Buy Now, Pay Later & AI: How Algorithms Decide If You’re Creditworthy in Under 3 Seconds

Okay so here’s something that happened to me a while back.

I was shopping online — Walmart’s site, actually — looking at an air fryer. Nothing expensive. $74. I get to checkout and there’s this little option I’d never really paid attention to before: “Split into 4 payments of $18.50. No interest.”

I thought, sure, why not. Clicked it. Put in my email, my debit card number, one more thing I can’t even remember. And it said approved. Just like that.

I looked at my phone for a second. The whole thing took maybe four seconds. I didn’t submit a credit application. I didn’t wait for an email. A machine looked at something about me — I didn’t know what — and decided I was good for it.

That bothered me a little. Not in a scary way. More in a “I should probably understand what just happened” way.

So I started looking into it. And what I found was honestly more interesting than I expected.

The Buy Now, Pay Later space — Klarna, Affirm, Afterpay, PayPal’s version — these companies are running credit AI that in some ways beats what traditional banks use for way bigger loans. They’re making millions of decisions a day, each one in under two seconds, and they’re getting pretty good at it.

How? What are they actually looking at? And is the whole thing fair?

That’s what this piece is about. I’ll try to keep it straight and skip the jargon where I can.

If you’ve already used one of these services, feel free to skip ahead. But for folks who haven’t:

Buy Now, Pay Later is a short-term loan that happens at checkout. The most common setup splits your purchase into four equal payments. You pay one chunk right now, then three more every two weeks for six weeks. No interest if you pay on time. No credit card needed.

That’s the basic model. Clean and simple on the surface.

What’s not so simple is how big this thing has gotten. The U.S. BNPL market hit $70 billion in transaction value in 2025. Globally it’s headed past $560 billion this year. Just during last year’s holiday weekend — Black Friday through Cyber Monday — Americans put over $10 billion on BNPL apps. Ten billion. In four days.

And it’s not just Gen Z buying sneakers anymore. Middle-income households are using it to manage budgets. People in their 50s and 60s are some of the fastest-growing users in the UK right now. This is a mainstream financial tool now, not a niche checkout gimmick.

Every single one of those approvals — millions every day — came from an AI that had about two seconds to make a call.

To get why BNPL companies went all-in on AI, you need to understand what they were dealing with before.

For decades, lenders used FICO scores. FICO looks at five things: how you’ve paid past debts, how much of your credit limit you’re using, how long you’ve had accounts, recent credit applications, and what types of credit you have.

For people with years of credit history — a card they’ve had since college, a car loan, maybe a mortgage — FICO works okay. It’s not perfect, but it gives lenders something to work with.

But here’s the problem. About 32 million Americans have no credit score at all. Zero. These people are called “credit invisibles.” They haven’t borrowed before. They’re young. They’re new to the country. Some of them just handle money with cash and have never needed credit. Under the old system, no score means automatic denial. Doesn’t matter if they’ve paid their rent every month for five years without fail. FICO can’t see that.

Speed was another killer. Traditional credit checks can take days. Sometimes longer. BNPL runs at checkout speed — if a customer has to wait more than a few seconds, they’re gone. Studies show cart abandonment goes above 70% when checkout gets slow or complicated. That’s not sustainable for any lender.

And accuracy? Old rule-based systems would flag good borrowers as risky all the time. False positive rates of 15 to 25%. Every one of those is a turned-away customer who would’ve paid back every dollar. A lot of lost money on both sides.

‘AI addresses all three problems at the same time. That’s the whole pitch — and for the most part, the data backs it up.

Two main technologies show up in most BNPL credit decisions.

The first is called Gradient Boosting — models like XGBoost and LightGBM. These are very good at making decisions on structured financial data fast. We’re talking under 200 milliseconds in some cases. You feed them your bank account behavior, income patterns, spending history — and they find the combinations of signals that best predict whether someone will repay. They’re not flashy. They’re just really efficient.

The second is Neural Networks. These handle messier, harder-to-structure data — device behavior, how you move through an app, patterns that don’t fit a spreadsheet. They pick up on things simpler models miss.

Put the two together and you get 92 to 96 percent accuracy on credit decisions. Old rule-based systems hit 78 to 85 percent. That sounds like a small gap. When you’re running millions of decisions a day, it’s not.

