When Technology Meets Finance: The Industries Being Rebuilt Around Money

Finance is the quietest industry to be disrupted and the loudest to talk about it. Every few months a new announcement claims that banking, payments or insurance is about to be rebuilt, and every few months the actual institutions go on moving money roughly the way they did the year before, only with a few more automated steps behind the screen. After spending the late summer of 2026 reading the funding data, the regulatory filings, the surveys and the company disclosures, I think both of those observations are correct at the same time. Something genuinely large is happening in the plumbing of money, and most of it is invisible to the customer. This study tries to describe it concretely: where the capital is going, what artificial intelligence is really doing inside banks and insurers, how tokenized assets and stablecoins have moved from experiment to infrastructure, why the measurements of all of these things disagree so badly, and what it means for people who build financial products, particularly in a market like Bangladesh where mobile money, and not cards or branches, is how most people meet the financial system. As in my other reviews this year, I have preferred primary sources, named surveys and regulators, and I have flagged claims that come from vendors or from secondary summaries.

It helps to start with why finance behaves differently from other industries when new technology arrives. Money is a promise, and promises depend on trust, law and settlement. A retailer can try a new checkout flow and lose a few sales if it fails. A bank that tries a new settlement system and gets it wrong can lose other people's money, break the law and destroy its licence in one afternoon. This is why financial institutions adopt technology in a characteristic pattern: slowly at the front, where mistakes are visible and regulated, and quickly at the back, where they are contained and can be checked. It is also why the most important changes in finance often come from the rules, which decide what is permitted, and from infrastructure, which decides what is cheap, and not from products, which decide what is fashionable. Keep that in mind, because almost every section below turns out to be a story about rules and plumbing.

Begin with the money flowing into financial technology itself, because it shows what investors believe. The headline numbers differ depending on who is counting, and the differences are instructive. Crunchbase reported that fintech start-ups raised about 28.6 billion dollars globally in the first half of 2026, up almost 23 percent from the first half of 2025 but down more than 17 percent from the second half of last year, while the number of deals fell by roughly a quarter to about 1,600. CB Insights counted 26.4 billion dollars across about 1,700 deals for the same half year, with deal volume falling 25 percent from the first quarter to the second, and described the first quarter's 762 deals as a multi year low. KPMG's Pulse of Fintech, built on PitchBook data, reported 103.1 billion dollars of global fintech investment in the first half, but that number is dominated by a single 24.3 billion dollar acquisition of a large payments company and by private equity and acquisition activity, not by venture investment. Three counts, three very different totals, and all of them honest. What they agree on is the shape: fewer deals, bigger cheques, capital concentrating behind established companies, and incumbents buying their way into the fast growing segments. Capital One's completion of its 5.15 billion dollar acquisition of Brex, which gave it a position in business spend management, is the example CB Insights highlights.

There are three lessons in that picture. The first is that fintech has matured from a land grab into a consolidation. The era in which a hundred start-ups each tried to reinvent a slice of banking has given way to one in which the survivors are being absorbed or have grown large enough to compete on scale. The second is that the regional picture is uneven. KPMG reports that the Americas attracted 86.9 billion dollars in the half year, of which the United States accounted for 80.8 billion, while funding in Asia Pacific fell from 7.1 billion to 4.6 billion, and India's own tracker puts its first half total at 2.2 billion dollars, the best half year since 2023, concentrated in lending and payments and in companies that were already unicorns. The third is that the sectors drawing the most interest are infrastructure, wealth management, enterprise automation and digital assets, which tells you that investors now prefer to fund the pipes more than the storefronts. Insurance technology drew about 3.7 billion dollars in KPMG's count, and fintech focused cybersecurity a modest 550 million, which strikes me as low given what the next sections describe.

