ChatGPT for Customer Service

What is ChatGPT?

ChatGPT (Generative Pre-Trained Transformer) is a cutting-edge large language conversational AI model developed by OpenAI. 

Built on top of a Large Language Model (LLM), ChatGPT interacts in a dialogue format; understanding a diverse range of prompts and generating human-like responses by predicting future words. 

Hold up! What are Large Language Models? 

Large Language Models are AI tools that are trained on massive data sets to read, summarize, and translate text, then generate sentences by predicting future words. The more data a Large Language Model is trained on the more it is able to learn, remember, and form connections between different words, concepts, and things. 

Not only is ChatGPT trained on an extremely large data set, but it has over 175 billion parameters that allow it to learn more and make more accurate predictions.  

What are ChatGPT’s capabilities?

A fine-tuned variant of the GPT-3.5 model, which was trained on a wide array of internet text, ChatGPT has the ability to perform a range of Natural Language Processing (NLP) tasks. Some of the most common ChatGPT use cases include language translation, text summarization, question answering, and conversation generating. But ChatGPT’s capabilities don’t stop there!

Text classification, text completion, and text generation are also among the most common uses for ChatGPT. It can be used for other language-based tasks as well, depending on the needs and applications of the user. 

While its capabilities are incredible, it is imperative to be aware of ChatGPT’s limitations. Most notably that its output is not always accurate. So although the responses may sound plausible, the answer ChatGPT is providing could be entirely incorrect. 

Stick with us as we’ll be exploring some of ChatGPT’s limitations in the context of customer support later on!

What is ChatGPT used for?

A huge part of the excitement surrounding ChatGPT is that it can be used for many different things. So, what can you do with ChatGPT?

ChatGPT’s use cases include, but are definitely not limited to...

- Writing music, sonnets, social media captions, cover letters, and even code 

- Breaking down and explaining complicated topics

- Outlining blog posts  

- Solving math problems

And this list goes on! 

Can ChatGPT be used for customer service?

While ChatGPT was not originally built for customer service, its human-like responses and generative chat interface will definitely change the future of customer communications as we know it.

At this point in time, ChatGPT is not quite ready to be fully deployed into your customer service ecosystem -. but Thankful is actively testing ways to make this a reality. As the bridge between ChatGPT and customer support is being built, CX leaders can expand their horizons and get ahead of the competition by familiarizing themselves with ChatGPT and what it could do for their customer service operations.

In order for ChatGPT to be leveraged for customer support, it would need to be trained on a dataset of customer service interactions. After all, AI is only as good as the data it is trained on! Similar to other conversational AI automation tools (e.g. an AI Agent, chatbot for customer support, or virtual agent), ChatGPT will require this training to accurately understand common customer service inquiries and respond in a natural and human-like manner. 

A tool like ChatGPT would be best suited for those more basic and repetitive support tickets your team is typically inundated with, such as queries pertaining to shipping status or account information. Using ChatGPT to generate automated responses to your most frequently asked questions can help ensure customers quickly find the information they need without human assistance. And since AI solutions don’t need to clock in and clock out, brands can provide that quick service 24/7. 

In addition to automating FAQs, some of ChatGPT’s other capabilities would lend nicely to customer service applications. ⬇️

Language translation: Provide support to a wider range of customers by leveraging ChatGPT to assist customers in different languages.

Sentiment analysis: ChatGPT can be trained to understand and classify the sentiment of customer inquiries. By identifying customers' emotions, brands can be more equipped to respond in an appropriate manner. 

Personalized responses: By training and fine-tuning ChatGPT with customer-specific data, it has the potential to generate responses that are tailored to individual customers.

This is just the beginning. OpenAI is already developing the next version of ChatGPT, which means more possibilities and opportunities are underway. 

So what will the future of ChatGPT in customer service look like? Thankful has some ideas... 

In the near future, we foresee that the LLMs will start to integrate with external systems (e.g. policy engines and order management systems). This will allow the AI to go beyond answering FAQs and actually start making changes to customers’ accounts and orders. By incorporating integrations, AI will be able to resolve a higher number of customer support tickets. 

We also anticipate that ChatGPT will develop an understanding of when it does know an answer to something. This future advancement could be achieved by a low confidence interval or by AI managers performing reinforcement training. This will be critical in delivering accurate and effective service experiences; when the AI understands it cannot answer a question it can escalate the problem to a human agent who can take action and close out the ticket. 

