Conversational AI for Customer Service: How to Get it Right
Prior to working on the content team here at Help Scout, I spent several years working in customer support. As you can imagine, with a career focused on customer service and creative content, my reaction to the release of ChatGPT — a large language model released by OpenAI in 2022 — has been mixed.
On one hand, I think using AI in customer service is pretty exciting. There are so many opportunities for artificial intelligence (AI) to elevate the work support teams are doing and to make a positive impact on the customer service field in general.
On the other hand, with technology capable of writing, speaking, troubleshooting, and creating original content, it’s hard not to feel a bit insecure.
In this post, we take a closer look at conversational AI, an area of AI technology now playing a large role in customer support experiences. We’ll cover what it is, how it works, how it can be used as part of your support strategy, and answer the all-important question:
Can conversational AI replace your support team?
What is conversational AI?
Conversational AI is the technology that enables humans to have realistic text or speech-based conversations with machines and applications such as chatbots, smart devices, wearables, and virtual assistants.
Historically, conversational AI hasn’t always provided the best user experiences. However, with recent advances in the field, namely the release of ChatGPT, more people have looked for ways to incorporate the technology into their lives, both personally and professionally.
How does conversational AI work?
While most people have probably (much to my chagrin) asked Alexa a question before, the technology behind Alexa’s answer — conversational AI — is likely still a mystery.
The first thing one needs to do to make sense of how conversational AI works is to understand the related technologies that come together to make human-computer conversations possible.
The AI technologies that make up conversational AI
Conversational AI is built on two major branches of artificial intelligence: Natural Language Processing (NLP) and Machine Learning (ML).
Natural Language Processing (NLP)
Natural Language Processing, or NLP, is the AI subfield that facilitates natural conversation between humans and computers. It’s a complicated task, as computers need to not only understand what words and phrases people are saying or writing, but they also need to understand the context and sentiment behind that language.
When you hear about the discipline of NLP, you’re likely to run into three other associated terms:
Natural Language Understanding (NLU): This is the aspect of NLP that analyzes human speech or text for context and sentiment, ensuring that the computer (phone, chatbot, etc.) understands the intended meaning.
Natural Language Generation (NLG): This is the part of NLP that allows a machine to respond to human speech in a conversational way, basing its responses on structured data (databases, user guides, FAQs etc.).
Generative AI: Generative AI is another technology that is capable of facilitating natural conversation; however, it is also capable of producing other types of content like images and music, and its outputs are not necessarily tied to structured data. Large Language Models (LLMs) like ChatGPT utilize generative AI with a focus on NLG to produce responses to user prompts.
Along with the additional technology that helps parse and segment speech or written word into something a computer can understand, NLU, NLG, and generative AI combined are the primary building blocks for conversation between people and machines.
Machine Learning (ML)
While NLP is impressive on its own, the real benefits start to emerge when it is combined with Machine Learning.
Machine learning is another branch of AI that uses algorithms and data to continuously “learn” and improve its output over time with minimal human involvement.
When it comes to conversational AI, machine learning makes it possible for a computer application to take the results of all previous conversations it has had with a human, as well as any additional data provided, and use it to deliver better responses in the future without additional programming.
How do AI technologies work together to create a realistic conversation?
Now that you know what NLP and ML are, let’s take a look at how they might work together to create a useful conversation between a human and a computer.
We’ll start with a question — “How can I knock my next job interview out of the park?”
First, you’ll speak or type your question into an AI-powered app’s UI. If you speak, the computer will use speech-to-text or Automatic Speech Recognition (ASR) technology to digitize the language.
The computer will parse and segment your request into a form it can understand.
Next, it will use NLU to understand the question. This means untangling things like what it means to “knock something out of the park” — is doing so a good or bad thing? Does it relate to baseball? What does it have to do with a job interview?
Once the computer understands that you want tips on how to do well in a job interview, it will use machine learning to consider how questions like this one have been answered before. Were the previous responses considered useful by the human who received them? Did it result in follow-up questions?
Finally, the computer takes this understanding and uses NLG and/or generative AI to provide you with interview tips using relatable human language.
Can you trust conversational AI?
How reliable the computer’s response might be depends on the specific LLM — deep learning algorithms trained on large quantities of data — the computer is using to process and generate language.
Recently released LLMs like Open AI’s GPT-3.5 and GPT-4 have been at the heart of the past year’s AI surge. However, not all LLMs and AI products are created equal. And even with high-performing models, hiccups can occur. For instance, sometimes AI can suffer what is known as a hallucination — an incorrect response that is presented confidently as fact.
This means that while there is substantial promise in the field of AI and accuracy is constantly improving, we’re not yet at the point where you would want AI flying solo with anything that could have a serious impact on your business, such as customer service.
However, as a co-pilot, there is potential.
10 benefits of using AI in customer service
Though conversational AI isn’t ready to handle every customer conversation, it’s still a powerful tool that your staff can tap into to improve the support your team provides and the experience your customers receive.
Here are 10 potential benefits of adding conversational AI tools to your customer support strategy:
Increases support coverage: AI allows your business to extend support coverage beyond the standard workday. Conversational AI chatbots and IVR systems can manage basic requests, enabling customers to get help when it is convenient for them. This is especially handy for teams covering multiple time zones.