Now turn from capital to what is actually happening inside the institutions, starting with banks and artificial intelligence. The visible pattern at the largest banks is two tracked. On one track are assistants for employees, such as JPMorgan's internal language model suite, reported to be available to more than 200,000 staff, and Goldman Sachs's firm wide assistant, reported to have reached roughly 46,000 employees. On the other track are narrower systems in high stakes operations such as fraud scoring, anti money laundering, document processing and trade reconciliation. Goldman Sachs has said it worked with Anthropic to develop autonomous agents for trade accounting, reconciliation, client onboarding and compliance, and large banks including JPMorgan, Citigroup and others are reported to be embedding models across middle and back office work. I mention the Goldman and Anthropic arrangement because it was widely reported and because it illustrates a general point, while noting that I am relying on press coverage of the banks' own statements and that I work for no one in this story.

What do the surveys say about how far this has gone? Capgemini's 2026 research on financial services found that banks name customer service, fraud detection, loan processing and customer onboarding as the top processes for deploying AI agents at scale, at 75, 64, 61 and 59 percent of respondents respectively, but also that only about one in ten institutions had actually implemented agents at scale. Roughly a third of banks said they were building agents in house, and nearly half of financial institutions were creating new jobs to supervise agents. That last figure is the one I find most revealing. When a bank creates a role whose job is to watch the machine, it is acknowledging that the technology changes the work without removing the accountability. A Goldman Sachs research note from March, as summarised by secondary sources, concluded that AI had not yet produced a meaningful economy wide productivity effect, which is consistent with the pattern I described in my broader review: heavy adoption, rare transformation.

I would like to be careful with the many precise figures that circulate about banking AI, such as savings of billions in fraud prevention or accuracy rates of ninety eight percent. Some originate in company statements about specific systems, but in aggregator articles they are routinely detached from their source, rounded up and repeated until they look like established fact. The credible, repeatable observations are the ones that survive independent scrutiny: fraud detection and document processing are the most mature uses because their outputs can be checked against known outcomes; customer onboarding and know your customer checks are speeding up because they involve reading documents and comparing records, which models do well; and software engineering inside banks is one of the clearest gains, with some institutions reporting very large productivity improvements on coding tasks. The riskier uses, such as advice, credit decisions and anything where a model's output reaches a customer without review, move slowly because regulators demand explanations, and because the consequences of a mistake fall on the bank.

That leads to what I consider the central point about artificial intelligence in banking: the scarce resource is not intelligence but permission. Every bank can buy capable models. What distinguishes the leaders is clean data they are allowed to use, governance that lets them deploy systems without breaching rules, and the organisational ability to redesign a process around a machine rather than stapling the machine to an old process. A bank that has grown through acquisitions and carries five incompatible data systems will find its agent cannot see two of them. A bank with tight controls and good data will put the same agent to work in weeks. For a small fintech, this is a hopeful observation, because the same logic means that a young company with clean data and clear controls can move faster than an incumbent whose advantage is size.

Tokenization, the practice of representing ownership of financial assets as digital tokens on shared ledgers, is where the gap between the story and the numbers is widest, so it deserves a careful look. The cleanest recent data come from RWA.xyz, which tracks tokenized real world assets. As of June 2026 it showed roughly 26.7 billion dollars of what it calls distributed assets, meaning tokens that actually trade and can be transferred, excluding stablecoins, set against more than 345 billion dollars of what it calls represented value, which includes off chain assets not yet issued as transferable tokens. Tokenized United States Treasuries and money market products were the largest category, at about 14.8 billion dollars across 82 products and some 65,000 holders in early June, up from about 8.9 billion at the start of the year, with a seven day yield around 3.35 percent. By July, one data summary put the category near 16 billion. Other counts in circulation, including one reporting more than 34 billion dollars in May, differ because they define scope differently. The biggest single products include BlackRock's tokenized money market fund, reported around 2.8 billion dollars in July, and a Circle product reported near 3 billion, with Franklin Templeton's fund also in the billions.

