BNPL Loan Values Are Rising—AI Can Keep Risk in Check

A 14% jump doesn’t sound dramatic until you multiply it by millions of checkout moments.

According to a CFPB market study covering 2022–2023 activity, the average yearly dollar value of a buy now, pay later (BNPL) loan rose to $848 in 2023, up from $745 in 2022. That’s not just “people spending a bit more.” It’s a signal that BNPL is shifting from “impulse-friendly payments option” into a more material slice of consumer credit—and that has direct consequences for fraud, underwriting, collections, and regulatory reporting.

This is where the AI in Payments & Fintech Infrastructure conversation gets practical. When BNPL loan values rise, you don’t solve the resulting risk with more spreadsheets or a few extra manual reviews. You solve it with better infrastructure: real-time risk signals, AI-powered fraud detection, automated oversight, and audit-ready controls built into the transaction flow.

What the CFPB data actually says (and what it doesn’t)

The CFPB study gives a clear, numbers-first view of the BNPL market—at least for the slice it can see.

In the report:

• Average yearly BNPL loan value increased 14%: from $745 (2022) to $848 (2023).
• Total dollar value of BNPL originations grew 23% in 2023 vs. 2022 (26% inflation-adjusted).
• BNPL growth continued, but at a slower pace than the earlier period measured.
• Late fees declined: 4.1% of loans assessed a late fee in 2023 vs. 5.2% in 2022.
• Charge-offs declined: 1.83% in 2023 vs. 2.63% in 2022.

The dataset is also constrained:

• It’s based on six large BNPL providers (Affirm, Afterpay, Klarna, PayPal, Sezzle, Zip).
• That represents roughly 40% of the point-of-sale financing market in 2023.
• The CFPB explicitly notes it’s not necessarily representative of the entire BNPL market.

The late-fee debate is a measurement problem

One reason BNPL gets heated in policy circles is that two “truths” can coexist:

1. Transaction data may show relatively low late-fee incidence.
2. Consumers may still feel strain, especially when autopay triggers overdrafts or cash-flow gaps.

A consumer advocate quoted in the coverage points out that late fees and charge-offs are partial measures of distress. Meanwhile, a separate consumer survey (not transaction data) reported a much higher share of BNPL users paying late.

If you operate BNPL—or enable it as a platform—this gap matters because it tells you something uncomfortable: risk isn’t only about whether someone paid late. It’s about whether the product design causes hidden harm. That’s exactly the type of issue regulators (and plaintiffs’ attorneys) care about.

BNPL is growing up—and your risk stack has to grow up too

Rising loan values change the economics of BNPL.

When the average ticket grows, each failure mode becomes more expensive:

• Fraud losses scale with transaction size.
• Synthetic identity and account takeover become more attractive.
• Credit losses concentrate faster if underwriting isn’t dynamic.
• Disputes and returns create bigger reconciliation headaches.
• Regulatory expectations increase because the product looks more like mainstream credit.

And it’s December—peak season for BNPL volume. This is when weak controls show up as:

• sudden spikes in first-payment defaults,
• elevated merchant dispute rates,
• mismatched identity signals across devices,
• and customer support queues that balloon overnight.

Most companies get this wrong by treating BNPL risk as a single model problem. It isn’t.

BNPL risk is an infrastructure problem: multiple systems, multiple parties (merchant, BNPL lender, payment processor, bank), and multiple decision points—all happening in seconds.

Where AI actually helps: oversight, fraud detection, and infrastructure

AI is most valuable in BNPL when it’s used to connect signals across the lifecycle—not when it’s bolted onto one step.

1) AI-powered fraud detection at checkout (before authorization)

BNPL fraud doesn’t look exactly like card fraud. The attacker often aims to pass identity checks and “look normal” just long enough to walk away with goods.

AI improves outcomes by combining:

• Device intelligence (new device, emulator signals, velocity)
• Behavioral biometrics (typing cadence, navigation patterns)
• Identity graph signals (email/phone reuse patterns)
• Merchant context (SKUs, returnability, shipping risk)
• Payment context (funding source, token history)

A practical stance I’ve found works: separate “identity confidence” from “ability-to-pay”. Many BNPL stacks blend them into one score. That’s how fraud sneaks in: a fraudster can look solvent, and a real customer can look risky if they’re new-to-file.

AI lets you maintain two parallel decision tracks:

• Is this the right person?
• Is this the right amount and term for them right now?

2) Smarter underwriting and credit modeling (especially with rising loan values)

A 14% increase in average yearly BNPL value means your underwriting model faces more “edge cases”: customers stacking plans, shifting cash flow, and splitting purchases across providers.