There’s also a transparency piece. Regulators are increasingly asking: if AI denies someone credit, what’s the reason? Tools called SHAP and LIME frameworks let companies point at any single decision and say exactly which data points pushed it toward yes or no. That auditability is becoming a requirement, not a nice-to-have.

Here’s where it gets genuinely interesting. And for some people, a little unsettling.

The AI isn’t just checking your credit score. Depending on the provider, it may not be looking at your credit score at all. What it’s actually pulling in falls into two buckets.

Cash-flow data — the stuff that actually makes sense

This is the core of what’s called “alternative credit data.” It includes things like:

•        How much money comes into your bank account each month on average

•        Your overdraft history — how often you go negative, and by how much

•        Whether you pay rent and utilities on time

•        How your account balance trends over time

•        How long your primary bank account has been open

Think about what that actually shows. Someone whose direct deposit hits every two weeks, who’s paid their electric bill on time for three years, who last overdrafted in 2021 — that’s a solid borrower. The old FICO model couldn’t see any of that. Cash-flow data can.

There’s a finding from the research I kept coming back to. Consumers who look like deep subprime borrowers on paper — the kind traditional banks decline automatically — still pay back their BNPL loans 96 percent of the time. Researchers call these people “invisible primes.” Low-risk borrowers who got mislabeled because the measuring tool was wrong. AI built around cash-flow data can actually find them. That’s a big deal for people who’ve been shut out of the credit system for no good reason.

Behavioral and digital signals — more complicated territory

Some systems also look at things that have nothing directly to do with your money. How long you’ve had your email address. Your device usage patterns. How you navigate within an app during a session.

The idea is that stable digital habits signal stable financial habits. Someone who’s had the same email for six years, uses the same phone consistently, has a long account history with the provider — that person is statistically lower risk than someone who just created a new account two minutes ago.

Is that fair? I genuinely go back and forth on this. It probably disadvantages people who change phones often, use multiple email addresses for privacy, or just don’t have a long digital trail for perfectly normal reasons. That tends to include older people, lower-income people, and privacy-conscious people — none of whom need more barriers to credit access. Regulators are starting to ask the same questions.

Here’s the sequence, roughly:

•        Data gets pulled. APIs grab your bank info if you’ve linked an account, device data, and any history you have with that provider. Happens simultaneously, not one after another.

•        Features get built. Raw data gets turned into inputs the model can use. “Spending volatility over the last 90 days.” “Net monthly cash flow.” “Days since last overdraft.” Dozens of these, calculated in real time.

•        The model scores you. Not just yes or no — a probability of default that shapes the offer terms and how much credit to extend.

•        Fraud gets screened. A separate identity verification layer checks that you are who you say you are.

•        Response lands on your screen. Usually before you’ve had time to second-guess the purchase.

All of that under two seconds. On a Black Friday morning when millions of people are hitting checkout at the same time. The engineering involved in doing that reliably at scale is genuinely something.

BNPL had a pretty comfortable run without much regulatory oversight. Most providers weren’t subject to the same rules as credit card companies. No required affordability checks in most states. No standard consumer protections. Companies kind of made up their own rules as they went.

That’s changing.

In 2024 the Consumer Financial Protection Bureau pushed to classify BNPL lenders as credit card providers under the Truth in Lending Act. That would have pulled them into a whole new compliance framework. The enforcement of that specific rule got complicated — legal pushback, political shifts — and wasn’t being actively enforced by 2025. But the broader signal is clear: Washington is paying attention and more structure is coming.

The UK is moving faster. New rules expected in 2026 will require real affordability checks before any BNPL approval. That’s a meaningful change from the current setup where a lot of approvals happen with basically no income verification.

One change that already happened in the US: Affirm started reporting repayment behavior to Equifax and other national credit bureaus. That’s worth paying attention to. It means if you use Affirm and pay on time, it shows up on your credit report. For someone with a thin file or no file at all, that could be a real pathway into the credit system. Other providers are slower to do the same thing, but the direction the industry is moving seems clear.

The AI is impressive and the financial inclusion piece is real. But glossing over the downsides would be doing you a disservice.

Loan stacking — more widespread than most realize

Here’s a stat that doesn’t get nearly enough coverage: 63 percent of BNPL borrowers are currently carrying multiple loans from different providers at the same time.

