AI in Customer Service: How To Use It – And When Not To

Luke Walker
February 15, 2024

Picture this: You’re a frustrated customer of an eCommerce company trying to return a faulty product. You’ve already spent 10 minutes scouring the website for a phone number before you come across the company’s new chatbot. Worth a try, you think. But it just sends you back to the page you just saw, which directs you back to the chatbot again. The customer journey goes around in circles until you’ve lost all patience and given up.

That’ll be the last time you buy from that retailer...

We’ve all been there. Whatever technology is being used, it’s great when it makes customer service easier, and a real frustration when it just gets in the way. As AI increasingly begins to dominate organizations’ priorities ahead of 2024, this creates a key challenge for customer service teams. While the race is on to successfully implement AI in customer service, there’s a real imperative to do it in a way that actually makes life easier for consumers.

So where do you start?

🔮 Fact vs. Fluff: Mythbusting AI in Customer Service

An important challenge for CS teams right now is to align their tools with business interests. We’re now coming out of an AI hype cycle, with everyone wanting to implement as much tech as possible.

But companies are realizing AI can't do every single part of the job. You still have to align your tools, including those powered by AI, with your business processes and your desired outcomes.

Julian Hertzog
Head of Sales at Babelforce

Since the overnight sensation of ChatGPT, organizations have raced to implement AI technology. Now, 56% of customer service leaders are already exploring new Gen-AI vendors for customer service, according to Zendesk. But the research also suggests that these same CX leaders know all too well that their current chatbots fall short of customer expectations, with just 22% saying they’re akin to digital agents. So what’s gone wrong?

Let’s return to the eCommerce returns example. The problem here wasn’t that AI was being used, it was the role it played in the customer journey and the specific touchpoints it was designed to automate. If the chatbot had recognized the customer’s request and triaged them through to a human agent, the problem could have been resolved in minutes. Often, the challenge is not whether to use AI – it’s how.

The example at the top of this blog is a classic faux-pas of AI in customer service: Using AI for processes it’s simply not designed for. While a chatbot can’t manage a complex, end-to-end returns process – it is capable of triaging a customer to a human agent who can. If you roll out AI on processes it’s not designed for, there’s a real risk you do more harm than good.

Let’s be clear: While intelligent CX certainly includes AI, automation, and data analytics, it’s important to think of those advances not as discrete, standalone elements but as parts of a greater whole. These tools must work together across a business’ experiences.

Zendesk, Customer Experience Trends 2024

At the same time, not all AI tools are built the same. Though Gen-AI products like ChatGPT have become the most well-known, they’re not the only show in town. AI is a whole web of overlapping tools and functionalities, including machine learning algorithms, internal knowledge base automation, voice recognition, automated transcription, and more.

So how do you strike the balance? The truth is, AI is a powerful tool, but it’s not a magic wand. To use it effectively, you need to understand which touchpoints and workflows it can be used for, and what it can’t.

🔧 When To Automate – And When Not To

On average, across all tweets, and regardless of whether the customer used a negative, neutral, or positive tone, we found customers who received any kind of response were willing to pay almost $9 more for a ticket in the future.

Harvard Business Review

Often, when customers get in touch, they’re already frustrated about something that hasn’t gone to plan. Bad customer service can make a poor customer relationship infinitely worse. But on the flip side, good customer service can turn frustrated customers into loyal ones. In fact, according to a Harvard Business Review survey, customer service interactions correlated with increased customer spending even among those customers who were already frustrated before the interaction.

The stakes, therefore, are clear. Customer service is make or break for your customer relationship. In short, there’s a lot riding on you getting it right.

📖 READ: Zendesk: CX Trends Report 2024

For customers, the need is clear: Smooth, seamless, and enjoyable experiences, regardless of the technology being used or the people involved. If AI makes it easier for them to do that, then the customer experience will benefit from it. 

Here are the three main areas where AI can enhance customer service operations:

1. 📋 Task Automation

Customers have made it increasingly clear that they want businesses to use the mountains of personal data they possess to offer warm, personalized experiences.

