How does Klarna use AI to make credit decisions?
Klarna underwrites every purchase individually, in real time, with machine-learning credit models. Its FY2025 20-F says the fully automated process draws on Klarna's own records and purchase behavior from about 3.4 million transactions a day, plus merchant data, credit bureau reports and open banking data, and decides in a matter of seconds. Klarna says its US Gini score, a measure of predictive power, rose from 0.36 in 2019 to 0.78 in the fourth quarter of 2025, against 0.43 for VantageScore 4.0 by its own credit scoring model. US provision for credit losses fell from approximately 3.6% of GMV in 2021 to approximately 0.63% in 2025. Exposure is small and short: an average balance of $124 per active consumer and a weighted average life of about 39 days in 2025. In Q2 2026, provisions were $192 million, or 0.52% of GMV, and Klarna says it has underwritten over $0.7 trillion since inception with provisions of around 0.6%. Its filings flag fair-lending risk under ECOA and possible EU AI Act duties for its ML-based risk scoring model, and the UK brought interest-free BNPL under FCA regulation from July 15, 2026.[1][2][3][4]
Data as of Oct 19, 2026 · Klarna profile updated Oct 19, 2026 · 11 sources cited on this page · 52 fintechs profiled on this site · Markdown
What goes into a Klarna credit decision?
As Klarna describes it in its FY2025 20-F and 2026 earnings releases. Klarna has not published its model variables or architecture.
| Step | What Klarna discloses | Date |
|---|---|---|
| Inputs | Klarna history and purchase behavior from about 3.4 million transactions a day by 118 million active consumers, plus merchant data, credit bureau reports and open banking data | February 26, 2026[1] |
| Sequence | Identification and authentication; credit and fraud policies to prevent over-indebtedness; ML-based risk scoring; comparison against Klarna's internal risk appetite; final decision | February 26, 2026[1] |
| Limits | A small spending capacity that gradually increases as consumers spend responsibly with Klarna | February 26, 2026[1] |
| Repeat underwriting | "We re-underwrite essentially every transaction"; most point-of-sale installment borrowers arrive with a Pay Later repayment history, which Klarna calls the foundation of how it underwrites them | May 14, 2026[5] |
| Agent checkouts | Pay-in-four and longer-term financing in Google's Gemini app and Search are each underwritten with an affordability check | May 12, 2026[6] |
| UK credit files | Soft credit search for Pay in 30 and Pay in 3; Klarna said it would share on-time, late and unpaid BNPL purchases with Experian and TransUnion | May 4, 2022[7] |
| US partner bank | WebBank originates Klarna's US Fair Financing loans, which Klarna then purchases and services | September 10, 2025[3] |
How well does Klarna say its credit models perform?
Klarna's own figures. The Gini comparison is "according to our credit scoring model"; no independent validation is published.
| Metric | Value | Period |
|---|---|---|
| US Gini score (Klarna vs VantageScore 4.0) | 0.36 in 2019; 0.72 vs 0.38 in Q1 2025; 0.78 vs 0.43 in Q4 2025 | 2019–2025[3][1] |
| US provision for credit losses as % of GMV | Approximately 3.6% in 2021 to approximately 0.63% in 2025, while US GMV grew approximately 213% | 2021–2025[1] |
| Group provision for credit losses | 0.56% of GMV in Q2 2025 (0.41% a year earlier, as US and Fair Financing shares rose); 0.65% in Q4 2025; $192 million, 0.52% of GMV, in Q2 2026 | 2025–2026[3][8][2] |
| Delinquencies | US Financing 30+ days past due improved 36 bps from the Q2 2025 peak (Q1 2026); Fair Financing 30+ fell about 20 bps and Pay Later about 30 bps quarter over quarter (Q2 2026) | 2026[5][2] |
| US Fair Financing charge-offs | The Q2 2025 cohort carries modestly higher charge-offs after the Klarna Card and 3-month tenor ramp and a model recalibration; later 2025 cohorts within an expected 3–4% range | Q2 2026[2] |
| Repayment | 99% of consumer loans paid on time | 12 months to June 30, 2025[3] |
| 2020 tightening | US Pay in 4 approval rates cut from 67% to 57% between March and April 2020; credit losses fell 60% from the March to the April 2020 cohort | 2020[3] |
What rules apply to Klarna's credit models?
| Rule or authority | What it means for Klarna | Date |
|---|---|---|
| ECOA (US) | Klarna's prospectus says a variable in its model could be deemed a proxy for a protected characteristic and that ML could cause disparate impact; it reviews selected variables through Legal and Compliance and runs post-implementation testing | September 10, 2025[3] |
| EU AI Act and GDPR | Klarna says its ML-based risk scoring model may oblige it to comply with the AI Act; EU and UK GDPR regulate automated decisions with legal or similarly significant effects | February 26, 2026[1] |
| FCA (UK) | Lenders offering Deferred Payment Credit (interest-free BNPL repayable within 12 months) come under FCA regulation from July 15, 2026, with rules on affordable borrowing; DPC lending grew from £0.06bn in 2017 to over £13bn in 2024 | February 11, 2026[4] |
| CFPB (US) | The Bureau said in March 2025 it planned to revoke its interpretive rule treating Pay in 4 loans as credit cards, after a suit by the Financial Technology Association, whose members include Klarna | March 28, 2025[9] |
| FDIC and Utah DFI | Klarna applied to form Klarna Bank USA, an FDIC-insured Utah industrial bank | July 6, 2026[10] |
What does Klarna not say about its credit models?
