager at Ada. AI may not offer the same level of empathy as a human, but thoughtfully providing AI tools with more context about the humans they're interacting with makes it possible to offer all the benefits of AI-powered customer service with more of that human touch.

In fact, AI’s ability for tapping into and analyzing large stores of information makes it well-suited for contextualizing a customer’s unique situation in order to serve them accordingly.

“Ideally, the AI agent knows all the stuff this person has experienced in the past, how it was solved and what steps were taken, so we can pick up where we left off. I think that's something that's lacking but quickly coming down the pipeline,” Oyston said.

For a customer who reached out recently or has had to repeatedly get in touch about the same problem, having to start from scratch with each interaction is frustrating. It can be a huge barrier to actually solving the issue. AI tools and agents should “know” about the customer’s history with the company, including how long they’ve been a customer, the products they use, and their past customer service interactions.

For example, an AI agent for an internet company should provide different instructions about resetting a router to a first-time caller versus a customer who's had the same problem three times in the last month. And if the AI agent knows a customer is visually impaired and using adaptive technology, it can adapt the service based on this context.

According to an Adobe survey of consumers and marketers, improved customer experience is the top benefit of personalization. Even for human customer service agents, 64% said AI helps them personalize their correspondences, according to HubSpot’s 2023 State of AI report .

Beyond supplying both human and AI agents with more context, AI itself can be instrumental in collecting and organizing information. AI provides a great way to get needed context directly from customers by just having a conversation. AI can organize that information, compile notes, and create summaries.

There are even benefits for support documentation. For a technical support specialist helping a customer with a unique issue, having to scour through various articles and manuals for answers can be difficult and time consuming, whereas AI can more easily scan through and synthesize large amounts of content.

“AI can refer to a large quantity of data so quickly,” said Oyston. “It has so much data at its fingertips.” The challenge is getting the right information to the AI model. 

Retrieval-augmented generation, known as RAG, is an emerging technique for working with LLMs . It involves directing a model to retrieve information it wasn’t trained on so it can consider that information in its response. RAG is a useful technique for fostering these types of context-empowered, more personalized customer interactions, but dumping in everything a customer has ever said and telling the model to solve a new problem doesn’t necessarily work.

“We need to retrieve the specific parts that are going to be relevant to help us solve this particular issue,” Oyston said. “So that, I think, is the next challenge as far as figuring out what context is relevant. It's about giving information that's actually going to help in the moment to solve the current problem they're facing.”

Serving customers in more languages

While 88% of support teams offer customer support in more than one language, just 28% of end users say they actually see support offered in their native language, according to an Intercom survey of 170 non-native English speaking SaaS customers and 135 support team leads. Yet the benefits of offering multilingual customer service are abundant.

Being able to interact with customers in their native languages reduces misunderstandings, fosters trust, improves customer satisfaction, and can enhance global reach and market opportunity.

“Communicating in their native language, it makes them feel more respected, more like a valued customer. It shows a level of commitment to offer the highest level of support,” said Oyston, adding that especially for global businesses, expecting that all of your customers are comfortable in English is a “big, big risk.”

Even for customers familiar with English, offering support in their native languages can make a significant difference in their experience and customer loyalty. In a 2020 CSA Research survey of consumers in 29 countries, 40% of online shoppers said they won’t buy from websites that aren’t in their native language. And in the B2B space, 29% of businesses said they’ve lost customers because they don’t offer multilingual support, according to the Intercom survey. What’s more, 

70% of B2B end users respond that they feel more loyal to companies that provide support in their native language.

Offering support in customers’ native languages also sets the stage for smoother, more understanding interactions. Among B2B customers who responded to the Intercom survey,

62% said they're more likely to tolerate problems with a product if they can interact with support in their native language, while 58% said they’d be willing to wait longer for customer support if it was available in their native language. On the other hand, 35% of end users would even be willing to switch products to one that offers support in their native language.

The challenge is that offering customer support in numerous languages has historically been extremely resource-intensive and logistically difficult.

“If you're a global company, to offer human support staff who speak 40, 50 different languages, especially if you're trying to staff that 24/7, it very quickly becomes unimaginably expensive,” said Oyston. And even beyond being cost-prohibitive, it’s incredibly difficult logistically in terms of hiring, scheduling, and getting the customer to the right representative speaking their language, during their working hours and without a significant wait.

This is where AI presents a huge opportunity. AI is already proving to be adept at translation, with various models trained specifically for this purpose and capable of instantly translating text and speech into a wide variety of languages almost instantaneously. As AI translation technology improves, and as efforts to train LLMs on more languages than just English-language data push forward, it can be a game-changer for customer service.

These capabilities will not only make it possible for businesses to seamlessly connect with more customers in their native languages, but can also help those businesses better strategize their support and even expand into new regions. For example, if 3% of a business’s customers speak Albanian, the company might wonder if providing an Albanian support staff is worth the cost. This has historically been difficult to measure, and it’s a chicken-and-the-egg problem because the business may not be experiencing enough growth in Albania to warrant Albanian language support because it doesn’t have the language support. 

“We often see companies begin rolling out AI support in these languages and then measure the demand,” said Oyston, describing how AI can help companies navigate this type of dilemma. “So if they're not sure if they need to hire human staff who speak Albanian, well, let's start by offering automated support because that's a low cost barrier to entry to experiment with. And then measure the demand to see what percent of our customers are choosing to speak in that language.”

Offering richer customer service interactions

While exchanging words with customer service agents — human or AI — often leads to resolution, sometimes these interactions would be a lot easier if we could better tap into more dimensions of how we actually experience the world.

For example, last month Microsoft announced a partnership with Be My Eyes to offer an AI visual assistant tool to help blind users quickly resolve issues. For a task like setting up a new laptop, the blind user can take a photo and OpenAI’s GPT-4 vision model will generate descriptions of that photo and use AI-based natural language conversation technology to speak directions out loud. 

“After testing the tool earlier this year on Microsoft users, Be My Eyes said the tool resolved inquiries in four minutes on average, which is less than half the average call time with human agents. In addition, only 10 percent chose to talk to a human customer service representative after interacting with the AI tool,” reported The Verge .

In this case, the use of AI to bring together images, text, and speech is providing much-needed accessibility. But more broadly, the introduction and combining of additional modalities like images and video has the potential to offer vastly richer, more immersive, and more helpful customer service interactions.

“Most of these machine learning models are trained on huge quantities of text,” said Oyston. “What if we could train them on a combination of text, images, and videos? What would that mean for customer service? Imagine if next time a part in your car's engine needed replacing, you received a fully personalized, automatically generated walkthrough video explaining how to solve your precise issue.”

Recent advancements in AI are quickly ushering in this reality. Earlier this month, Google announced its long-awaited Gemini model, which was trained on and can handle combinations of text, image, video, and code prompts. Up until this point, multimodal models have existed only in limited and premature capacities where capabilities around other modalities like images were essentially built on top of text-based models after the fact.

This could soon pave new opportunities, with companies like Meta, OpenAI, and Microsoft all believing advancements in multimodality will finally enable them to create smart glasses people actually want, according to The Information . AI-powered augmented reality glasses have already proven useful in specific use cases like helping industrial technicians with repair, and they could open up richer, more personalized customer service, too.

“There's so much to be excited about that we still haven't fully solved,” said Oyston. “Not just for customer service, but for AI in general.”