In July 2025 Commonwealth Bank of Australia cut 45 customer service jobs after launching an AI voice bot. A month later it reversed the decision and apologized, because call volumes had gone up, not down. Meanwhile, a data-driven fraud program that includes machine learning helped the US Treasury prevent and recover more than $4 billion in fraudulent and improper payments in a single fiscal year. Both results are real, and together they set the terms for the fintech trends that matter in 2026: investors, customers and regulators now reward what works in production, at scale and under audit.
The fintech market in 2026: capital returns, for fewer and larger companies
After three years of falling investment, the global fintech market turned in 2025. KPMG's Pulse of Fintech, the fintech report built on PitchBook data, counts $122.8 billion of venture capital, private equity and M&A investment in 2025, and $103.1 billion in the first half of 2026 alone, which puts the year on course for a four-year high.
The number of deals tells the other half of the story.
It fell from 5,965 in 2023 to 5,328 in 2025, and only 2,100 deals closed in the first six months of 2026. The ten largest deals took 62% of the money, led by Global Payments' $24.3 billion acquisition of Worldpay and FIS's $13.5 billion purchase of Global Payments' issuer business. Capital is concentrating on mature fintech companies with proven business models. Emerging startups still raise money, $9.8 billion at the early stage in the half year by KPMG's count, and many of them are built around AI or financial infrastructure.
From growth at any cost to profitable growth
Revenue has caught up with the story.
BCG's Global Fintech Report 2026, written with FT Partners, puts global fintech revenue at $504 billion in 2025, up 22%, growing more than four times as fast as incumbent financial institutions. 74% of the largest public fintechs are profitable, and their average EBITDA margin rose by 400 basis points to 20%. BCG sells advice to this market, so these are a consultancy's figures rather than audited totals, but the direction matches the investment data.
Two more numbers from the same report show where the fintech industry is heading. Fintech revenue is still only about 4% of global financial services revenue, so the room to grow is large. And in 2025, for the first time on BCG's record, fintechs completed more acquisitions than banks and other incumbents did: 659 deals against 589. The fintech industry spent a decade disrupting traditional financial services; now the scaled fintechs buy licenses, customer books and products from them, while banks buy the capabilities they could not build in time.
For finance professionals and institutional investors, the market trends point the same way.
Growth alone no longer earns a premium. Unit economics do, along with a license or a bank partner that survives scrutiny, and a cost base that falls as volume rises. A financial technology platform that needs more people for every new million customers shows up as a risk in due diligence.
AI in financial services: where it pays off and where it has not
Every fintech report of the past two years ranks artificial intelligence as the leading theme, and the money confirms it: KPMG counts $21.4 billion invested in AI-related fintech deals in the first half of 2026, close to the $23.6 billion of the whole of 2025. For a leader, the useful test is narrower than adoption. In which financial processes has AI produced a measured result in production, and where has a launch been followed by a quiet reversal?
Adoption is broad, understanding is thin
The most detailed regulator survey of AI in financial services comes from the Bank of England and the FCA. In their 2024 survey of UK firms, 75% already used AI and another 10% planned to within three years. Foundation models, the family that includes large language models, made up 17% of use cases, and a third of all use cases were implementations bought from third parties. 46% of firms said they had only a partial understanding of the AI they use, against 34% who claimed complete understanding, mostly because of models bought from vendors. Most use is still low-stakes: firms rated 62% of their use cases as low materiality and 16% as high, and only 2% of use cases decide fully on their own.
Fraud detection and AML: the strongest measured results
Fraud detection is where machine learning has the longest record and the clearest numbers, because every prevented loss is counted. The US Treasury's fraud and improper payment work prevented and recovered more than $4 billion in fiscal 2024, up from $652.7 million the year before. Identifying and prioritizing high-risk transactions accounted for $2.5 billion of it, and machine learning applied to check fraud recovered another $1 billion. Firms in the Bank of England survey rank anti-money laundering and fraud among the three areas with the highest perceived benefits from AI today.
The limits are just as practical. Fraud models need labeled data, retraining as criminals change tactics, and a way to explain to a customer, and to a regulator, why a payment was stopped. Criminals use AI too: UK Finance's 2026 fraud report describes gangs exploiting it to industrialize their operations. A fraud model that is not retrained loses value every month it runs.
Customer service at scale: Erica, Klarna and the route back to a human
Bank of America's Erica shows what patient automation achieves. Launched in 2018, the assistant passed 3 billion client interactions by August 2025, averaging more than 58 million a month, and the bank reports that more than 98% of users find the information they need. Bank of America presents Erica as a way to cut call center volume and free its specialists for complex conversations, and it took seven years to reach those numbers.