What should we make of this? On one hand, the growth is real and fast: tokenized Treasuries crossed ten billion dollars for the first time in February and roughly tenfold since mid 2024, according to several trackers. On the other hand, thirty billion dollars is tiny against the roughly trillions of dollars in conventional money markets, and the gap between distributed and represented value shows that most of what is called tokenization still sits in structures that are not yet usable as open financial instruments. I read the growth as a story about cash management, not about revolution. The most successful tokenized products are the ones that do something treasurers already want, namely holding short term government debt with yield and moving it around outside banking hours, and they succeed because they fit into existing legal structures: a token wrapping a regulated fund share, issued by a regulator approved manager. Forecasts that tokenized assets will reach trillions by 2030, made by consultancies, are projections of ambition. The measured fact is that a few large asset managers have shown that a regulated fund can live on a shared ledger without breaking the law, and that is a meaningful step even if it is not yet a transformation.

Stablecoins have moved furthest from experiment to infrastructure, and the data here are among the most useful in this review, provided you notice that supply and usage tell different stories. Total stablecoin supply stood around 302 to 306 billion dollars in late September, with Tether at about 184 billion and Circle's USDC at about 75 billion, which together account for more than eighty percent. That is below the peak of roughly 320 billion reached in May, so the market has contracted slightly over the summer, even as usage has risen. Rise's third quarter report, drawing on Allium data, identified payments of at least 401 billion dollars in the first eight months of 2026, up about 42 percent, while other analyses put adjusted transfer volume in the tens of trillions of dollars, with USDC accounting for the majority of adjusted volume despite its smaller supply. These two sets of figures describe the same market in incompatible ways. The 401 billion is a lower bound on identifiable payments for goods and services. The tens of trillions include trading, arbitrage and automated flows between wallets. Anyone who quotes a single stablecoin volume number without saying which kind it is, is telling you almost nothing.

The regulatory side, which I covered in my review of the year, has moved decisively. The United States has built a federal framework through its GENIUS Act, and federal agencies have been publishing the implementing rules throughout 2026: the Comptroller of the Currency in March, the FDIC and the Treasury's financial crimes and sanctions bodies in April, Treasury again in August, and the Federal Reserve in late September. The Act takes effect at the earlier of January 2027 or a hundred and twenty days after final rules. The practical consequence is that stablecoin issuance in the United States is becoming a regulated activity resembling banking, with reserve, redemption, capital and compliance obligations, and that banks and licensed entities will increasingly be the ones issuing them. Combined with comparable licensing regimes elsewhere, this is the change that turns a crypto trading instrument into something a bank treasurer or a payments company can responsibly use, which is why I regard regulation, not technology, as the main driver of the stablecoin story.

The claims made about stablecoins as a payment method need the same sober treatment. Industry sources say business to business cross border payments are the fastest growing segment, with monthly volumes surging many times over during 2025, and that Asia accounts for the majority of payment volume. Those claims come mostly from stablecoin companies and their research arms, so I would hold them loosely, but they match what I would expect: the corridors where correspondent banking is slow and expensive, such as remittances and supplier payments between emerging markets, are where an alternative can win. Everyday retail payments in markets that already have cheap instant systems are a very different matter, and there I expect little displacement. A stablecoin offers faster settlement across borders. It does not offer a better way to buy a cup of tea in Dhaka than a mobile wallet already does.

That brings the discussion to Bangladesh, where the central fact of financial life is mobile money. bKash, Nagad and Rocket, supervised by Bangladesh Bank under its mobile financial services rules, have turned the phone into the main account for tens of millions of people, and industry press reports daily transaction values in the hundreds of millions of dollars. Figures vary widely by source and year. One promotional release for a crypto wallet put the number of mobile financial services accounts above 238 million in a country of about 174 million people, with daily transaction values above 260 million dollars, while another industry summary claimed roughly 440 million dollars a day and more than 120 million users across the two largest platforms. Those cannot all be right, and part of the explanation is that accounts are not users, since many people hold several. The structural developments are clearer than the headline counts. Bangladesh Bank has been building interoperability between banks, wallets and payment service providers, first through the national payment switch and, according to reports, through a new interoperable instant payment system built on open source software that was announced in September 2025, designed to give the country ownership of its own payments rail. Observers have also raised governance concerns about the sector, including regulatory disputes and the use of wallets for illegal gambling and money laundering, documented by Transparency International Bangladesh, which is a reminder that scale without controls creates its own risks.

