- Key insights: Affirm's founder and CEO discusses how he sees agentic AI reshaping shopping and personal finance and why that's good for consumers, and Affirm.
- What's at stake: Payment companies this year have been in an arms race, rushing to deploy artificial intelligence into all facets of the payments lifecycle, from payment orchestration to fraud prevention and client-facing treasury management services.
- Forward look: Agentic payments are the next nut the industry will need to crack.
Affirm's Max Levchin can see the future of commerce, and generative AI's role in it, in the way he buys bicycle parts. But that's just the beginning.
"I don't buy a whole lot of things, but I am a highly committed cyclist, and I buy way too many gadgets and widgets that make my aging body go faster on the road bike," Levchin, who was one of the founders of PayPal and now runs one of the largest BNPL companies in the world, told American Banker a day after his
"I've gone to local bike shops my entire life. It's a very specific experience. It's very curated. The person behind the counter is also the mechanic. They typically know a lot more about bikes than you do. You come in to buy a replacement chain and walk out with new pedals, some chain lube, and a service appointment scheduled and probably some snacks," he said.
For the local bike shop, most of the profit of that transaction comes from the margin made on the snacks and the upsell of the chain lube, and very little on the actual part that the consumer came in to buy or the labor to service the bike.
"Labor is not super expensive. It's not free, and yet they're going to try to think nothing about it at all, because just the value of having you there [in the store] and interacting with the inventory is actually quite valuable," Levchin said.
Fast forward to the online bike shop Levchin recently used, and the ecommerce model is not much different from the brick-and-mortar business model.
"I probably found this European bike shop by searching for a very specific bolt length on Google." Levchin said, noting his "fairly horrible leg-length mismatch."
The bike shop "probably paid a lot of money for what is going to be literally a 2.99 euro purchase, plus the shipping," Levchin said. "And, yet it's probably a profitable transaction for them for two reasons. One, I also bought some shims and a couple of interesting 3-D printed things for my bike. And I'm now aware of this brand and I'm going to come back to it."
That model was the status quo for decades, where search was the yardstick by which one could measure consumers' buying intent.
But that dichotomy was turned upside down when OpenAI released ChatGPT on November 30, 2022.
By January 2023 – less than two months after its launch – the large language model boasted an estimated 100 million users. Two and half years later, ChatGPT had 700 million users, or 10% of the global population, who were sending a total of 18 billion messages per week as of July, according to a September working paper from the National Bureau of Economic Research.
Seeking information about purchasable products was one of the top 24 most common topics of conversations on ChatGPT, according to NBER. And that's breaking traditional models of commerce in a potentially drastic way, according to Levchin.
The nut to crack, though, is in agentic payments. If an LLM can already streamline the shopping experience and return a list of products that fit a consumers' specifications, complete with recommendations, what happens to traditional flows of commerce once agents are the arbiter of purchasing decisions?
"I'm now going to go talk to my favorite chatbot about my leg-length discrepancy and what I should do about it. And it will tell me, 'What you need is a shim for your shoe.' It probably [costs] a dollar. And if I can just click on buy, don't care where from, within the chatbot, that kind of wrecks the business model in a couple different ways" by eliminating the merchant's opportunity to upsell the consumer to make that transaction profitable. It also limits any brand equity that companies could build with new customers.
"The notion of customer acquisition cost to lifetime value, which ecommerce has been grounded in, from day one till right now, is that changing? Is there a notion of lifetime value if the AI decides what will be the right source of whatever it is you're buying?"
Chatbot as financial advisor
Levchin doesn't claim to know exactly how Affirm will harness this shift in consumer behavior. But he sees potential for major upside for his company because it will ultimately increase transparency around consumer finance products, especially as chatbots increasingly take on the role of financial advisor.
