Artificial Intelligence in Fintech Market

Upstart grew loan originations 86% in 2025 while its headcount rose just 18%, and 91% of its fourth-quarter loans were originated end to end with no human involvement. Artificial intelligence (AI) in fintech covers the machine-learning and generative models that financial firms buy or license to score credit, stop fraud, answer customers and automate compliance; the artificial intelligence in fintech market comprises software, model platforms, AI lending network fees and related services sold to lenders, insurers and payment firms. Douglas Insights sizes it at USD 19.10 billion in 2025, built as 46,850 buying institutions × USD 407,600 of average annual AI spend per institution. With 4.35% yearly growth in buyers and 12.35% growth in spend per buyer, the compound annual growth rate (CAGR) is 17.24%, enough to reach USD 93.7 billion by 2035. The rulebook moved mid-forecast: the Digital Omnibus on AI, Regulation (EU) 2026/1744, entered into force on 27 July 2026 and pushed the deadline for stand-alone high-risk systems such as credit scoring to 2 December 2027. Readers can set these figures beside our enterprise software research, and every number follows the Douglas Insights research methodology.

Which AI in fintech segment makes the money: fraud models, credit underwriting or chat assistants?

At 27.4% of 2025 AI in fintech spend, or USD 5.23 billion on Douglas Insights estimates, Fraud and anti-money laundering is the biggest earner because every card payment and transfer needs a real-time risk score. Customer service and conversational AI grows fastest, at 20.93% a year, as generative assistants move from pilot desks into bank contact centres.

Fraud and anti-money laundering holds USD 5.23 billion because card networks and banks score each payment before it settles, and the Bank of England found fraud detection among the top three AI uses, cited by 33% of firms. Credit scoring and underwriting holds USD 4.60 billion, 24.1%, and overtakes fraud by 2035 at USD 23.4 billion; AI lenders sell approval decisions, not seats. Customer service and conversational AI holds USD 3.02 billion and is fastest at 20.93% a year, because a chat assistant replaces call minutes that banks already price. Risk, compliance and regtech holds USD 2.60 billion, 13.6%, built on transaction monitoring and know-your-customer (KYC) checks. Trading and wealth advice holds USD 1.97 billion and grows slowest, at 13.82%, since quantitative desks have run models for decades. Back-office automation holds USD 1.68 billion, 8.8%, from document reading in loan files and claims.

Three further cuts of the USD 19.10 billion total sit in the data tables: by component (Solutions and Services), by deployment (Cloud and On-premises) and by buyer (Banks, Credit unions, Insurers, Payment firms and Fintech lenders). Services grow faster than Solutions in our build because 33% of AI implementations at UK firms already come from third parties, up from 17% in 2022. The wider model and chip spend that sits under these tools is sized in our Artificial Intelligence Applications Market report.

Why are fraud scoring and automated loan approvals boosting AI in fintech budgets?

Each year 4.35% more institutions start paying for AI in fintech tools in our model, lifting buyers from 46,850 in 2025 to 71,719 by 2035. Late-adopting banks and insurers add 1.85 points, fintech lenders and credit unions joining AI lending networks add 1.40 points, and payment firms buying fraud scoring add 1.10 points.

Late adopters among banks and insurers

The Bank of England and Financial Conduct Authority (FCA) survey of 118 firms, published on 21 November 2024, found 75% already using AI and another 10% planning to within three years. With insurers at 95% use, the remaining buyers sit among smaller lenders, brokers and asset managers. Douglas Insights credits this late wave with 1.85 points of yearly buyer growth, about 10,576 of the 24,869 institutions the AI in fintech market adds by 2035. The same survey shows why spend per buyer climbs 12.35% a year: the median firm expects its AI use cases to rise from 9 to 21 within three years.

Credit unions and fintech lenders plugging into AI underwriting

AI lending networks recruit buyers by the dozen. Upstart reported that 70% of funding for its fourth-quarter auto and home loans came from 11 partners, with an additional 13 signed for the coming year, and its 2025 loan count more than doubled to 1,497,149, up 115%. Networks scale fast. Pagaya runs a network with more than 30 lending partners. Each partner bank or credit union becomes a paying AI in fintech buyer without hiring a modelling team. We assign 1.40 points of yearly growth to this route, roughly 8,004 additional institutions by 2035.

