AI & Tech Intelligence · Finance
AI in Finance and Banking: How It Works, Who Uses It, and What It Risks
How banks use AI in 2026: fraud detection, credit scoring, trading and generative AI — with adoption data from McKinsey, the Bank of England, the IMF and Evident.
TL;DR — AI in finance is the use of machine learning and, increasingly, generative AI to automate and improve tasks such as fraud detection, credit scoring, trading, compliance and customer service across banks, insurers and fintechs.
McKinsey · Bank of England / FCA · IMF · Financial Stability Board | 2,651 words · 16 sections | data: CSV + JSON
Executive summary
AI in finance is the application of machine learning and generative AI to core banking, insurance and investment tasks — fraud detection, credit decisions, trading, compliance and customer service. It is no longer experimental: the Bank of England and FCA found that 75% of UK financial-services firms used AI in 2024, up from 58% two years earlier, with insurers (95%) and international banks (94%) furthest ahead. The economic prize is large. McKinsey estimates generative AI alone could add $200–340 billion a year to global banking, equal to 2.8–4.7% of industry revenues, mainly through productivity gains. Real deployments back this up: JPMorgan tops the Evident AI Index for the fourth year running and runs an in-house 'LLM Suite' used by more than 200,000 employees; Klarna's OpenAI-powered assistant handled two-thirds of customer-service chats — the work of 700 agents — in its first month; and Stripe's Radar reduces fraud by an average of 32% by scoring payments across a $1.9-trillion network. Banks apply AI across four workhorse areas: catching fraud and money laundering in real time, scoring credit for thin-file borrowers, executing and optimising trades, and, most recently, deploying generative-AI copilots and agents. The risks are equally concrete. The IMF, FSB and Bank of England warn that AI-driven trading and dependence on a handful of model providers could deepen volatility and concentration risk, while bias in credit models draws regulatory scrutiny — the EU AI Act classifies creditworthiness assessment as high-risk from August 2026. The trajectory is clear: finance is one of AI's largest and fastest-scaling markets, but one where trust, explainability and regulation now set the pace.
“Across the global banking sector, generative AI could add between $200 billion and $340 billion in value annually, or 2.8 to 4.7 percent of total industry revenues, largely through increased productivity.”
Key findings
Generative AI is a $200–340bn annual prize for banking
McKinsey estimates generative AI could add $200 billion to $340 billion in value annually across global banking, chiefly by lifting productivity in software engineering, customer operations, marketing and risk. Banking is among the industries with the highest potential value relative to revenue, because so much of its work is language- and data-intensive.
Source: McKinsey — The economic potential of generative AI · 2023 · confidence: High
Three-quarters of financial firms already use AI
The Bank of England and FCA's 2024 survey of 118 firms found 75% already using AI and a further 10% planning to within three years. Adoption is highest in insurance (95%) and international banks (94%). Most firms reported ten or fewer live use cases, showing scaling is still early.
Source: Bank of England / FCA — AI in UK financial services 2024 · 2024 · confidence: High
AI fraud detection cuts losses by roughly a third
Stripe's Radar scores every transaction using signals from a network processing over $1.9 trillion in annual payments, reducing fraud by an average of 32% for businesses that use it. Machine-learning models also cut card-testing attacks in Stripe Checkout by around 80%, illustrating why fraud detection is finance's most mature AI use case.
Source: Stripe — Radar · 2024 · confidence: High
Where AI in finance stands in mid-2026
As of 30 July 2026, AI in finance has moved decisively from pilots to production. Generative-AI copilots are now standard at the largest banks — JPMorgan's LLM Suite reaches more than 200,000 employees and the firm has topped the Evident AI Index for a fourth consecutive year, ahead of Capital One and Royal Bank of Canada. NVIDIA's 2025 industry survey found roughly 70% of financial firms reporting AI-driven revenue gains and about 60% reporting cost reductions. At the same time, the regulatory perimeter is tightening: the EU AI Act's high-risk obligations — which cover AI used for creditworthiness assessment and credit scoring — begin to apply on 2 August 2026, and the IMF, FSB and Bank of England have all flagged model concentration and AI-amplified market volatility as emerging systemic concerns.