As ChatGPT advances over time and becomes more sophisticated, it will impact customer service operations, support teams, and the overall customer experience. 

One of the primary reasons CX leaders are leveraging or seeking AI automation technology is to increase levels of efficiency and productivity - and Chat GPT will be no different. 

By quickly and effectively handling large volumes of repetitive customer inquiries, ChatGPT will help support teams to scale even when there is a spike in ticket volume. This alleviates pressure on human agents while allowing them to focus on more complex and rewarding tickets. With increased scalability and greater team productivity, brands could reduce hiring projections - ultimately cutting costs. 

The scalability and cost savings won’t have to come at the expense of customers. In other words, brands could actually improve service experiences for their customers.

Here are a few ways ChatGPT could elevate CX. ⬇️

The wait is over 

The speed at which AI automation tools like ChatGPT are able to respond to support queries saves customers from being stuck waiting in long queues. Rather, when a customer sends a message reaching out for support, they will receive a prompt response. This will help support teams reduce their time to first response (TTFR), which is a win for CX leaders that are aiming to hit their KPIs!

Make it personal

Trained to detect and analyze sentiment, ChatGPT will have the ability to better understand customers’ support queries and respond accordingly. When each customer’s issues are addressed in a unique way with a human-like manner, the interaction will feel more personal and they are likely to feel more taken care of. 

When ChatGPT can be trained with customer-specific data, it will have the capacity to generate responses that are even more tailored to each individual customer. 

Boost customer satisfaction

Quick, efficient, and personalized support - what’s not to love? An AI trained on your support tickets can match or exceed the CSAT of even your best agents.

Before this promising future can become a reality, there are some essential events that must occur in order for ChatGPT to function within a customer service ecosystem.

ChatGPT is regarded by customer support communities as a self-service AI chatbot that can solve common surface-level support issues. Though efficient, a true service experience is when a customer can receive answers to complex questions and have their problems fully resolved. These full-service experiences can occur with conversational AI solutions like Thankful that connect with brands’ existing business intelligence systems, policies and processes, as well as customers’ purchase histories to provide holistic support. 

For ChatGPT to function within a customer support environment, the following needs to happen:

A system that Interacts with the generative text layer 

ChatGPT’s response can vary greatly depending on the prompt that is given. Asking the same question in two ways can yield very different outputs. For this reason, it is crucial that the prompts one gives to ChatGPT are right. 

To address this challenge a new field called “Prompt Engineering” is emerging. It would generate prompts in the most ideal form for ChatGPT to understand and take the right action.

ChatGPT needs to integrate into business systems

In order for ChatGPT’s actions to be effective in customer support, it has to use external systems. Without integrating with a company’s business systems, it cannot actually carry out the action. 

ChatGPT needs to apply brand policies and processes

ChatGPT needs to understand company policies and (when necessary) have the ability to quickly adjust those policies through an external policy engine.  

ChatGPT’s performance needs to be measured

Measuring performance over time is essential for all highly complex systems. It is particularly important in this case as prompts and underlying long language models are updated and improved as time goes on. 

Right now, there is no industry standard test to measure the performance of generative models. Before this test can be deployed into production, a mechanism and process to measure improvements and regressions needs to be designed. Without this, we cannot be sure that the AI model works as intended across the tasks it is expected to perform

Brand safety has to be guaranteed 

Every answer that ChatGPT provides a customer must use the appropriate brand voice. Brand alignment is essential in CX, but the importance of enforcing brand safety goes beyond that. 

One of the risks of ChatGPT is if a user knows they are interacting with AI, there’s an unfortunate possibility of them trying to “trick” the model into responding in a way that is inappropriate or doing something the designer didn’t intend for it to do. 

This is why it is absolutely critical that the AI model cannot be “tricked” into replying inappropriately.  

Logical consistency must be enforced

As previously mentioned, ChatGPT is not always correct in its responses - even if it sounds plausible. Inaccurate answers are unhelpful and frustrating for customers and can cause a big mess that your human agents will have to clean up.

ChatGPT must provide customers with answers that are logically consistent within an interaction or conversation. But that’s not all; it must also be consistent across every single message it sends on behalf of a brand. 

Here at Thankful, we are both excited and invested in all the possibilities of ChatGPT in customer support. We understand that some brands will want to remain rigid when responding to customers, while others will want to take a more freeform approach. Thankful’s plan is to give each brand control over how generative they want their conversations to be.