Payment firms and merchants buying fraud scores

Payments push fraud models down to smaller firms. Featurespace, which Visa bought in a deal completed on 19 December 2024, screens more than 100 billion payment events a year and protects 500 million consumers for more than 100,000 businesses, including HSBC, NatWest and Worldpay. Mastercard added Recorded Future to feed threat intelligence into its real-time fraud scoring. As networks bundle AI fraud tools into payment services, acquirers and processors become counted buyers. Douglas Insights puts this driver at 1.10 points, about 6,289 institutions by 2035, and together the three make up the 4.35-point volume leg. Card authentication, a neighbouring layer, is covered in our 3D Secure Authentication Market report.

Spend per institution is the larger leg. Douglas Insights expects average AI in fintech spend per buyer to climb from USD 407,600 in 2025 to USD 1.31 million by 2035. The omnibus delay to 2 December 2027 gives credit-scoring buyers 16 extra months to finish model documentation before the high-risk rules bite.

What limits AI in fintech spending as credit-scoring audits and bank mergers bite?

Compliance costs for high-risk credit models, data-protection limits and lender mergers together strip 1.20 points from yearly buyer growth in AI in fintech: 0.45, 0.35 and 0.40 points respectively. Gross of these drags, revenue growth would reach 18.59% rather than 17.24%, adding about USD 11.3 billion by 2035.

High-risk obligations under the EU Artificial Intelligence Act cover creditworthiness scoring, so a European Union (EU) lender buying an underwriting model must fund risk management, logging and human oversight. The omnibus that entered into force on 27 July 2026 delays these duties to 2 December 2027 but does not remove them. Douglas Insights subtracts 0.45 points a year, worth about USD 4.12 billion of 2035 AI in fintech spend.

Data rules are the most cited regulatory brake. In the Bank of England survey, 23% of firms named data protection and privacy a large constraint, 25% cited insufficient talent, and 46% reported only partial understanding of the AI systems they run, mostly because models come from third parties. Our model removes 0.35 points of buyer growth for these frictions, about USD 3.19 billion by 2035.

Consolidation cuts the buyer count. The National Credit Union Administration (NCUA) counted 4,214 federally insured credit unions at mid-2026, down 156 from 4,370 a year earlier, a 3.57% fall, while members rose to 146.1 million. Each merger folds two AI in fintech contracts into one. Fewer buyers, bigger tickets. We take 0.40 points from the volume leg, roughly USD 3.65 billion of 2035 spend, though part of that value reappears in higher spend per surviving buyer.

How much do lenders and banks pay per AI-decisioned loan and per platform subscription?

An average buyer spent USD 407,600 on AI in fintech tools in 2025, a Douglas Insights estimate that rises to USD 1.31 million per institution by 2035. Per decision, Upstart earned USD 634.5 of fee revenue for each of its 1,497,149 AI-underwritten loans in 2025.

Disclosed revenue lines set the realised price bands. Upstart’s USD 950 million of 2025 fee revenue on USD 11.0 billion of transaction volume equals 8.64% of each dollar it helped lend. Pagaya’s USD 1.30 billion of revenue and other income on USD 10.5 billion of network volume equals 12.4%, though that figure includes income beyond fees. Platform subscriptions price lower: nCino’s USD 523.1 million of fiscal 2026 subscription revenue across more than 1,500 depository customers implies at most USD 348,700 per customer a year, 14.4% below our all-buyer average.

Douglas Insights estimates three spend bands per buyer. Community banks and credit unions pay roughly USD 60,000 to USD 250,000 a year for fraud scores, a lending model and a chat assistant. Regional banks and mid-size insurers pay USD 0.8 million to USD 4 million across several models. Global banks spend above USD 40 million, mostly on in-house platforms, cloud compute and specialist services. The average of USD 407,600 sits closer to the low band because 8,791 United States (US) banks and credit unions, most of them small, dominate the buyer count.

Who are the AI underwriting and fraud-scoring vendors that banks buy from?

Disclosed revenue makes Pagaya the largest AI in fintech vendor, at USD 1.30 billion in 2025, which Douglas Insights estimates as 6.81% of spend. Add Upstart’s fee revenue and FICO’s software arm and the top three reach only 15.7% of USD 19.10 billion, a fragmented supply base.