Klarna moved faster. In February 2024 it announced that its OpenAI-powered assistant had handled 2.3 million conversations in its first month, two-thirds of all customer service chats, doing the work of 700 full-time agents and expected to add $40 million to profit in 2024. In May 2025 chief executive Sebastian Siemiatkowski told Bloomberg that the cost-cutting had gone too far, and Klarna started recruiting people again so that customers could always reach a human. Commonwealth Bank's reversal followed three months later.
The lesson concerns design.
An assistant that answers routine questions and hands the rest to a person earns its keep; one that stands between a customer who has just been defrauded and a human being costs more in complaints, churn and regulatory attention than it saves on salaries.
Underwriting and credit: explainability decides
Credit decisions are where large language models meet their hardest limit. A model that can produce a fluent, wrong answer and cannot show which inputs drove a decision does not fit a lender's duty to explain a refusal, and the EU AI Act classifies AI used to assess the creditworthiness of individuals as high-risk. The Digital Omnibus agreed in 2026 moved the obligations for such stand-alone high-risk systems to 2 December 2027. Lenders get more time, and the destination is unchanged: risk management, data governance, logging and human oversight for each model.
The pattern that works keeps a machine learning model with established explainability methods in charge of the decision, and uses generative AI around it, to read bank statements, payslips and documents. 81% of the AI-using firms in the Bank of England survey already apply some explainability method, most often feature importance and Shapley values. Model risk management, long a discipline at large banks, now applies to every fintech that lends.
Product configuration: AI drafts the change, people approve it
Some of the most useful AI in financial services never talks to a customer.
Launching a new loan, a fee schedule or an insurance product has traditionally meant a requirements document, a specification and weeks of developer time inside a core system, followed by a release window. AI can now draft that configuration directly from the business description, which moves product changes from the IT backlog back to the people who own the product.
Openkoda, an insurance policy administration system built by Stratoflow's sister company, shows how this works in a regulated product. In its AI Product Builder, a product manager describes a new product or a change in plain English, and the AI drafts a typed, reviewable change-set of coverages, pricing, rules and workflow steps. The change-set is built from a fixed catalogue of operations rather than free-form code, nothing is applied until a person approves it, and every change is versioned and recorded in the audit trail, so policies already sold stay on the version they were bound under.
That design answers the objections that stall generative AI elsewhere in finance. The output is checked before it takes effect, the model cannot invent an operation the system does not support, and an auditor can see who approved what and when. AI proposing and people deciding, inside a system of record, is the pattern we expect to last in financial product work.
Coding assistants follow the same rule. 84% of developers use or plan to use AI tools, according to the 2025 Stack Overflow Developer Survey, but 46% distrust the accuracy of the output, so the gains hold only inside code review and automated tests.
The missed promises: pilots, agents and running costs
The gap between pilot and production is wide.
Gartner, writing about all industries, expects over 40% of agentic AI projects to be canceled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls. KPMG's fintech investors say the same in their own terms: as companies learn what artificial intelligence costs to run, money is moving from pilot projects to companies that can prove value. Inference bills, evaluation work, data clean-up and compliance reviews rarely appear in the first business case, and they decide whether a pilot reaches production.
The matrix below sorts the use cases by the evidence available in 2026. When we put AI into a financial system, as a scoped pilot with evaluation and access controls, the first deliverable is the number that decides whether it goes further.
Real-time and account-to-account payments become the default
In Europe, instant payment is now a legal requirement.
The Instant Payments Regulation required euro area payment providers to receive instant credit transfers from January 2025, and to send them from 9 October 2025, at no more than the price of a standard transfer and with a Verification of Payee check that matches the payee's name to the account before the money leaves.
The effect shows in ECB statistics. Instant transfers made up 15% of the credit transfers processed by euro area retail payment systems in the first half of 2024, and 25% in the second half of 2025. By value the share is 8%, so the shift starts with everyday digital payments.
In the UK, open banking payments, which move money directly from one bank account to another without a card, now exceed 40 million a month, with more than 19 million active user connections, according to Open Banking Limited. Variable recurring payments, which let a customer authorize a series of account-to-account payments under agreed limits, are the next step for bills and subscriptions, where account-to-account payments compete directly with cards.
For payment businesses, instant has two consequences that reach well beyond the payment screen. Fraud decisions now happen in the seconds before an irreversible payment leaves, not in a next-day review. And the ledger, liquidity management and support all have to run around the clock, weekends included, which traditional systems built around overnight batches were never designed to do. KPMG counts $44.2 billion of payments investment in the first half of 2026, the largest of any segment of the fintech industry. Our guide to payment integration covers the build side.