I should add a caution about one item I came across. A press release in March announced that a crypto wallet company had enabled users to convert stablecoins into taka and send them to bKash and Nagad accounts. I could not verify the legal status of such a service in Bangladesh, whose central bank has historically taken a restrictive position on dealing in cryptocurrencies, and a company announcement is not evidence that the activity is lawful or widely used. I mention it because it illustrates how global stablecoin infrastructure is reaching local wallets by informal and partly unregulated routes, and because a founder building in this space must treat the legal question as the first question, not an afterthought. In the meantime, the lawful and more consequential development is domestic: faster, cheaper and more interoperable local rails, which reduce the very friction that stablecoins claim to solve.

Insurance is the part of finance where the technology story is easiest to tell and hardest to verify, so I will be disciplined about it. Conning's survey, as relayed by trade sources, found full adoption of artificial intelligence among insurers rising from 8 percent to 34 percent in a year, and adoption of large language models from 18 percent to 63 percent, with about nine in ten insurers somewhere between pilot and production. GlobalData's July 2026 survey found that 34.4 percent of respondents expected underwriting and risk profiling to benefit most, followed by customer service at 20.4 percent and claims management at 18.3 percent. Specific reported results include a major British insurer's claims transformation reported to have saved close to 100 million pounds, an expansion of its AI underwriting summary tool into critical illness cover, and Pacific Life's survey of 103 senior underwriting executives finding that AI is delivering value through efficiency and better use of data and not by replacing human judgement. A new AI native carrier aimed at start-ups raised 108 million dollars at a 630 million dollar valuation in January.

As elsewhere, the gap between adoption and returns is the real story. One survey summary captured it well: insurance is further into AI adoption than most industries and no further into returns. The pattern I find convincing is the one that applies to every regulated industry. Claims is the most mature use because the work involves reading documents and photos, a speed gain is easy to see, and a human can review the exceptions. Underwriting is the most valuable use because it affects the price of risk, which is the heart of the business, but it is also the most sensitive, since an opaque model that sets prices can create discrimination and regulatory problems. My expectation is that the next two years bring steady progress in claims and in underwriting support, a growing role for models in fraud detection, and a slower, more contested arrival of models that directly set prices, with regulators asking insurers to prove that their systems are fair and explainable.

There is an uncomfortable counter-current running through all of this, and it concerns fraud. Every technology that makes it cheaper to produce convincing text, voices and documents also makes it cheaper to forge them. Banks and insurers that deploy models to detect fraud are defending against criminals who use the same models to attack, and the economics favour the attacker in the short run, since an attack needs to succeed only once. Voice cloning of customers and executives, synthetic identities that pass automated onboarding, forged claims documents and invoices that look perfect are all real problems that the industry is already facing, though precise measurements are hard to find. I regard fintech security, which drew only 550 million dollars of investment in the half year according to KPMG, as underfunded relative to the risk. If I had to name a category likely to be more valuable in two years than the market currently believes, it would be verification: proving who a person is, that a document is genuine, that an instruction really came from the account holder, and that an automated action was authorised.

What patterns can we draw from banking, tokenization, stablecoins, payments and insurance together? The first is that incumbents are not losing. In nearly every part of finance I examined, the established institutions are adopting the new technology on their own terms, acquiring the start-ups that prove a concept, and using their data and their licences as advantages. The large banks have the data, the budgets and the regulatory relationships. The asset managers that tokenized Treasury funds did so inside existing legal wrappers. The stablecoin rules favour entities that can meet bank like requirements. The story of finance in 2026 is not disruption of incumbents by outsiders but absorption of the new by the old, with a few outsiders becoming incumbents themselves.

