How ChatGPT will power ApexEdge’s next generation product and workforce (Episode 2/2)

The public launch of generative AI in the form of ChatGPT greased the creative wheels for the impact of these types of technology. We’re seeing experts and firms scurrying, just trying to wrap their minds and their hands around AI. 

We had something similar a few years ago when chatbots for financial services started getting interesting. The thing is – these were primarily black-box technologies that seemed very abstract from daily life for consumers and for the teams of developers at banks tasked with managing them.

ChatGPT changed that and with a slight change in interface, it’s unlocked a lot of use cases for financial services. We’re early for sure in this evolution but companies that have been studying the space are already beginning to make a move. ApexEdge provides consumers – through their banks and fintechs – subscription management tools. Want to easily unsubscribe from a subscription to a streaming service you’re not using much? ApexEdge makes that easy. It even has created technology to help you negotiate how much you’re paying for your subscription services.

The firm, which has been curiously studying the evolution of generative AI since ChatGPT-1, is already at work building the next generation of its product with ChatGPT at its center – including embedding it into how its flagship product works, as well as implementing AI tools internally to boost efficiency and increase productivity. 

For ApexEdge, it’s a lot about scaling efficiently. 

This is one of the first cases we’ve seen of a fintech firm embracing the next generation of AI tools so openly. 

This story is part of a collaboration between ApexEdge and Tearsheet Studios. To read more, please head over to Tearsheet where there’s a downloadable guide and two episodes of a podcast about how ApexEdge is using AI as a foundation for the next generation of its product, as well as an important resource inside the firm. 

Michel Mora is the CTO of ApexEdge, which runs Billshark, a popular consumer app that enables people to easily cancel subscription payments they’re making to streaming services, for example. ApexEdge utilizes the same technology to enable banks, FIs, and PFMs to offer similar services to their customers.

Subscribe: Apple Podcasts I SoundCloud I Spotify I Google Podcasts

The following excerpts were edited for clarity.

Learning to build tech in Cuba

I was born in Cuba andI went to school all the way to college without seeing a computer. Things are super tough there to be able to acquire anything, especially a computer 25 years ago. By my first year of college, I hadn’t touched a computer before. Not a single computer. My parents never passed high school. We didn’t know anything about computers in my house. I did think that it was related to math, so I chose it when I was planning to go to school.

When I was there, for the first few months, I realized that oh, man, this is amazing. I started my own business buying computer parts and putting them together and selling computers. It was kind of a niche way to use to start doing something. I’m making some money, saving some money to be able to buy my computer. It took me almost a year, but I was able to buy one for myself. But I did that for a couple of years, just to be able to make some money. So when people were talking about computers, I was putting together some super core graphic cards for people that like to play games, which had to work anywhere because you didn’t have the internet then. I had to figure it all out by myself 20 to 25 years ago.

AI and the need to stay in character

AI is definitely impactful in the technology world, as the internet was. At some point, AI is going to change the way that we use search for information on the internet for sure. We’ve been following this for a while and the main thing that we noticed, it all depends on what you need to use from ChatGPT. For our purposes, we need the AI to stay in character. So if you say to ChatGPT, I want you to be this character, and it doesn’t matter what question somebody’s going to ask, you must stay is that character — ChatGPT-3 wasn’t able to do it as well as ChatGPT-4 is.

In our use case, ChatGPT is going to interact with agents on the other side, representatives of the merchants our users are trying to cancel. There is a chance that they realize this is not a human and then the phone call is going to be over.

Skills in AI need to built new products

In terms of skills, it’s always good to have a background in AI. To understand how these things work, that always gives you an edge in terms of how you organize the architecture of the project. But I will say, though, that these days, 99% of developers, they just use the technology and don’t create it. So for example, we use AWS, everything that we use is hosted on Amazon today. So we don’t manage computers, we don’t manage servers, we don’t manage OS.

Similarly, with ChatGPT, you don’t manage the data behind it, and in some cases, you don’t even have to train it. We tried doing that — we tried to train it with the hundreds of conversations. It didn’t improve — not at all. When you think about how much data is behind ChatGPT, there is no way you’re going to be able to do anything to train it, especially ChatGPT-4, which I believe has almost all the data in the world up to 2022. What else are you going to add there? I mean, it’s just billions and billions of words and conversations and everything.

The new AI-powered service

We have multiple things that we’re working on right now. We’re trying to create something that is going to be able to interact with service providers by itself. I think using ChatGPT is about 30% to 40% of the entire project. And from our use case, it’s the easiest part to do, just because you’re just using an API. You just need to understand how it works.

But the other piece of what we’re building takes a lot more work than using ChatGPT. In terms of the other projects, we’re trying to apply to development so that we can start creating software with whatever charges we can generate and grow from there. It’s really amazing. You can see that at least there are three good products out there that you can start using that can generate code, at least again to start with.

Is AI game ready for different channels?

The core of the tool that we’re building is going to work pretty much the same for any channel. We’re not ready to use AI over the phone. We’re not ready for that — the other side is going to notice it, whether it’s going to be delays, when you’re trying to give that tool what to say on the phone, or whether it’s going to be the voice, that needs to be as good as you need it to be. We’re not there yet.

But whatever we’re building, we’re going to be able to use it for emails and chats, which are our other two main channels. It’s going to be just one tool that you can put together to interact with providers. In terms of the ChatGPT piece, the main and most important thing is the generation of the prompt. We’re building something that is going to generate what you’re going to feed to ChatGPT, so that it can go into a conversation. That’s the most important piece. And you have to be able to think about all the different angles and feel for every use case. Obviously, you have to do that automatically. Otherwise, you’re not going to be able to scale that up.

Autonomous chatting

The other piece, which I was saying will be about probably 60% of the project that we’re working on, is how you communicate with those providers via chat. How do you build a tool that’s going to sit on top of that, and get the data that we need, and put it back to ChatGPT? Then you need to get it from ChatGPT and send it to the provider. There is no chat tool that is built for interact with providers over API. So there are going to be a lot of things that we’re going to have to be doing.

The other complexity today is that every single chat out there that providers use is going to be different. You’re not going to be able to reuse the same one for two providers. For whatever reason, they don’t get along with each other.

The team building on top of ChatGPT

For the most part, it’s part of our internal team. We are lucky enough that one of the guys on our team that is helping us to build this has a masters degree and some tools that are related to what we need. And now he’s doing a doctorate in something related, so he’s aligned with what we’re doing. It’s really powerful to have somebody on the team that can do what he’s doing. Building on generative AI, you’re going to find things that are going stop you. It’s how you go around them and find a solution that defines how you move forward.

Need to learn AI internally

If you don’t learn how to use AI, you’re going to be behind in a couple of years. I can give you one example of something amazing that was built a few years back for some of the companies in the hotel industry. A few years ago if you wanted to book a hotel, you would have to go searching on multiple hotels’ websites. Firms like Expedia and Booking.com put them all together to create a system that is all in one place. Now you go to one place and you book. Now, if you’re a hotel and you’re not a part of that, most likely, you’re not going to have enough customers at some point because they’re going to take everything.

ChatGPT is doing something similar. If you’re not going to learn how to use ChatGPT, you’re going to be behind. And eventually you will be left out. Today, it is a nice to have — in a few years, it’s going be part of the core. It’s like when the internet came out, who doesn’t have a website today? It’s the same thing from my point of view.