Open banking and open finance: regulation sets the pace
Open banking has become a payments and data channel with real volume.
Its next stage, open finance, which extends consent-based data sharing to savings, pensions, insurance and lending, depends on rules that are still moving on both sides of the Atlantic.
Europe: PSD3 and the Payment Services Regulation
The European Parliament and the Council reached provisional political agreement on PSD3 and the new Payment Services Regulation on 27 November 2025, and the Parliament's ECON committee approved the text in May 2026. As of September 2026 the package still awaits formal adoption. Beyond the open banking rules, the Regulation moves fraud liability. When a scammer poses as a payment provider's employee and tricks a customer into approving a payment, the provider has to refund the full amount once the customer has reported it to the police and the provider. Online platforms that fail to remove fraudulent content after being told about it become liable to the providers that refunded the victims.
United States: the 1033 rule in limbo
The CFPB finalized its Section 1033 personal financial data rights rule in October 2024. In 2026 the rule is on the books but effectively unenforceable: a federal court in Kentucky enjoined the Bureau from enforcing it, the appeal is stayed, and the CFPB is rewriting the rule, including whether banks may charge fees for data access (Ballard Spahr's Consumer Finance Monitor, June 2026). For US fintech companies whose business models rely on aggregated account data, the fee question matters more than any technical standard.
Consent-based account data also shortens onboarding. Income and account ownership checks that once needed uploaded statements can run in the background, which is what makes instant account opening practical for lenders, brokers and neobanks, and what makes the US uncertainty expensive.
Embedded finance after Synapse: the ledger is the product
Embedded finance puts payments, accounts, cards, loans or insurance inside a non-financial product: a marketplace paying its sellers, restaurant software offering working capital, a ride-hailing app with a driver card. Stripe, which supplies much of the plumbing, says businesses running on its platform generated $1.9 trillion in 2025, equivalent to 1.6% of global GDP.
The model has also shown its weak point. In April 2024 Synapse, a banking-as-a-service intermediary between fintech apps and their partner banks, filed for bankruptcy, and customers lost access to their money for weeks or months. The CFPB's action of August 2025 alleged that Synapse failed to keep adequate records of where consumers' funds were and to make those records match the partner banks' records. The shortfall was put at $60 million to $90 million.
Embedded finance is a ledger problem before it is a distribution opportunity.
Every brand that offers a financial product needs a record of who owns which money, reconciled with the bank every day, that still holds when a partner fails.
This is work we know from the inside. For Legerity we were the main development team on FastPost, a rule-driven accounting platform whose in-memory ledger processes a billion financial transactions in under an hour in the cloud and answers queries over billions of balances in under a second. The engagement ran for nearly ten years, and Legerity was acquired by insightsoftware in April 2022. At that speed, reconciliation stays a routine daily job instead of a quarter-end project.
Fraud prevention: liability moves to the payment provider
Fraud costs fintechs and financial institutions money every day, and it is the area where regulation has changed the economics fastest. UK Finance's Annual Fraud Report 2026 counts £1.28 billion stolen in 2025. Unauthorized fraud, such as card fraud and account takeover, fell 5% to £703.4 million, and the banking industry prevented 70p of every £1 attempted. Authorized push payment (APP) fraud, where victims are tricked into sending money themselves, rose 19% to £576.4 million across 248,070 cases, the most since comparable records began.
Investment scams are the largest and fastest-growing type, at £221.5 million, up 40%. Losses to scams that redirect a payment, such as invoice fraud and impersonation of police or bank staff, fell to their lowest in years, which suggests that payee name checks and bank warnings work against the scams they were designed for. The growth comes from what UK Finance calls malicious payees: the victim pays exactly the account they intended, a fake investment manager or a romance scammer, and no name check can help.
Since 7 October 2024 the UK's Payment Systems Regulator has required payment providers to reimburse APP fraud victims up to £85,000 per claim, with the sending and receiving providers splitting the cost equally. In 2025 banks returned £354.3 million, 61% of all APP losses, and 88% of in-scope losses were reimbursed in the first year of the rules. The EU's Payment Services Regulation will add its own refund duty for impersonation fraud.
Every reimbursed pound is now a direct cost, which turns fraud prevention into a product decision with its own profit-and-loss line. Receiving providers, often fintechs that compete on fast onboarding, pay half of each claim, so spotting mule accounts at sign-up has a price. The tools that work combine fraud detection models with behavioral economics: a short pause and a specific warning before a first payment to a new investment platform interrupts the urgency that scammers manufacture, where a generic warning gets clicked through.