"No AI model is going to pay snooze fees or reminder fees or late fees or fall for deferred interest," Levchin said. "Nothing is more annoying than being stunned with a $30 late fee on a $15 transaction, yet it happens to almost everyone. But it's just not going to happen when AI looks after your personal financial state. You will not be late without knowing what's going on. You will not auto renew or get hit with insufficient funds or fall for a 0% scheme or some kind of predatory loan.
"All of that is just this side of being in the rearview mirror because these machines have a PhD-level [education] in consumer finance, and most people do not," he said.
Levchin admits he's "talking his own book" a bit in the sense that Affirm does not charge late fees and fixes interest rates at the time of transaction.
"That accretes to us," Levchin said. "I don't know if it inherently makes us better known [but] today, if you search 'Best BNPL provider with no late fees' on ChatGPT, it will tell you, 'You should use Affirm.' That's profoundly a good thing for us."
The notion of an internet browser with embedded AI serving as an underlying observer and helper is a powerful one, Levchin said, and will likely be what ecommerce looks like in the near future.
"My best guess right now – and I reserve the right to be proven wrong – is you will be in your fairly normal browser experience, and the AI will be an extremely souped up version of a comparison shopping / browser toolbar," Levchin said.
"The norm will be real AI, which will constantly comment like, 'Hey that 3-D mount is going to make your bike that much more aerodynamic… By the way, I think you should get more wax for your chain.' That will be your constant talk track on your shopping life," he said.
Similarly, AI can help consumers make better choices in an angel-on-your shoulder dynamic. "Having an AI that controls your browser to at least some extent, saying, 'I know you're about to click on a thing, I'm pretty sure it's a bad idea for you to click on it,'" Levchin said. "'We're not going to do that right now' is easily extensible into, 'Hey, that thing that looks like a giant 0% with a tiny asterisk. The asterisk means it's not to be trusted. Don't touch that. And that helps Affirm because we never have an asterisk."
Are the bots ready?
Whether generative AI models are sophisticated enough today to be a PhD-level financial advisor is up for debate. Chatbots have been known to display sycophantic behavior, and have even gone so far as to falsely confirm a new
"AI agents want you to think you're right. It wants to praise you. It wants you to think that's a good idea," Daniela Hawkins, a partner at Capco, told American Banker. "When it comes to finances, that may be a guardrail that needs to be considered."
So how will AI be able to help a consumer make good financial decisions, such as holding off on a purchase while other debt is being paid down? Levchin thinks it will take time.
"I don't want AI telling my consumers, 'Oh yeah, what a great idea. Borrow some more, go into debt," Levchin said. "I would love for AI to be like, 'Hey, let's examine your personal balance sheet. It doesn't seem like a good idea for you to borrow this much money for this thing. Consider saving up and postponing this purchase for a little while."
Levchin believes that the bots will be measured by how helpful they are, and that the market will adjust accordingly.
"If [chat bots] end up driving people into a bad financial choice, they will be displaced by the ones that do drive positive outcomes because as much as we love new shiny toys, if my bike doesn't ride as fast as I was hoping for, I will switch bikes. I think the same is true for AI models," he said.
Access to consumers' financial data will also be paramount. Levchin puts consumer delinquencies into two buckets: The individuals that are lying to Affirm and have no intention to pay, and the individuals that are fooling themselves in their ability to repay.
"If we do it right – which I think we will – we're going to give these models full context into our personal finance, and that will really help people who are lying to themselves get a little bit more sober," Levchin said.
"As the chatbots become partner advisors, they're also going to need more context, and as they gain context, they'll be able to give us better advice, and ultimately, would be full participants in a conversation around what's the best financial position for you," he said.
And with access to your financial accounts, and with the autonomy that agentic payments promise, the bots will be able to actually do the things consumers don't have time for.
"You get very close to the water, but the machine cannot make you drink," Levchin said. "With AI, we will actually be capable of saying, 'Hey, drink for me. Go set up my 401K. Create a retirement path. Assign some trusts for my kid. Whatever, financial planning or financial cleaning you need to do in your life, AI is probably very close to being capable of doing so."
