Innovating Customer Service: Harnessing the Power of Generative AI

Generative AI is potentially highly suitable for customer service and support due to its ability to quickly and accurately generate human-like responses for a wide range of customer inquiries.

It allows organizations to scale their operations without hiring additional staff, resulting in cost savings and increased efficiency. It can also provide deeper insights into customer needs and behaviors, enabling data-driven decisions that enhance the overall customer experience.

With large language models like ChatGPT, this technology has the potential to revolutionize the customer service industry.

Let's take a look at a few examples of how companies are using generative AI to enhance customer service workflows and how it can help businesses improve their customer satisfaction rates.

Pre-built chatbot platforms that integrate with ChatGPT like Dialogflow, Microsoft bot framework, IBM Watson assistant, and Amazon Lex, are something many businesses are exploring to enhance their customer experience.

But let's take a closer look at how the technology can be used in such platforms.

In January, Intercom, a software company that specializes in customer messaging platforms, released beta features that integrated Open AI's ChatGPT technology into its customer service support platforms focused on a few different uses of the technology;

• Conversation summarization: Helping handovers between agents by generating a summary of conversations between customers and agents with the simple click of a "summarize" button.

• Composing messages to customers: They found this is where customer service agents spent almost half of their time, so they wanted to help by adding toolbar buttons for things like editing tone, rephrasing, and expanding.

• Generating help center articles: This is something they found companies struggle with finding the time to do.

But at the time, they were hesitant to implement the technology into their chatbots that are answering customer questions because of the risk of misinformation.

Then, just last week, they announced they would go full force with their GPT-powered chatbot using GPT4 technology that is able to minimize hallucinations, and have created their chatbot they call Fin.

They demonstrated Fin's power by giving an example prompt that they typed into a customer service chat for a fictional company they created for demo purposes called Staybnb (a vacation rental company).

"Is there a charge for canceling my reservation?"

The bot searches the company's existing support content and help center documents and provides an answer;

"If your reservation is still pending, you can cancel without being charged for the reservation or the service fees."

It follows up with "Was that helpful?" to which you can confirm, or even ask a follow-up question. Just like speaking to a real human in plain English;

"So how do I actually cancel?"

Again it searches through the information that already exists in the company's documents so the chatbot also provides you the option to go to the source document for more information. And when it doesn’t know something or gets asked a complex question, it will say it doesn’t know and hand it over to a support rep.

So the technology used here is a far step forward from the traditional customer service chatbots that use "IF" followed by the "THEN" reasoning to provide answers. It provides a more human-like interaction.

The experts at Intercom pointed out that if given the option between receiving an immediate response that is fairly good versus waiting for 15 minutes to receive a personally crafted answer, most people would likely choose the instant option, so businesses should use the technology to solve simple issues allowing their reps to focus on more complex ones.

They also note that in order to set your bots up for the most success, you need to have your As we know, the better the input, the better the output. And in this case, the input is your support documents. So companies should focus on writing those documents in an unambiguous way so GPT-powered bots have a clear source to draw answers from.

Intercom's use of GPT4 in their chatbots is a good indication that other companies will, or are already, using it in theirs.

Other companies like PolyAI are using similar technologies to create voice assistants that allow companies to answer customer service calls 24/7/365, no agents needed.

A PolyAI voice assistant can carry on a natural conversation to solve the customer’s problem, whereas previous voice assistants on calling platforms worked by recognizing keywords that the customer says and took a lot of effort to train.

The obvious benefit of this use is cutting costs of hiring customer service representatives to be available at all hours, but others include multilingual capabilities, as well as the ability to understand customer sentiment and offer personalized services.

The interaction on their demo goes something like this;

PolyAI: "Hi Brooke, thanks for calling Polymobile. How can I help?"

Customer: "Hi, I just need to renew my plan, okay?"

PolyAI: "Is this for the number you're calling from?"

Customer: "Yep."