This is called loan stacking. And it creates a visibility gap that’s genuinely dangerous. Because most BNPL companies don’t share data with each other — and because most still don’t report to credit bureaus — no single lender knows what you owe across all the apps. You could have $1,200 in active BNPL obligations spread across four different platforms, and each lender thinks your total exposure is $300.

When cash gets tight, they all come due at the same time. Nobody saw it coming. Not them, not you.

Late fees are sneakier than the marketing suggests

A 2025 survey found 41 percent of BNPL users made at least one late payment in the past year. Late fees vary by provider but they add up fast, especially for someone already stretched thin.

What bothers me more is this: only about 52 percent of users say they clearly understood the fee structure before signing up. So roughly half of people using these services didn’t fully know what would happen if they missed a payment. That’s a problem that starts before the product is even used.

Read the terms. I know it’s boring. Two minutes, maybe three. It matters.

The “it’s only $22 a month” trap

BNPL can push average order values up by as much as 60 percent. Merchants love this number. For a shopper on a tight budget, it means the psychological cost of a purchase feels much lower than the real cost — because you’re only thinking about $22 today, not $90 total across four apps adding up at once.

The installment structure makes spending easier to justify in the moment. That’s not always a good thing.

Short answer: depends what kind of risk you’re talking about.

On debt accumulation, BNPL has a real structural advantage. Loans are short — six weeks max — and there’s no revolving balance quietly compounding at 27 percent APR. BNPL charge-off rates were around 1.83 percent in 2023. Credit cards were at 4.19 percent the same year. You can’t fall into the minimum-payment trap because there is no minimum payment. You either pay in six weeks or you don’t.

On consumer protections, traditional credit cards still have the edge. Dispute rights are clearer. Payments build your credit file automatically. There’s one institution to call when something goes wrong instead of juggling five different app support channels with five different policies.

My take: BNPL makes sense for a specific, one-time purchase you know you can cover in six weeks. It doesn’t make sense as a way to manage ongoing money problems spread across multiple platforms. That second use case is where things go sideways

If you use BNPL already — or you’re thinking about trying it — here’s what I’d actually suggest:

•        Use providers that report to credit bureaus. Affirm does. Paying on time builds your credit history, not just your checkout history.

•        Track what you owe across all apps combined, not just per app. It adds up faster than you think when you’re looking at it as one number.

•        Read the late fee structure before you commit. Seriously — five minutes of reading can save you real money.

•        If you link your bank account, the AI gets a cleaner picture of your actual cash flow. That can actually help your approval odds if your finances are in decent shape.

•        If you have a thin or empty credit file, a bureau-reporting BNPL service might genuinely be one of the more practical ways to start building credit right now.

That three-second decision you got at checkout wasn’t random and it wasn’t magic. A machine had just looked at your banking behavior, your spending patterns, maybe your device history — and made a probabilistic call, with pretty high confidence, about whether you’d pay it back.

For the 32 million Americans who’ve been invisible to traditional lenders their whole lives, this technology is genuinely opening doors. The invisible primes — people who pay their bills on time and handle money responsibly but have never had a reason to borrow — the AI can find them in a way FICO never could. That’s real.

But loan stacking is widespread and under-tracked. The regulatory framework is still getting built. And the behavioral data feeding these models is broader than most people who hit “approve” ever realize.

Understanding how it works doesn’t make you paranoid. It makes you a smarter user of something that’s already woven into everyday American financial life — and that’s only going to get more sophisticated from here.

Financial Disclaimer

The information published on Tech Capital Hub is intended for educational and informational purposes only. Nothing on this website — including articles, guides, analysis, or commentary on AI, fintech, blockchain, cryptocurrency, or stocks — should be interpreted as financial advice, investment advice, trading recommendations, or any other form of professional financial guidance.

All investments carry risk, including the potential loss of principal. Past performance of any financial instrument, strategy, or technology is not a reliable indicator of future results. Cryptocurrency and blockchain-based assets are particularly volatile and speculative in nature, and their value can fluctuate significantly in short periods of time.

Tech Capital Hub, Marcus Delray, and any associated contributors do not hold responsibility for any financial decisions you make based on content published on this site. Before making any investment or financial decision, we strongly encourage you to conduct your own independent research and consult with a licensed financial advisor, accountant, or legal professional who understands your personal financial situation.

Any links to third-party websites, tools, or platforms are provided for convenience and informational purposes only. Tech Capital Hub does not endorse or take responsibility for the content, accuracy, or practices of any third-party sites.