Financial inclusion: the next customers are already online
The World Bank's Global Findex 2025 finds that 79% of adults worldwide now have an account, and that 40% of adults in developing economies saved in a financial account in 2024, up 16 percentage points since 2021 and the fastest rise in more than a decade. 84% of adults in low- and middle-income economies own a mobile phone, and the World Bank credits mobile money and digitally enabled accounts with much of the change, so the next wave of financial inclusion will arrive through financial technology. Yet only 56% of adults worldwide could reliably find extra money in an emergency.
The work of financial inclusion has moved from opening accounts to products that people keep using: savings that start small, credit priced on cash-flow data instead of a thin credit file, and financial education built into the app at the moment of a decision rather than offered as a separate course. Behavioral economics earns its place here too, in defaults and round-ups that make saving the easy option. For the fintech industry, financial inclusion is a growth segment with its own cost discipline: products for low-balance customers only work on a cost base that stays low as volume rises.
Compliance requirements are now product requirements
A financial technology company that treats regulation as a legal review at the end of a release finds out late, and expensively, what it should have built. The rules below change products directly, for fintechs and incumbent financial institutions alike: what a payment screen shows, who pays for a scam, what an AI model must log. This calendar is worth pinning to the roadmap.
| Rule | Where | Status in October 2026 | What it changes in the product |
|---|---|---|---|
| DORA | EU | Applies since 17 January 2025 | ICT risk management, incident reporting, resilience testing and contracts with critical ICT providers |
| Instant Payments Regulation | EU | Euro area: receiving since January 2025, sending and Verification of Payee since 9 October 2025; other member states follow through 2027 | Instant transfers at the price of a standard transfer, a payee name check before every transfer, round-the-clock operations |
| APP fraud reimbursement | UK | Since 7 October 2024 | Sending and receiving providers split refunds of up to £85,000 per claim, which puts a price on weak onboarding and weak payment checks |
| PSD3 and Payment Services Regulation | EU | Provisional agreement 27 November 2025; formal adoption pending | Full refunds for bank-impersonation fraud, platform liability for fraudulent content, new open banking rules |
| AI Act, high-risk systems | EU | Obligations for stand-alone high-risk systems, including credit scoring of individuals, from 2 December 2027 | Risk management, data governance, logging, human oversight and documentation for each model |
| Section 1033 data rights rule | US | Final rule of October 2024 enjoined; CFPB rewriting it | Open questions on who may access account data and whether banks may charge for it |
For traditional financial services firms, the same calendar lands on older core systems, which is where most of the cost sits; our companion piece on banking technology trends looks at it from the bank's side. Where a core system blocks a deadline, legacy system modernization in tested steps usually beats a replacement program that lands after the rule does.
What we left out of the 2026 fintech trends, and why
There is no section here on the metaverse, Web3 or blockchain in everything. Emerging technologies earn a place in a list like this with adoption, revenue or regulation behind them, and those three have shown none of it at scale. Digital assets get a paragraph because the evidence has changed: the US GENIUS Act of July 2025 created a federal framework for payment stablecoins, and KPMG counted $11.1 billion of digital-asset investment in the first half of 2026, while the CLARITY Act for the wider crypto market was still stalled. Stablecoin settlement is worth a watching brief and a pilot where cross-border costs are high; for most financial institutions it is not yet a mainstream rail.
Six moves for fintech leaders in the next twelve months
- Put a number on every AI use case before it leaves the pilot: losses prevented, handling time or conversion, against the full running cost, inference and review included.
- Keep a route to a human in every automated customer journey, and track complaints as closely as the share of chats the bot closes.
- Give fraud reimbursement an owner in product, and price friction in the payment flow against the losses it prevents.
- Make the ledger the source of truth for every embedded product, reconciled daily with each partner bank.
- Run treasury, fraud monitoring and support for instant payments around the clock, weekends included.
- Write the regulatory dates into the roadmap: AI Act high-risk obligations in December 2027, PSD3 and the Payment Services Regulation once adopted, and the 1033 rewrite in the US.
Most of these moves meet the same obstacle: systems built for overnight batches and quarterly releases. Financial innovation in 2026 depends less on technological innovation itself than on how fast a team can change a ledger, a payment flow or a fraud model without breaking what already runs. If you need a partner for that work, our comparison of the best fintech software development companies sets ten of them side by side. Since 2013 Stratoflow has built and modernized business-critical Java systems for financial technology companies, from the FastPost ledger to payment integrations, in weekly increments and without risky rewrites. If one of these trends is stuck on your systems, a thirty-minute call with one of our architects will show whether we are the right team for it.