AI can improve credit decisioning by:

• Updating risk in near real time as repayment behavior changes.
• Detecting plan stacking patterns (multiple active BNPL plans across merchants).
• Adapting limits based on cash-flow proxies (consistent payroll deposits, bill payment stability) where permissible.

The goal isn’t to approve more at any cost. The goal is stable approvals: fewer “approve then charge-off” cases and fewer “decline good customers” mistakes.

3) Automated oversight and audit-ready reporting

Regulatory attention on BNPL is uneven right now, but the direction of travel is clear: more scrutiny, more data requests, more expectations around consumer outcomes.

AI supports oversight when it’s used for:

• Policy monitoring: alert when actual approvals drift from written policy.
• Fair lending testing: detect proxy bias and disparate outcomes early.
• Explainability workflows: generate human-readable reasons tied to model features.
• Complaint triage: classify complaint themes (autopay, refunds, disputes) and link them to product changes.

A simple rule: if you can’t explain your BNPL decisions to a regulator in plain language, you don’t have an AI problem—you have a governance problem.

4) AI-driven transaction routing and payments infrastructure

BNPL isn’t only credit; it’s payments plumbing.

Behind the scenes you’re managing:

• authorization flows,
• settlement timing,
• refunds and partial returns,
• retries and autopay scheduling,
• and integrations across processors and merchants.

AI helps here by optimizing:

• Retry logic (when a payment fails, choose the best time and rail to retry).
• Dynamic routing (select rails/providers based on fraud risk, cost, and success probability).
• Exception handling (auto-match refunds to plans; flag mismatches before they become chargebacks).

This is the “unsexy” part of BNPL that quietly drives margin. Better routing and reconciliation can reduce avoidable loss without tightening credit at all.

A practical BNPL risk playbook for 2026 planning

If BNPL loan values are rising, the correct response is to treat 2026 as an infrastructure year. Here’s a concrete playbook you can use whether you’re a BNPL provider, a PSP, a bank partner, or a merchant enabling BNPL.

Step 1: Instrument the lifecycle (not just the moment of approval)

You need visibility from application to last installment:

• application and identity signals
• approval/decline decisions and reasons
• repayment events (on-time, retries, failures)
• refunds/returns mapped to plans
• disputes, chargebacks, and complaints

If your data is siloed, your AI will be blind.

Step 2: Build a “risk ledger” alongside the money ledger

A risk ledger is a structured event stream that records why decisions were made:

• model version and features used
• rules fired
• vendor signals
• human overrides

This becomes your backbone for audits, model monitoring, and faster incident response.

Step 3: Use AI where it reduces manual load and improves outcomes

High-ROI AI use cases in BNPL tend to be:

• identity verification escalation (only send to manual review when signals conflict)
• fraud ring detection across merchants
• dispute prediction (flag merchants/SKUs with elevated return or “friendly fraud” risk)
• collections prioritization (who needs outreach vs. who just needs a retry)

If your AI use case doesn’t reduce loss, reduce ops cost, or reduce customer harm, it’s probably a science project.

Step 4: Measure consumer harm signals, not only late fees

Late fees declining in transaction data can still coexist with customers struggling.

Add operational metrics that capture cash-flow stress:

• autopay retries per plan
• overdraft-adjacent signals (NSF patterns where available and permitted)
• “short pay” frequency
• refund timing vs. installment schedule
• customer support contacts per active plan

A blunt but useful stance: If customers need support to understand their repayment schedule, the product UX is part of your risk model.

People also ask: what does “AI-powered BNPL oversight” look like?

What’s the difference between BNPL fraud and card fraud?
BNPL fraud often targets identity and account creation, not just stolen card numbers. The attacker wants approval for a short-term loan and fast fulfillment.

Why would BNPL loan values rising increase compliance pressure?
Higher average loan values make BNPL look more like mainstream consumer credit. That attracts more scrutiny around underwriting, disclosures, repayment practices, and customer outcomes.

Does AI replace rules-based risk controls?
No. The strongest BNPL stacks use both: deterministic rules for known bad patterns and AI models for pattern discovery, anomaly detection, and decision consistency.

What to do next if you support BNPL infrastructure

BNPL isn’t “cool fintech credit” anymore. It’s turning into a scaled consumer finance product, and the CFPB’s data—especially the higher average loan values—is a reminder that risk, fraud, and oversight have to keep pace.

If you’re building payments or fintech infrastructure, this is the moment to ask a sharper question than “Are our losses okay?”

Ask: Could we explain our BNPL decisions, repayment mechanics, and dispute handling end-to-end—using data—if a regulator, partner bank, or enterprise merchant asked tomorrow?

That’s where AI earns its keep: not as a buzzword, but as a system for real-time fraud detection, consistent credit decisioning, and audit-ready oversight as BNPL keeps growing.