Business leaders wrapping their heads around the impressive capabilities of ChatGPT are still wondering how LLMs can impact customer service.
Should you use ChatGPT in your CX organization?
There’s a lot of excitement about ChatGPT and what it means for the future of customer support. And here at Ada, we share that excitement. Having worked with LLMs for the last three years and more specifically OpenAI, the creators of ChatGPT, for the last year, LLMs are the backbone of our automated customer support platform.
We receive a lot of questions from support and technology leaders on how to leverage this new technology, applied specifically to customer support. While ChatGPT is impressive, it wasn’t developed for customer support applications specifically, and has some important differences when compared to a purpose-built platform like Ada.
In this example, we have an Ada bot connected to ChatGPT. Imagine a major airline using ChatGPT for support. I’ll ask it a creative question that traditional chatbots might struggle with: "Suggest a vacation destination for me and my family somewhere United flies."
You can see that generative AI can respond with a creative, original suggestion based on the user’s query. However, there are some key limitations preventing immediate widespread adoption.
Let’s imagine I’m ready to book my flight to Costa Rica and I ask about the cost of checked bags.
Now, you can see here that the bot returns a lengthy and complex message, which says that on Economy and Premium Economy tickets, the first checked bag is free for a large number of flights including to Central America and the Caribbean.
If we check United’s website, we can see that if I look for flights from San Francisco to San Jose, Costa Rica, in Economy class, the first bag costs $35 and the second costs $45.
Let's try asking again in a different way: "How many bags are free on United on International flights?"
We get a different response now, but it’s still inaccurate, saying it’s one free bag plus a personal item on International flights. This presents a few problems for us.
There are four key requirements for automated customer support that are missed by ChatGPT:
- Accuracy: the information provided wasn't accurate.
- Explainability: I don’t know why ChatGPT is returning this false information to the customers or where it got this information from.
- Safety: I need to ensure that whatever my bot is saying to my customers, it represents my brand, and doesn’t risk harming customers or our brand's reputation.
Continuous improvement: We need tools to continuously improve our bot, but with ChatGPT, I don’t have the ability to make corrections — there’s no back-end admin interface available.
At Ada, we’re excited about the opportunities ChatGPT and other language language models unlock. ChatGPT, and LLMs in general, are a game changer for businesses that rely on natural language processing. When applied specifically to customer support, businesses can automate more resolutions with less effort.
As a general-use model, ChatGPT isn’t a replacement for tools like Ada, which is designed to solve the specific problems of enterprise customer support, but it represents a new wave of what’s possible with AI, and Ada will continue to leverage the underlying technology across our product in new and innovative ways.
Grant Oyston is a Product Marketing Manager at Ada. Before working in tech, he was an innkeeper, a crepe chef, a nanny, a bartender, and an ecommerce support agent, where he personally answered over 100,000 emails asking about where someone's order was.