Pagaya’s 2025 revenue grew 26% while its network volume rose 9%, so it earned more per approved loan. Upstart’s total revenue rose 64% to USD 1.0 billion, and its fee revenue alone gives a 4.97% share. FICO reported USD 740.1 million of software revenue in fiscal 2025, up from USD 711.3 million, plus USD 1.17 billion from Scores, which we exclude as credit bureau scoring rather than licensed AI. nCino grew total revenue 10% to USD 594.8 million. Visa and Mastercard compete through acquisitions: Visa closed Featurespace on 19 December 2024, and Mastercard folded Recorded Future into its fraud scoring.

Where do banks spend most on AI in fintech, and which region adds credit-model spend fastest?

With USD 7.56 billion of 2025 spend on Douglas Insights estimates, 39.6% of the world total, North America leads AI in fintech, as thousands of US lenders buy from the deepest vendor bench. Latin America grows fastest, at 20.36% a year, from USD 936 million as digital lenders scale to USD 5.97 billion by 2035.

North America rises from USD 7.56 billion to USD 32.7 billion at 15.76% a year; the Federal Deposit Insurance Corporation (FDIC) counted 4,421 insured institutions in the second quarter of 2025, and with 4,370 credit unions that makes 8,791 US deposit-takers. Europe climbs from USD 5.02 billion to USD 22.4 billion at 16.11%, held back slightly by high-risk documentation for credit scoring. Asia Pacific grows from USD 4.81 billion to USD 28.0 billion at 19.27%, the largest absolute gain after North America, as super-app lenders and digital banks score thin-file borrowers. Latin America, at 20.36%, is fastest because new digital lenders build on AI underwriting from day one rather than retrofitting it. Middle East and Africa, the wildcard, moves from USD 764 million to USD 4.63 billion at 19.73%; mobile-money growth there could outrun our base case in any single year.

Which EU AI Act rules govern creditworthiness scoring models sold to lenders?

Creditworthiness scoring is a high-risk use under Annex III of the EU AI Act, so AI in fintech vendors selling credit models must meet risk-management, data-governance, logging and human-oversight rules. Regulation (EU) 2026/1744, dated 8 July 2026, moves the compliance date for these stand-alone systems to 2 December 2027.

For lenders, the rule turns a model into a documented product. Each credit-scoring model needs technical files, bias testing on training data, event logs and a named human able to override decisions. Fraud and anti-money laundering models face lighter duties because fraud detection is not listed as high-risk in the same way. Douglas Insights prices the compliance burden at 0.45 points of yearly buyer growth, concentrated in Credit scoring and underwriting, our USD 4.60 billion segment.

What did the Digital Omnibus on AI move for high-risk credit-scoring deadlines?

The Digital Omnibus on AI moved the deadline for stand-alone high-risk AI, including credit scoring, from 2 August 2026 to 2 December 2027, a 16-month delay. Regulation (EU) 2026/1744 was adopted on 8 July 2026 and entered into force on 27 July 2026.

Negotiators reached a provisional political agreement on 6 May 2026, and member state representatives confirmed it on 13 May 2026, so lenders had less than three months of certainty before the original date; the final text then entered into force on 27 July 2026. The delay does not cut the AI in fintech market; it spreads compliance spending across 2026, 2027 and 2028. Douglas Insights treats the 16 months as a timing shift worth roughly 0.10 points of 2026 buyer growth recovered later in the decade, inside our 0.45-point compliance restraint rather than on top of it.

How many payment events do fraud-scoring networks such as Featurespace screen each year?

Featurespace screens more than 100 billion payment events a year for over 100,000 businesses, according to Visa, which completed the purchase on 19 December 2024. That volume explains why Fraud and anti-money laundering is the largest AI in fintech segment, at USD 5.23 billion in 2025.

Scale matters. A fraud model trained on 100 billion events sees rare attack patterns that a single bank’s data never shows, so payment networks sell scores as a shared service and smaller banks rent them. Douglas Insights expects Fraud and anti-money laundering to grow 15.31% a year to USD 21.7 billion by 2035, slower than underwriting because network bundling holds unit prices down.

How far has end-to-end loan automation gone at AI lenders like Upstart and Pagaya?

Upstart originated 91% of its fourth-quarter 2025 loans end to end with no human involvement, its results show, on 1,497,149 loans for the full year. Its conversion rate rose to 19.4% from 15.1% in 2024, so more applicants reach a funded AI in fintech decision.