Faster response times: In addition to receiving responses outside of business hours, conversational AI tools provide responses almost instantaneously, helping customers get answers more quickly.
Scalability: AI-powered tools can process multiple customer requests at once, making it easy to grow your business without overwhelming your team.
Lowers support costs: Along the same lines, conversational AI tools can take the pressure off of your support team without the need for hiring additional staff.
Improved consistency: Along with knowledge base content and saved replies, the use of conversational AI tools can help ensure that your team is providing consistent information across all channels.
Better onboarding: Agents can use AI to help draft support responses or query an internal AI knowledge base if they have questions during training, helping them gain confidence in the queue more quickly.
Boosts team productivity: When FAQs are being answered through self-service channels like virtual agents and knowledge bases, your team can focus on cases and tasks that require a higher level of expertise.
Increases customer engagement: Conversational AI is capable of proactively reaching out to customers with personalized messages. From product recommendations to feedback requests, AI can help keep your customers engaged and excited about your brand.
Improves customer experience (CX): Customers want fast, accurate, and personalized support. While that is a tall order, AI-enhanced channels can independently provide this level of support for simpler requests, and for those that are more complex, AI can provide your team with additional information to make their job easier.
Provides room for growth: While many support team members worry about being replaced by AI, the technology can actually provide space for more interesting assignments, building skills, and making progress toward individual career goals.
What does conversational AI look like in customer service?
Conversational AI shows up in customer service in a number of places, though predominantly it is encountered in self-service experiences like:
Chatbots and virtual agents: AI-powered chatbots — sometimes referred to as virtual agents — provide a much improved CX over their rule-based predecessors. Where users may have previously reached a dead end when the interaction strayed from the assumed support flow, AI-powered chatbots can better handle unexpected situations and provide more accurate, human-like responses.
Interactive Voice Response (IVR) systems: While traditional IVR systems are rule-based and can sometimes struggle to understand spoken word when it is complicated by accents, slang, or a bad phone connection, conversational IVR systems rely on ASR, NLP, NLU, and ML to deliver better support.
Knowledge bases: Even with good architecture and a solid search engine, it can still be tough for customers to find the information they’re looking for in a standard knowledge base. When a knowledge base is enhanced with conversational AI, people can locate the right resource fast by simply asking the knowledge base a question in plain language and having it “understand” and respond in kind.
What types of customer service tasks can conversational AI handle?
Using the channels above, conversational AI can perform a number of customer support tasks that can improve CX and lighten the load for your support team.
A few of the main tasks you may choose to streamline using conversational AI include:
Responding to frequently asked questions (FAQs): Answering customer requests, even FAQs, can be time consuming. AI can give your team back their time by taking care of common questions.
Providing order updates: AI is great for providing order information such as shipping statuses and tracking numbers.
Returns and exchanges: In a similar vein, AI-enabled chatbots and IVR systems can process simple returns and exchanges.
Product or content recommendations: Conversational AI can take prior communications, browsing behavior, and known personal preferences into consideration to provide customers with suggestions on products to buy or content to consume.
Collecting data and customer information: Conversational AI chatbots and IVR systems excel at asking customers questions and gathering data such as contact information and user feedback.
Triaging support requests: Conversational AI tools are capable of assessing an incoming support query and routing it to the appropriate agent or team.
Troubleshooting simple technical issues: AI can run through simple troubleshooting steps like performing resets and checking device settings.
Providing multilingual support: AI technology can instantly translate customer requests and provide support across multiple languages.
Appointment scheduling and reminders: Conversational AI support tools can create service appointments or schedule support callbacks, as well as remind customers of upcoming service visits.
Suggesting responses: AI can analyze the content in an incoming support request and provide a suggested response. Agents can then review and modify the response as needed before hitting send.
While conversational AI shines in the scenarios above, there is still a lot that it can’t (and you shouldn’t want it to) do.
Here are a few tasks best left to members of your customer support team:
Crisis management: AI should never handle situations where the stakes are high for either the customer or your business.
Complex troubleshooting: When troubleshooting moves beyond the basics, it’s best to escalate to a higher tier of support. Even simple troubleshooting can be frustrating for a customer, and humans are uniquely qualified for creating a connection and balancing information with patience and empathy.
Public support requests: While conversational AI has come a long way, it’s still best to avoid letting it provide responses in public forums like your brand’s Facebook page or Instagram comments without human supervision.
Requests that involve legal, security, or privacy issues: Whenever you’re dealing with a legal, security, or privacy issue, your company is exposed to potential liability. While conversational AI may have the skills necessary to complete the request, it’s best that these types of requests are managed by a human, just in case.
Issues that involve emotions, ethics, or judgment calls: While conversational AI tools can detect sentiment and even mimic tone, they lack the ability to empathize. When emotions are running high or a situation requires your team to operate outside of protocol, it’s best for the case to be handled by someone on your team.
Using conversational AI responsibly in customer service
There is a lot of (justified) excitement around all of the different ways that AI can improve the way your customer service organization operates. At Help Scout, we have changed our view on using AI in customer service settings and have even launched some AI features that we think are pretty great.