What is AI in finance, and why now?
AI in finance is the use of machine learning and generative AI to automate, augment and improve financial tasks — including fraud detection, credit scoring, algorithmic trading, anti-money-laundering (AML) checks, regulatory compliance, customer service and portfolio management. It spans banks, insurers, asset managers and fintechs, and ranges from decades-old statistical models to the large language models (LLMs) deployed since 2023.
The 'why now' is a mix of capability and economics. Modern models can read unstructured text, code, images and transaction streams at a scale humans cannot, and finance is unusually data-rich and language-heavy. McKinsey estimates generative AI could add $200–340 billion a year to global banking, equal to 2.8–4.7% of industry revenues, largely through productivity (McKinsey, 2023). That potential, combined with falling inference costs, has pushed AI from back-office experiment to boardroom priority.
Adoption confirms the shift. The Bank of England and FCA found 75% of UK financial firms using AI in 2024, up from 58% in 2022 (Bank of England / FCA, 2024), while Stanford's AI Index reports organisational AI adoption climbing above 78% across sectors in 2024 (Stanford HAI, 2025). Finance is now one of the largest and fastest-scaling markets for AI.
How do banks actually use AI?
Banks use AI across four workhorse areas — risk and fraud, lending, markets, and customer-facing operations — plus a growing layer of internal productivity tools. Most institutions run many narrow models rather than one general system: the Bank of England found the majority of firms had ten or fewer live use cases in 2024 (Bank of England / FCA, 2024).
The most valuable deployments are concrete and measurable. JPMorgan's document-review tool COiN interprets commercial-loan agreements in seconds, saving an estimated 360,000 lawyer-hours a year, and its in-house LLM Suite now reaches more than 200,000 employees. The table below maps the leading use cases to what they do, the benefit they deliver, and a real-world example.
| Use case | What AI does | Benefit | Real example |
|---|---|---|---|
| Fraud detection | Scores every transaction in real time against learned patterns | Fewer losses and false declines | Stripe Radar cuts fraud by ~32% on average |
| Anti-money-laundering (AML) | Flags suspicious networks and reduces false positives | Lower compliance cost, better detection | HSBC uses Google Cloud AML AI to screen transactions |
| Credit scoring & lending | Assesses creditworthiness, including thin-file borrowers | Broader access, faster decisions | Upstart uses ML for consumer-loan underwriting |
| Algorithmic trading | Executes and optimises trades on signals and market data | Speed, execution quality, lower cost | Renaissance, Two Sigma and bank desks use ML strategies |
| Customer service | Handles chats and queries via conversational AI | 24/7 service, lower cost per contact | Klarna's AI handles 2/3 of service chats |
| Document & code work | Summarises contracts, drafts memos, writes code | Large productivity gains for staff | JPMorgan COiN saves ~360,000 legal hours/yr |
Is AI used in fraud detection and anti-money-laundering?
Yes — fraud detection is the single most mature and widely deployed use of AI in finance. Machine-learning systems score every payment in milliseconds against thousands of behavioural signals, catching patterns that static rules miss. Stripe's Radar, which learns from a network processing over $1.9 trillion in payments a year, reduces fraud by an average of 32% for businesses that use it and has blocked tens of millions of additional high-risk transactions using probabilistic stolen-card models (Stripe).
The same techniques power anti-money-laundering (AML). Traditional rules-based AML systems generate enormous false-positive rates — often above 90% — forcing costly manual review. AI models trained on transaction networks cut those false positives while surfacing genuinely suspicious activity; HSBC, for example, screens transactions with Google Cloud's AML AI, and most tier-one banks now run machine-learning transaction monitoring alongside legacy rules.
The advantage is speed and adaptability: fraud and laundering typologies change constantly, and models retrained on fresh data adapt faster than hand-written rules. The Bank of England's 2024 survey found firms rank cybersecurity and fraud among the top areas where AI both helps and introduces new risk (Bank of England / FCA, 2024), since the same generative tools also lower the cost of committing fraud.
How is AI used in credit scoring and lending?