PolyAI: "Looks like your silver level contract is nearly over. Did you want to renew it for another year?"

Customer: "Yes, please."

PolyAI: "Okay, it looks like you went over your data limit a couple of times. So if you'd like, I can upgrade you to the unlimited gold plan. It's only an extra ten dollars because you've been a loyal customer for two years."

Customer: "How much would it be per month then?"

PolyAI: "$40."

Customer: "That's right. Yes, it's $40 for the next year instead of $50. Yeah, alright. That sounds good. Thanks."

PolyAI: "Great! You'll be on the gold plan next year. Are your billing details still the same?"

Now, PolyAI has created its own proprietary language learning model called ConveRT which uses the retrieval model meaning it understands the meaning of the input and retrieves the answer from a pool of responses whereas ChatGPT can generate responses from scratch based on the data it has been trained on. ConveRT has been trained on millions of conversational samples, making it around 10% more accurate for this specific field.

Skeptics wonder if generative model tools like ChatGPT will ever be useful for big enterprise customer service because of the risk of inaccuracies and biases. In fact, Paweł Budzianowski, leader of the machine learning team at PolyAI commented, "It’s unlikely that enterprises will want to put their customer service entirely in the hands of generative models. In the same way call centers follow scripts, conversational assistants for customer service should always have an on-brand pre-approved bank of responses."

But what is particularly interesting is how PolyAI is using ChatGPT behind the scenes of their platform.

PolyAI uses ChatGPT to role-play as customers and simulate a real-world testing environment for its conversational assistants to create voice assistants with higher levels of accuracy. This type of AI-to-AI training, also known as adversarial training, is similar to the training method used to train the AI game, AlphaGo Zero.

They essentially engineer prompts for ChatGPT that trigger it to create similar utterances to those they would expect from customers.

For example, for a logistics company, they use a prompt like;

“Imagine you ordered a package, but it never arrived. Get on the phone with a contact center agent and find out what happened.”

PolyAI is also using ChatGPT to summarize conversations into short statements, like;

Prediction: Satisfied.

Conversation Overview: The agent informed the call with an update for their order query.

Dialogue explanation: The agent collected the delivery order and confirmed the first delivery had arrived and the second was due to arrive at a later date.

This also gives the company a real-time view of customer satisfaction without requiring callers to sit through lengthy surveys that most people opt out of in order to save time.

PolyAI isn't the only one harnessing the power of AI to understand how its customers are feeling.

Sentiment analysis is an increasingly important part of customer service in today’s world where people are quick to express their opinions online.

Bots can analyze text and then categorize it as positive, negative, or neutral in tone making it one of the most popular use cases in the customer service sector.

Viable is a platform used by big names like Uber, that uses GPT technology to identify emotions and sentiments from various sources such as surveys, help desk tickets, live chat logs, and reviews to generate insights from this feedback and provide a summary in seconds.

For example, if asked;

"What’s frustrating our customers about the checkout experience?"

Viable might say;

"Customers are frustrated with the checkout flow because it takes too long to load. They also want a way to edit their address in checkout and save multiple payment methods."

A great use of AI for customer service, but still dependent on your own company data.

Talkwalker's "Quick Search" is a sentiment analysis tool that is optimized for social media channels. It works by analyzing mentions, comments, engagements, and other data to better understand customer sentiment.

And Reputation is a platform that looks at multiple digital channels or locations to determine what people are commenting on and whether the comments are positive or negative.

Harnessing the power of AI presents exciting opportunities for innovating customer service. From chatbots that can handle complex inquiries with natural language processing, to sentiment analysis tools that provide valuable insights from customer feedback, companies can provide faster, more personalized, and more effective customer service, while freeing up human agents to focus on more complex issues and building stronger relationships with customers.

As we continue to push the boundaries of artificial intelligence, the possibilities for innovative customer service solutions are vast, and businesses that embrace this technology stand to gain a competitive edge in today's fast-paced digital landscape.