That said, when you’re in the business of helping people, the stakes for getting AI right are high. Even just one poor customer support experience with your brand can cost you a customer, so it’s important to balance excitement over new technology with the responsibility of maintaining a good experience for people — both your customers and employees.
Here are some things to keep in mind when looking to responsibly offer conversational AI experiences.
Be transparent and set expectations
For a customer, it can be hard to tell the difference between a chat reply crafted by a customer support agent and one confidently generated by AI. Your experience should make it obvious who (or what) your customer is interacting with at all times.
In addition to ensuring that customers know who they’re talking to, be honest with them about the customer experience your AI solution offers. If it can only handle specific types of queries like returns and exchanges, then be upfront with that information so that customers aren’t disappointed if your experience doesn’t live up to their expectations.
Always provide a pathway to a human
There are lots of reasons why a customer might prefer to speak to a human over a chatbot or virtual agent. Whatever their reason, customers who are being engaged by AI should always be provided with a way to speak to a real person if they would prefer it.
As part of that transition, let customers know what the expected response time for the new channel will be. For instance, if the customer wants to speak to someone via email or phone, let them know when they should expect to receive a message from your team. Or if they’re requesting a live chat, let them know what the wait time to speak to a person is.
This will allow customers to choose the option that will not only make them feel most comfortable but will also align with the urgency of their request.
Train your tech with relevant data
When setting up your AI experience, be sure that the data used to train your solution — this can be anything from knowledge base articles to CRM information to past support interactions — is well-written, comprehensive, and accurate. Also, ensure that it doesn’t include anything that you wouldn’t want it to consider (e.g., proprietary data) when formulating a customer-facing answer.
If you’re building the experience in-house, you’ll have more control than if you’re using third-party software. If you are using a third-party option, ask their team what data is used when formulating responses and whether the technology utilizes machine learning to improve responses over time.
Test and monitor your CX
Once the AI is trained up, be sure to thoroughly test the results internally before rolling it out to your customers. This will help ensure that the customer experience is good, and if there are shortcomings, you’ll be able to set customer expectations.
But your involvement isn’t over once your AI solution is live. Post rollout, your AI experiences will require regular monitoring to spot issues and identify opportunities for improvement.
Collect customer feedback
Part of the monitoring process is creating a feedback loop with your customers to gauge satisfaction with your AI-powered features.
Closely monitor CSAT and NPS scores and consider other feedback options such as targeted in-app surveys or video calls with key customers to learn what is working and what needs attention.
Keep privacy in mind
While there are plenty of great use cases for conversational AI in fields like law or medicine, companies with strict security requirements need to explore how the AI tools they use work, who has access to client data, and what kind of security precautions are used to keep data safe.
Before adding conversational AI solutions to your customer experience, double-check that your chosen AI tools are following all security and privacy practices that your business requires.
Can conversational AI replace your support team?
One of the biggest draws for companies to implement AI solutions is cost savings; however, I’m just going to cut to the chase with this one:
Conversational AI cannot replace your support team.
Customer support is more than just providing the right answer as fast as possible. It’s about creating positive brand experiences and customer connections, improving your product or service, creating a feedback loop between executives, engineers, and end users, humanizing data, and many other things that a computer just isn’t capable of.
As you approach integrating AI into your customer service processes, remember that conversational AI is not an alternative to having a customer support team, but rather a tool that empowers them to do their best work.
How to introduce conversational AI into your customer support experience
Ready to introduce conversational AI into your customer support experience? Here’s how to get started:
Set goals. Do you want to reduce first response time? Lower contact volume? Give your team time to focus on more complex tasks? Knowing why conversational AI is a good fit for your support strategy is an important first step.
Get your team on board. The introduction of AI may make your team uneasy, so it is important to get their buy-in early. Meet with the team, listen to and work through any concerns, and get their opinions on where in the support experience AI would be most helpful.
Determine resources. Implementing AI may involve technical resources or additional budget. Figure out what you have to work with so that you can seek out the right solution for your team.
Consider tools. Many help desks offer AI features, or they can be added on through an integration. Alternatively, you may prefer to build your own solution. If you’re going with a third-party option, you may want to try them out to see which works best for you.
Seek feedback. Check in with your team and customers regularly to figure out what is and isn’t working, and make changes accordingly.
Supercharge your support with AI
At Help Scout, we’ve made using AI feel effortless. Learn more about how our AI features can save you time and energy on every conversation.
Learn More
Customer service is changing
AI is introducing new ways for customers to receive faster, more personalized support, and customer service teams are able to leverage that technology to ease the burden of busy work and grow their own roles beyond the queue.
However, despite all of that good stuff, it’s okay if you or your team are feeling nervous.
This space is new and evolving quickly, and feeling unease around change is completely natural. The important thing to remember is that as good as AI is, it is no match for the creativity, experience, and heart that support teams bring to the table each day. Support roles are changing, but they aren’t going away.
Alexa Ahrens
The Supportive Weekly: A newsletter for people who want to deliver exceptional customer service.