Max Levchin
Analyst · William Blair. Please go ahead
That's a great question. I would spend a lot of time on it. So we are applying, and unfortunately the world has gotten really confused about what is and isn't AI under today's definition versus yesterday's definition. So, I'll try to be very precise. So we've used machine learning since the day of our founding. In fact, the idea was to build a credit score that was built on alternative data and modern machine learning techniques and obviously we've been pretty successful with it. So, we use that all the time, every part of the business, et cetera. So that's just the baseline. GenAI or LLMs, large language models, have been really, really helpful in a bunch of places. Internally, we use them in all kinds of ways, including, and we're actually quite actively investing into internal adoption of AI, where we have teams that are tasked with finding use cases for GenAI specifically inside their teams or increasing productivity, anything from we have literally hundreds of thousands of legal contracts with merchants, you need to find a clause that we need to modify for whatever reason, that's a great task for an LLM, read 435,000 contracts or whatever the current number is, find the clause that we need to change which is very subtly different contract to contract, summarize it, and construct a thing that we need to go out there and get re-signed. So that's the thing that humans would take thousands of hours for potentially can be done in minutes with AI. This isn't a made-up example, it's very real. The thing that I refer to in the letter, a big part of consumer delight is, at least for us anyway, is transparency and speed to resolution. And the ability to speak to a human and explain your case is really powerful. That's one thing that I don't think will ever go away, like, having empathy from someone who gets it and can actually speak to you with a human emotion in their voice. Machines aren't there yet, then it may actually be irreducible. But if you kind of know what you want and you don't want to wait, and you have it all figured out, most importantly, you have the evidence that there's a dispute and it's just going to go a very specific way and we can use machine learning to adjudicate the outcome, you can package the entire thing with GenAI and say, all right, let's have an interaction. And you understand you're talking to a robot but this robot can take in all your evidence, process it, run it through machine learning model in a backend and say yeah this basically means you get a refund. We'll tell you right now, the refund is coming, you're going to be okay, we'll take care of the rest. That is a huge booster of customer satisfaction and so that's actually it -- that's what I'm referring to in the letter. That's also an example of both GenAI as a user interface and machine learning on a back-end as a resolution device. Machine learning is much more stable if you will, the models are highly, highly predictable and repeatable. That's why they're favored in precise things like credit underwriting and GenAI models are somewhat less predictable. Everybody's heard about hallucinations, but they make for an amazing, very flexible user interface. And we use both very actively. I'll skip the part where I talk about how we have lots of engineers using lots of AI to write code and so on, but obviously we're doing that as well. So we don't spend a ton of time talking about it. We are very enthusiastic adopters of all this new stuff. But we also kind of think that it's a bit of a set of tools. You use what is available to you, and you try to find the very best ones, and you maybe don't make such a big story out of it.
Matthew O’Neil: Appreciate it. Thank you.