Zendesk, Customer Experience Trends 2024

One of the best uses of AI in customer service is to automate individual tasks. Whether it’s customer service agents, sales executives, or account managers, every second that technology can save them is a second they can spend making customers’ lives easier. 

The truth is that much of the work that customer-facing teams do is tedious, manual, and surprisingly simple to automate. In many cases, AI is well qualified to take on these tasks:

Retrieving information from a knowledge base, CRM, or other tool. This saves the agent from clicking between different apps and gives them the complete picture of a customer’s journey in one place. 

Customer authentication is another great use case for AI. Here, customers can be pre-screened by chatbots, before being passed over to a human agent if necessary. This is a vital yet time-consuming task that often adds precious seconds to calls and in-person live chats. 

Automated call transcription can also save time and give employees a much clearer, more detailed, and more accessible record of what was said in calls or meetings. 

Sentiment analysis for calls and feedback surveys can also save employees from manually aggregating information - and makes it easier to find the consensus across different responses. 

2. 🤳 Self-Service

Effective automation can also have a much more direct and tangible impact on the customer experience. Recent Zendesk research suggests that as many as 67% of customers prefer self-service to calling a human agent for simple tasks. If an automated process can resolve their issue faster and with less hassle, it’s clearly in everybody’s best interests. 

But the same research also suggests that 40% of customers will call a contact center if they can’t resolve the query via self-service. Clearly, AI has a role to play here, but as soon as it gets in the way, it’s likely to do more harm than good. 

Here are a few examples of tasks that AI can accelerate:

Changing customer information and other basic admin tasks are perfect for automation. Generally, this can be easily achieved via a chatbot or online portal.

Basic search inquiries like finding information about a product, feature, or case study can also be automated. In this case, bots and Gen-AI tools can understand the request more effectively than plain text searches and respond with relevant content in seconds.

Voice-based inquiries are another exciting development for customers. Just like with basic search inquiries, they enable customers to easily find answers and solve problems, only via conversation rather than typing. 

3. ✅ Customer Experience

An effective customer experience requires the smoothest possible journey from end to end. But if sales and customer service teams are spending all their time on complex, highly manual tasks, it becomes impossible to properly deliver this. 

Luckily, there are several areas where AI can help here as well:

Email parsing can reduce the amount of time agents spend manually compiling data from emails. This could include things like names, contact details, phone numbers, and other details. This essentially acts like an intelligent, responsive inbox search engine. 

Pre-screening support tickets also makes life easier for agents, essentially automating the process of deciding which tickets require which responses. This removes the manual work of tagging tickets and ensures agents can save time by resolving multiple similar tasks in one go.

Pre-qualifying leads can also save significant time for sales teams. Chatbots can pre-screen potential leads by asking their name, company, job title, and other important information. This means sales teams can focus on the most relevant leads, and avoid spending their time qualifying ones that aren’t going anywhere. 

📖 READ: Revolutionizing Customer Service With Automation: Turning Tickets Into Seamless Workflows

🙅 Don’ts of AI: When AI Can’t Deliver

Understanding where AI can deliver is a great first step – but it’s not the whole story. It’s also important to understand the limitations of the technology, and where more robust solutions might be needed.

According to McKinsey, some 40% of work has the potential to be automated. But not all automation is the same - especially when customer expectations depend on the successful delivery of a task. The more complex the task is, the more difficult it is to automate. 

For more basic tasks, AI tools and more basic out-of-the-box integrations are usually sufficient. But the more complex the process, the more likely it is you’ll need a more robust solution, often involving some level of human involvement in the process. According to the Zendesk 2024 CX Trends report, 71% of organizations now use digital channels for first contact and voice for resolving complex customer issues or escalation. While digital channels have long since become the norm for simple tasks, more complex processes remain manual by default. 

So how do you tell the difference between a simple and complex process? Generally, they will fall into one of these four categories:

1. Multiple Teams, Systems, or Touchpoints

Anything that involves a complex workflow is unlikely to be a good fit. This complexity can be defined as anything that involves multiple systems, customer touchpoints, teams, or stakeholders. Generally, a good rule of thumb is that a complex workflow is anything a customer service representative can’t resolve themselves.