AI is used in credit scoring to predict a borrower's likelihood of default from far more data than a traditional scorecard, and to underwrite loans faster and, in some cases, for people with thin or no credit files. Machine-learning underwriters such as those used by Upstart and Zest AI incorporate hundreds of variables — cash-flow, education, employment and behavioural signals — rather than a handful of bureau attributes, and lenders report both higher approval rates at the same loss level and faster decisions.
The upside is financial inclusion and efficiency; the risk is bias and opacity. Models trained on historical lending data can reproduce or amplify discrimination against protected groups, and complex models are harder to explain to a declined applicant — a legal requirement in many jurisdictions. This is why regulators treat credit AI as especially sensitive.
The EU AI Act classifies AI systems used to evaluate the creditworthiness of natural persons, or to establish their credit score, as high-risk (Annex III), with obligations on data governance, human oversight, transparency and bias testing applying from 2 August 2026. In the US, the Equal Credit Opportunity Act's adverse-action rules already require lenders to give specific reasons for denials, constraining pure 'black-box' scoring. Explainable AI is therefore a compliance necessity in lending, not a nice-to-have.
How is AI used in trading and financial markets?
AI is used in trading to generate signals, execute orders and optimise portfolios at speeds and scales beyond human capacity. Quantitative funds such as Renaissance Technologies and Two Sigma have used machine learning for years, and bank trading desks apply it to execution, market-making and transaction-cost analysis. NVIDIA's 2025 industry survey found trading and portfolio optimisation among the highest-ROI generative-AI use cases reported by financial firms (NVIDIA, 2025).
The benefits are real but bounded: faster execution, tighter spreads and better liquidity in normal conditions. The concern is what happens under stress. The IMF's October 2024 Global Financial Stability Report warned that greater use of AI-driven trading could increase market speed, volatility and herding — with algorithms reacting similarly to the same macro news — and could shift activity toward less-regulated non-bank intermediaries (IMF, 2024).
Regulators including the Bank of England, ECB and the Financial Stability Board have flagged a related risk: concentration. If many firms rely on the same few third-party foundation models and cloud providers, correlated failures or coordinated behaviour could amplify shocks (FSB, 2024). AI in markets, in other words, may raise efficiency in calm periods while making tail events sharper.
What is generative AI doing in banking?
Generative AI in banking means using large language models to draft, summarise, code and converse — powering customer-service assistants, employee copilots and, increasingly, autonomous 'agents' that complete multi-step tasks. It is the fastest-growing layer of AI in finance, and the one that moved the field from back-office analytics to front-office productivity after 2023.
The customer-service case is the clearest. Klarna's OpenAI-powered assistant handled two-thirds of its customer-service chats — 2.3 million conversations — within its first month, did the work of 700 full-time agents, cut resolution time from 11 minutes to under 2, and was projected to drive a $40 million profit improvement in 2024 (Klarna / OpenAI, 2024). Klarna later rebalanced toward human agents for complex cases, a useful reminder that automation has limits.
Internal copilots are scaling even faster. JPMorgan's in-house LLM Suite reaches more than 200,000 employees for research, drafting and analysis, and the bank tops the Evident AI Index for the fourth year running (Evident, 2025). Across the sector, NVIDIA's 2025 survey found roughly 70% of financial firms reporting AI-driven revenue gains and about 60% reporting cost reductions (NVIDIA, 2025) — evidence that generative AI is delivering measured returns, not just pilots.
How widely is AI adopted in finance, and how big is the market?
AI adoption in finance is now mainstream and still climbing. The Bank of England and FCA found 75% of UK financial firms using AI in 2024, up from 58% in 2022, with a further 10% planning to adopt within three years (Bank of England / FCA, 2024). Stanford's AI Index reports organisational AI adoption above 78% across sectors in 2024, rising toward 88% in its 2026 edition (Stanford HAI).