Automation explains the gap. Upstart’s transaction volume rose 86% to USD 11.0 billion while headcount grew 18%. Pagaya grew revenue 26% on 9% volume growth to USD 10.5 billion. Douglas Insights counts USD 21.5 billion of credit decisioned by these two networks in 2025, against which they collected USD 2.25 billion, or 10.47 cents per dollar lent. That yield is the price signal behind our 17.64% growth rate for Credit scoring and underwriting.

What if AI underwriting adoption runs cooler or hotter through 2035?

Base-case AI in fintech spend reaches USD 93.7 billion in 2035. A cooler path, with buyers growing 2.60% and spend per buyer 10.10%, gives USD 64.6 billion; a hotter path at 5.90% and 14.20% gives USD 128 billion. Spend per buyer matters most.

The cooler path, at 12.96% a year, assumes European lenders pause credit-model purchases until the omnibus deadline of 2 December 2027 passes and credit-union mergers speed up. The hotter path, at 20.94% a year, assumes generative assistants replace most contact-centre minutes and AI lending networks sign partners faster than Upstart’s 13 new partners. USD 9.37 billion of 2035 value rides on each point of yearly buyer growth. Published forecasts we read run from 17.0% to 22.7% a year; our 17.24% sits at the low end because we exclude cloud compute billed outside AI products and FICO’s credit bureau scores.

Douglas Exclusive: the AI Credit Decision Register

Douglas Insights built the AI Credit Decision Register, our own model, from the 2025 disclosures of Upstart, Pagaya, nCino, FICO and Visa’s Featurespace. It is a modelled estimate, not a survey or a count: it divides each vendor’s disclosed AI in fintech revenue by its disclosed scale, per decision, per dollar lent or per customer.

What the register shows is plain: an AI credit decision is worth far more than a fraud score. Upstart collects USD 634.5 per approved loan, while a fraud network spreads its service across 100 billion events, so even a one-cent fee per event would total USD 1 billion. On our register, lending networks earned 10.47 cents per dollar lent in 2025, which is why Credit scoring and underwriting overtakes fraud by 2035 in our model. It is not an official register. It is our compilation of company disclosures, and it shows that per-customer software such as nCino’s earns less per institution than our USD 407,600 average.

Which methodology turns 46,850 buying institutions into the AI in fintech estimate?

In our model, 46,850 buying institutions at USD 407,600 each give USD 19.10 billion in 2025, and five regional blocks sum exactly to that figure. From there, buyers add 4.35% a year and spend per buyer rises 12.35%; the product compounds at 17.24% and ends 2035 at USD 93.7 billion.

Disclosed company figures, regulator counts and the Bank of England survey feed the AI in fintech build, with published growth rates read only as a check, and country anchors exist for the United States and the United Kingdom. By 2035, 71,719 buyers × USD 1.31 million gives USD 93.7 billion, and 2026 spend is USD 22.4 billion. Three checks hold. nCino’s implied USD 348,700 per customer lands 14.4% under our average, as a single-product vendor should. Pagaya, Upstart, FICO software and nCino disclose USD 3.59 billion between them, 18.8% of our total, leaving 81.2% for in-house builds, cloud providers and private vendors. And our 17.24% CAGR sits 0.24 points above the lowest published rate.

Sources

1. Upstart Holdings Upstart announces fourth quarter and full year 2025 results (2026)
2. EUR-Lex, Official Journal of the European Union Regulation (EU) 2026/1744 (Digital Omnibus on AI) (2026)
3. Bank of England and FCA Artificial intelligence in UK financial services 2024 (2024)
4. Visa Visa completes acquisition of Featurespace (2024)
5. Mastercard Mastercard finalizes acquisition of Recorded Future (2024)
6. FDIC Quarterly Banking Profile, second quarter 2025 (2025)
7. NCUA Quarterly credit union data summary, 2026 Q2 (2026)
8. SEC EDGAR FICO fourth quarter fiscal 2025 results (Form 8-K exhibit) (2025)
9. SEC EDGAR nCino fourth quarter and fiscal year 2026 results (Form 8-K) (2026)
10. Pagaya Technologies Pagaya fourth quarter and full year 2025 results (2026)