The second pattern is that measurement is the weakest part of the whole field. Fintech funding totals differ by a factor of three depending on the source. Tokenized asset totals differ by an order of magnitude depending on whether you count represented or distributed value. Stablecoin volume differs by two orders of magnitude depending on whether trading is included. Mobile money accounts in Bangladesh are reported anywhere from tens of millions to hundreds of millions. When the basic facts of a field are this unsettled, anyone building a business should be sceptical of any market size and should do the arithmetic from the bottom up, starting with real customers and real transactions in a defined niche. A founder who writes that a market is worth hundreds of billions of dollars in a business plan is usually writing a sentence that conveys no information.

The third pattern is that compliance is turning into a product. Reserve rules, sanctions screening, monitoring, reporting and supervision of automated agents all need software, and the companies that make compliance cheap and reliable for others will earn durable revenue because their customers cannot easily switch and cannot afford to fail. In the stablecoin world, regulators have effectively created a market for licensed infrastructure. In banking, the new roles for supervising agents create demand for tools that log, audit and constrain what those agents do. In insurance, explainability requirements create demand for systems that document how a price was derived. The founder who sees regulation as an obstacle will find it frustrating. The one who sees it as a specification for a product will find customers.

The fourth pattern is the shift from software that assists to software that acts. A model that helps a bank employee draft a note carries little risk. An agent that moves money, approves a claim or opens an account carries a great deal. The institutions that move toward acting systems need a different kind of discipline: strict limits on what an agent can do, logs of everything it did, a human who owns each outcome, and the ability to reverse mistakes. The principle I have applied in my own work on payment integrations, that every decision should be traceable, every action reversible where possible, and a named person accountable, becomes more important, not less, as automation deepens. I suspect that the first serious public failure of an autonomous financial agent, whenever it comes, will be a failure of authorisation and oversight, not of intelligence, and that it will reshape the rules.

Now for my own predictions, offered as judgement and not as forecasts from any institution. Over the next eighteen months I expect fintech funding to remain concentrated, with deal counts staying low and large rounds going to infrastructure, enterprise automation, wealth and digital assets, and I expect continued acquisitions of mid sized fintechs by banks and payments companies. I expect the regulated stablecoin market in the United States to open in early 2027 with a wave of announcements by banks and payment companies, followed by a quieter period in which actual use grows in business to business and treasury settings long before it reaches consumers. I expect tokenized money market funds and Treasuries to keep growing at a rapid percentage rate from a small base, driven by collateral and cash management uses, and I expect the headline numbers to remain a battle of definitions. I expect most large banks to report real savings in operations, fraud and software development, and few to report transformation of their customer facing business. I expect insurance to see the quickest measurable gains in claims, and I expect the first regulatory actions about unfair or opaque AI pricing to arrive within two years.

For Bangladesh in particular, my expectations are different, because the starting point is different. The country already has what most markets are still trying to build: a population that pays by phone, regulated mobile money providers with enormous reach, and a central bank that is investing in open, interoperable infrastructure. The next step is likely to be deeper interoperability and the entry of banks and digital banks into a space that wallets have dominated, rather than a leap to tokenized assets. The opportunities I see are in the layers beneath and around the wallets: reliable reconciliation for businesses that accept payments from several providers, merchant tools that make acceptance easy for small shops, compliance and fraud monitoring suited to local patterns, agent banking tools for rural areas, and credit scoring for people and small businesses that have transaction histories but no collateral. Language matters as well, because customer service, onboarding and fraud alerts in clear Bengali are valuable and still done poorly by many systems. And because stablecoin and crypto rules in the country are restrictive, I would treat any business that depends on them as carrying a legal risk that no technological advantage can offset.