Klarna does not publish its model variables, its model architecture, current approval rates or how it explains declines to applicants. The Gini scores and the comparison with VantageScore 4.0 come from Klarna's own credit scoring model.[1]
Klarna's two 2026 releases also describe the lifetime record differently: the Q1 2026 letter cites more than half a trillion dollars originated across 20 years with a through-cycle provision rate of approximately 0.6%, and the Q2 2026 release says over $0.7 trillion underwritten since inception with provisions of around 0.6%.[5][2]
What does the Klarna profile record on this topic?
The rows below are the entries of the Klarna AI strategy profile that bear on this page, as of Oct 19, 2026; the profile holds the full record of 11 dated moves, 7 use cases and 15 sources.
| Use case | What the record shows | Status | Source |
|---|---|---|---|
| Real-time underwriting | Klarna discloses using machine-learning techniques to enable its real-time credit-underwriting process, drawing on proprietary transaction data from its consumer and merchant network. | In production | [3] |
| Real-time fraud detection | Established ML techniques used for real-time fraud detection and prevention across the payments network, per the IPO prospectus. | In production | [3] |
| AML and sanctions screening | ML used for anti-money-laundering and sanctions screening; Finansinspektionen fined Klarna Bank AB SEK 500 million in December 2024 for AML rule violations found in a 2021-22 examination, including gaps in risk assessment and customer-due-diligence procedures. | In production | [3][11] |
| Date | Move | Detail | Source |
|---|---|---|---|
| Jul 6, 2026 | Klarna applies for a US industrial-bank charter | Klarna filed with the Utah Department of Financial Institutions and the FDIC to form Klarna Bank USA, an FDIC-insured Utah industrial bank, to fund loans with deposits and bring more payments, lending and merchant operations in-house. | [10] |
Does Klarna use AI for credit decisions?
Yes. Klarna's FY2025 20-F says its underwriting process uses ML-based credit models, is fully automated and makes a new decision for each transaction in a matter of seconds, using Klarna's own transaction history plus merchant data, credit bureau reports and open banking data.
How accurate is Klarna's credit model?
Klarna reports that its US Gini score, a 0-to-1 measure of predictive power, rose from 0.36 in 2019 to 0.78 in the fourth quarter of 2025, against 0.43 for VantageScore 4.0, according to its own credit scoring model. No independent validation of those figures is published.
What are Klarna's credit losses?
Klarna's provision for credit losses was $192 million, or 0.52% of GMV, in the second quarter of 2026. Its 20-F says US provisions fell from approximately 3.6% of GMV in 2021 to approximately 0.63% in 2025, and Klarna says its lifetime provision rate is around 0.6%.
Does Klarna run a credit check?
Klarna underwrites every purchase and uses credit bureau reports alongside its own data, per its 20-F. In the UK it said in 2022 that it runs a soft credit search for Pay in 30 and Pay in 3 and would share BNPL payment data with Experian and TransUnion.
Is Klarna's BNPL lending regulated?
In the UK, interest-free BNPL (Deferred Payment Credit) comes under FCA regulation from July 15, 2026. In the US, Klarna is subject to CFPB authority and fair-lending law; the CFPB said in March 2025 it would revoke a rule treating Pay in 4 loans as credit cards.
- Klarna Group plc Annual Report (Form 20-F) for the fiscal year ended December 31, 2025
- Klarna Q2'26 Earnings Release (Exhibit 99.2 to Form 6-K)
- Klarna Group plc Prospectus (Form 424B4)
- PS26/1: Regulation of Deferred Payment Credit (unregulated Buy Now Pay Later)
- Klarna First Quarter 2026 Results (Exhibit 99.2 to Form 6-K)
- Klarna's flexible payments are coming to Google Search and the Gemini app within Google Pay
- All you need to know about Klarna sharing BNPL payments with credit reference agencies
- Klarna Fourth Quarter 2025 Results (Exhibit 99.1 to Form 6-K)
- CFPB retreats from classifying BNPL like credit cards
- Klarna seeks U.S. bank charter in latest push beyond buy now, pay later
- Klarna får en anmärkning och sanktionsavgift
When Klarna moves on AI, you'll read it here first.
when a bank, fintech or regulator moves on AI, it is in the next morning's brief · six sourced stories · 7 am PT · free
plus every regulation, bank and agent page update, the morning after · leave any morning