The value at stake is measured in hundreds of billions. McKinsey puts generative AI's annual potential for global banking at $200–340 billion (McKinsey, 2023). The table below collects the headline adoption and value figures with their sources so each can be cited independently.
| Metric | Figure | Scope / year | Source |
|---|---|---|---|
| Gen AI value potential, banking | $200–340bn/yr (2.8–4.7% of revenue) | Global, 2023 | McKinsey |
| Financial firms using AI | 75% (up from 58% in 2022) | UK, 2024 | Bank of England / FCA |
| Firms reporting AI revenue gains | ~70% | Global, 2025 | NVIDIA State of AI in Financial Services |
| Organisational AI adoption | >78% (rising to ~88%) | Cross-sector, 2024–26 | Stanford HAI AI Index |
| Klarna chats handled by AI | 2/3 (2.3m in first month) | Global, 2024 | Klarna / OpenAI |
| JPMorgan LLM Suite users | >200,000 employees | Global, 2025 | JPMorgan / Evident |
What are the risks of AI in banking, and how are they regulated?
The main risks of AI in banking are bias in credit and pricing models, opacity ('black-box' decisions), data-privacy exposure, model concentration among a few providers, AI-amplified market volatility, and new fraud and cyber-attack vectors. The Bank of England's 2024 survey found that four of the top five perceived AI risks relate to data — privacy, quality, security and bias — with cybersecurity close behind (Bank of England / FCA, 2024).
Systemic risks are drawing central-bank attention. The IMF warned in October 2024 that AI trading could increase volatility and herding (IMF, 2024), and the FSB flagged that dependence on a small number of third-party model and cloud providers could create correlated points of failure (FSB, 2024). The table below pairs each major risk with its typical mitigation.
Regulation is catching up. The EU AI Act classifies credit scoring and creditworthiness assessment as high-risk, with obligations on data governance, transparency, human oversight and bias testing applying from 2 August 2026. Existing rules also bite: the US Equal Credit Opportunity Act requires explainable adverse-action reasons, GDPR governs automated decision-making, and model-risk-management expectations (such as the US SR 11-7 guidance) already apply to AI models in banks.
| Risk | Why it matters | Typical mitigation |
|---|---|---|
| Bias & discrimination | Credit/pricing models can reproduce historical bias | Fairness testing, protected-attribute audits, EU AI Act high-risk controls |
| Opacity ('black box') | Hard to explain declines; legal exposure | Explainable AI, adverse-action reasons, human oversight |
| Data privacy & security | Sensitive financial data at risk | Governance, anonymisation, GDPR compliance |
| Model concentration | Few providers = correlated failure | Vendor diversification, third-party risk management (FSB) |
| Market volatility | AI trading can herd and amplify shocks | Circuit breakers, stress testing, monitoring (IMF) |
| Fraud & cyber | Gen AI lowers cost of attacks | AI defence, red-teaming, transaction monitoring |
How are European and Nordic banks and fintechs using AI?
European finance combines strong AI deployment with the world's most developed AI rulebook. International banks operating in the UK reported 94% AI usage in 2024 (Bank of England / FCA, 2024), and the EU AI Act — the first comprehensive AI law — makes credit scoring a high-risk application from August 2026, shaping how European lenders build models.
The Nordic region is a notable early adopter, led by fintech and conversational AI. Sweden's Klarna delivered one of the sector's most-cited generative-AI results, with its assistant handling two-thirds of customer-service chats and doing the work of 700 agents in month one (Klarna / OpenAI, 2024). Nordic banks were also among the first to scale virtual assistants: Nordea's 'Nova' handles well over 200,000 conversations a month across the Nordics, SEB pioneered its 'Aida' assistant, and Swedbank deployed 'Nina' — with most of the region's largest banks using the same conversational-AI platform.
The Nordic pattern — high digital penetration, near-cashless payments and concentrated banking markets — makes the region a useful proving ground for AI in finance. It also shows the trade-off in sharp relief: rapid automation of service and operations alongside strict EU data-protection and, increasingly, AI-Act obligations that raise the bar for transparency and oversight.
What is the outlook for AI in finance?
The outlook is continued, fast scaling — but with returns and regulation, not hype, now driving decisions. NVIDIA's 2025 survey shows firms doubling down on AI investment as a majority report measurable revenue and cost benefits (NVIDIA, 2025), and McKinsey's $200–340 billion estimate remains only partly captured, leaving substantial headroom (McKinsey, 2023).