AI tools aren’t designed to coordinate complex processes like this - and there’s a real risk of making customer service worse if they’re required to. In these cases, an automated workflow will almost certainly offer a smoother, more reliable experience.

Examples include: Returns, repair requests, purchase orders, customer onboarding/offboarding

2. Highly Emotional or High-Impact Cases 

Sometimes, there’s just nothing like the personal touch. If your customer is in a crisis, no automated bot or digital agent is going to put their mind at rest quite like a real person can. Generally, these situations are highly unique and the use of automation here can create further confusion, panic, and stress. 

In this case, you should ensure customers can speak to a human agent as quickly as possible. Then, you can use automated resolution workflows to ensure a swift and efficient response, whatever the crisis. 

Examples include: Vehicle breakdowns, lockouts, service disruptions

Make or Break

This is anything where a poor experience is going to have a significant impact on your most valuable customers. This could include long-term partners, well-known brands, or high-value clients. Often, these customers require a bespoke process to ensure they get the quality control and attention they need. 

There are several forms this could take. You might, for instance, want to ensure tickets from these customers are prioritized for resolution first. Or, it might be preferable to triage these customers straight through to the specific account manager or CS rep they already have a relationship with. Often, a bespoke experience is the best approach here.

In these cases, the stakes are simply too high to rely on AI. Generally, you’ll need a separate workflow for these customers, so there’s no chance of poor service or missed tickets. Then, you can proactively identify the right solution and ensure they’re immediately directed to the right place for a swift and effective resolution. 

Examples include: Any customer service requests from high-value customers

4. Compliance and Quality Assurance Tasks 

Certain workflows are simply too sensitive to risk handing over to AI tools. This is particularly the case where compliance laws lay strict standards on things like how long a task should take, what information a customer should be given, when and how data should be saved and deleted, etc. In these cases, AI simply isn’t yet advanced or reliable enough to ensure these processes are complete 100% of the time. 

The same is true when it comes to particular quality assurance tasks, including anything that’s governed by service level agreements (SLAs). In these situations, there will be strict contractual standards for the turnaround of specific tasks and the quality customers can expect on delivery. In this case, it’s also important to use technology that can provide you and your customer with these assurances. 

Examples include: GDPR requests, anti-financial crime (AFC) / anti-money laundering (AML) processes, tax reporting, support tickets, maintenance requests, etc. 

📖 READ: 10 Customer Service Workflows Every Company Should Automate in 2024

🤝 Offering a Better Deal for Customers

AI in customer service is new, exciting, and powerful - and there are plenty of good reasons to take advantage of it. If done correctly, it can promise a better experience for the customer and lower costs for the organization. 

But it can’t handle everything. The more complex a workflow is, the less capable AI is going to be at managing and overseeing the whole process. In that case, a more robust workflow automation tool or in-person agent will be much more effective. 

If you’re still wondering if a workflow is too complex for AI, here are four questions you should ask: 

  • How many teams, stakeholders, customer touchpoints, or internal systems need to be coordinated to take the workflow from ticket to resolution?
  • What service expertise or technical knowledge is required to resolve the issue?
  • What, if any, third-party organizations need to be involved in the process?
  • When can the human touch be the best way to alleviate a difficult situation?

Ultimately, whatever products or services you might be selling, your goal should be simple: To make it easier for customers to get done what they need to do. Whether you’re relying on AI, workflow automation, human agents, or some combination of all three, it’s this attitude that will make the real difference to your customers and the relationship they form with your brand. 

Want to find out more about what customer service processes to automate? Check out our recent guide to find out more. 

📖 READ: 10 Customer Service Workflows Every Company Should Automate in 2024

About the author
Luke Walker is the Product Marketing Manager at Next Matter. He is a longtime process hacker, and writes about marketing, business digitization, leadership, and work-life balance. When he's not at work, you can find him listening to records or climbing rocks.

Start automating in Next Matter today