The next frontier is agentic AI — systems that don't just answer questions but execute multi-step workflows such as reconciling payments, preparing compliance filings or managing parts of the trade lifecycle. JPMorgan and other leaders already describe production 'agents,' and the Evident AI Index shows a widening gap between AI leaders and laggards (Evident, 2025), suggesting AI capability is becoming a competitive dividing line in banking.
The binding constraints will be trust and governance. As the IMF, FSB and Bank of England have warned, concentration, explainability and market-stability risks grow with deployment, and the EU AI Act's high-risk regime lands in August 2026. The banks that win will be those that pair aggressive deployment with auditable, explainable and well-governed models — because in finance, a wrong-but-plausible AI output is more dangerous than no output at all.
Scoreboard (machine-readable data)
Every headline indicator with its value, period, source and confidence. Free to reuse under CC BY 4.0.
| Indicator | Value | Period | Source | Conf. |
|---|---|---|---|---|
| Gen AI annual value potential, global banking (upper) | 340 usd_bn | 2023 | McKinsey — The economic potential of generative AI | High |
| UK financial firms using AI | 75 pct | 2024 | Bank of England / FCA — AI in UK financial services 2024 | High |
| International banks using AI | 94 pct | 2024 | Bank of England / FCA — AI in UK financial services 2024 | High |
| Average fraud reduction, Stripe Radar | 32 pct | 2024 | Stripe — Radar | High |
| Financial firms reporting AI revenue gains | 70 pct | 2025 | NVIDIA — State of AI in Financial Services 2025 | Medium |
| Klarna customer-service chats handled by AI | 66 pct | 2024 | Klarna / OpenAI | High |
Methodology & verification
This report synthesises figures from named primary and authoritative sources published in 2023–2026: McKinsey (generative-AI value estimates), the Bank of England and FCA's 2024 AI in UK Financial Services survey (adoption), the IMF's October 2024 Global Financial Stability Report and the FSB's 2024 report (systemic risk), the Stanford HAI AI Index (cross-sector adoption), the Evident AI Index 2025 (bank AI maturity ranking), NVIDIA's State of AI in Financial Services survey (industry ROI), and company disclosures from Klarna/OpenAI, JPMorgan and Stripe (use cases and results). Every statistic is attributed inline to its source. Figures were verified against these sources on 30 July 2026; where a single number could not be independently corroborated it was labelled with lower confidence or omitted. No figures were invented, estimated or interpolated.
Data dictionary
| Field | Type | Description |
|---|---|---|
| genai_value_potential | number (USD billions/year) | McKinsey's estimated annual value generative AI could add to global banking, largely via productivity. |
| ai_adoption_rate | number (percent) | Share of financial-services firms reporting active use of AI in a given survey and geography. |
| fraud_reduction_rate | number (percent) | Average reduction in fraud reported for businesses using a machine-learning fraud-detection system. |
Frequently asked questions
What is AI in finance?
AI in finance is the use of machine learning and generative AI to automate and improve financial tasks such as fraud detection, credit scoring, trading, compliance and customer service. It is used across banks, insurers, asset managers and fintechs, and ranges from long-established statistical models to large language models deployed since 2023.
How do banks use AI?
Banks use AI mainly for fraud detection and AML, credit scoring and lending, algorithmic trading, and customer service, plus internal productivity tools. In 2024, 75% of UK financial firms reported using AI (Bank of England / FCA), with most running ten or fewer live use cases.
Is AI used for fraud detection?
Yes — fraud detection is the most mature use of AI in finance. Systems score every transaction in real time against learned patterns; Stripe's Radar, learning from over $1.9 trillion in annual payments, reduces fraud by an average of 32% for businesses on its network (Stripe).
How is AI used in credit scoring?
AI predicts a borrower's default risk from far more data than a traditional scorecard, enabling faster decisions and lending to thin-file borrowers. Because it can encode bias, the EU AI Act classifies credit scoring as high-risk, with obligations on transparency and oversight applying from 2 August 2026.
How much value can generative AI add to banking?
McKinsey estimates generative AI could add $200 billion to $340 billion annually to global banking — 2.8 to 4.7 percent of industry revenues — largely through higher productivity (McKinsey, 2023).
Which bank leads in AI?
JPMorgan Chase tops the Evident AI Index for the fourth consecutive year in 2025, ahead of Capital One and Royal Bank of Canada, based on 70+ indicators across 50 major banks. Its in-house LLM Suite reaches more than 200,000 employees (Evident, 2025).
What are the risks of AI in banking?
The main risks are bias in credit models, opaque decisions, data-privacy exposure, model concentration among a few providers, AI-amplified market volatility, and new fraud vectors. The IMF, FSB and Bank of England have all warned of systemic risks from AI trading and provider concentration.
Is generative AI used in financial customer service?
Yes. Klarna's OpenAI-powered assistant handled two-thirds of customer-service chats — 2.3 million conversations and the work of 700 agents — in its first month, cutting resolution time from 11 minutes to under 2 (Klarna / OpenAI, 2024).
Glossary
- Algorithmic trading
- The use of computer programs, increasingly AI-driven, to execute and optimise trades based on market data and signals at high speed and scale. ↗
- Credit scoring
- The assessment of a borrower's creditworthiness; AI-based scoring uses many more variables than traditional scorecards and is classified as high-risk under the EU AI Act. ↗
- Generative AI
- AI models, typically large language models, that produce new text, code or images; in finance used for copilots, customer service and document work. ↗
- Anti-money-laundering (AML)
- Processes and systems used to detect and report suspicious financial activity; AI reduces false positives in transaction monitoring. ↗
Embed & cite this report
Free to reuse under CC BY 4.0. Embed the live-updating widget on your site, or cite the report directly — always with attribution to Affärslivet.
Embed (HTML) — auto-updating
<iframe src="https://xn--affrslivet-s5a.com/en/embed/ai-in-finance" width="100%" height="520" style="border:1px solid #e3e3e6" title="AI in Finance and Banking: How It Works, Who Uses It, and What It Risks — Affärslivet" loading="lazy"></iframe> <p style="font:12px sans-serif">Source: <a href="https://xn--affrslivet-s5a.com/en/reports/ai-in-finance">Affärslivet</a></p>
APA
Affärslivet Research. (2026). AI in Finance and Banking: How It Works, Who Uses It, and What It Risks. Affärslivet. Version 1.0. https://xn--affrslivet-s5a.com/en/reports/ai-in-finance
MLA
Affärslivet Research. "AI in Finance and Banking: How It Works, Who Uses It, and What It Risks." Affärslivet, 2026-07-30, https://xn--affrslivet-s5a.com/en/reports/ai-in-finance.
BibTeX
@techreport{affarslivet_ai_in_finance,
title = {AI in Finance and Banking: How It Works, Who Uses It, and What It Risks},
author = {{Affärslivet Research}},
year = {2026},
note = {Version 1.0},
url = {https://xn--affrslivet-s5a.com/en/reports/ai-in-finance}
} License CC BY 4.0 — free to cite, embed and republish with attribution to Affärslivet. Data also as CSV / JSON.
Sources
- McKinsey — The economic potential of generative AI: The next productivity frontier
- Bank of England / FCA — Artificial intelligence in UK financial services 2024
- IMF — Global Financial Stability Report, October 2024 (Chapter 3: AI and capital markets)
- Financial Stability Board — The Financial Stability Implications of Artificial Intelligence (2024)
- Evident AI Index for Banks (2025)
- NVIDIA — State of AI in Financial Services 2025
- Klarna / OpenAI — AI assistant case study
- Stanford HAI — AI Index Report
Part of Affärslivet AI Intelligence
This report is one part of Affärslivet's source-cited AI knowledge layer. Start with the big picture:
- The State of AI 2026 — our free, source-cited annual report
- AI Statistics 2026 — 230+ source-cited AI facts & figures
- The AI Intelligence hub — all our indexes, rankings, model & lab profiles