There are also risks to the whole picture that I want to name. A sharp correction in technology valuations or a stress event in stablecoin reserves could change the tone of the sector quickly, as has happened before. A major fraud or an AI related breach at a bank could prompt a regulatory backlash. The geopolitical shocks described in my other reviews could affect payments, remittances and cross border flows in unpredictable ways. And the legal framework for tokenized assets and stablecoins remains new, with the first big test cases still ahead. I hold my views on these topics with moderate confidence and I expect parts of them to be wrong. The sections of this study that I am most sure about are the structural ones: that incumbents are absorbing the new technology, that regulation drives infrastructure, that measurement is unreliable, and that trust and verification become more valuable as automation spreads. The sections I am least sure about are the quantitative forecasts, which I would treat as informed guesses.

A note on method, since finance is a field where numbers are often used to impress. Many of the figures above were found on aggregator sites, vendor blogs and promotional releases that repeat one another. Where I could, I have relied on the data providers themselves, such as Crunchbase, CB Insights, KPMG and RWA.xyz, on regulators and on named surveys, and I have said when two sources disagree. I have left out several striking statistics because I could not trace them to a credible origin. Anyone using a figure from this review should return to its source, check what exactly is being counted, and note the date. This study reflects what was available at the end of September 2026.

The short version is this. Technology is rebuilding finance from the back office outward, and the visible changes at the front are much smaller than the invisible ones behind. Capital is concentrating behind fewer, larger and more established companies, and incumbents are absorbing the winners. Artificial intelligence is widely used in banks and insurers but deeply embedded only where data, governance and process allow it, and the scarce resource is permission, not intelligence. Tokenization is growing rapidly in cash management while remaining tiny against conventional markets, and stablecoins are being pulled into regulation that will favour licensed institutions. In Bangladesh, mobile money is the foundation and the opportunity lies in the layers around it. Across all of it, measurement is poor, compliance is becoming a product, and the value of verification and accountability rises with every step towards automation. The people who will build lasting businesses in this space are those who treat trust as the product, who read the rules as a specification, and who keep a human accountable for every decision a machine makes with other people's money.

Sources: Crunchbase News, fintech funding in H1 2026 (https://news.crunchbase.com/fintech/funding-rises-deals-slump-h1-2026/); CB Insights, State of Fintech Q1 2026 (https://www.cbinsights.com/research/report/fintech-trends-q1-2026/); KPMG, Pulse of Fintech H1 2026 (https://kpmg.com/xx/en/what-we-do/industries/financial-services/pulse-of-fintech.html); CB Insights Q2 2026 fintech summary via freewritings (https://www.freewritings.law/2026/08/fintech-funding-trends/); Business Standard on India fintech funding in H1 2026 (https://www.business-standard.com/content/press-releases-ani/analytics-insight-releases-india-fintech-funding-report-for-h1-2026-companies-raise-2-2-billion-126081300724_1.html); Capgemini Research Institute, World Cloud Report in Financial Services 2026, via Business Wire (https://www.businesswire.com/news/home/20251112597684/en); Value Add VC on AI in financial services (https://valueaddvc.com/blog/ai-in-financial-services-2026-what-jpmorgan-goldman-and-blackrock-are-actually-doing); CCG Catalyst on banks and agentic AI (https://www.ccgcatalyst.com/thought-leadership/commentary/ai-in-banking-just-got-real/); RWA.xyz data as reported by Spoted Crypto (https://www.spotedcrypto.com/rwa-tokenization-market-size-2026-guide/) and by Asset Haus (https://asset.haus/blog/tokenized-us-treasuries-guide); MetaMask on real-world asset tokens (https://metamask.io/news/real-world-asset-tokens-what-crypto-wallet-users-need-to-know-in-2026); Stablecoin Beat market data (https://stablecoinbeat.com/charts/market-cap/); Rise Q3 2026 stablecoin payments report via Inside Deep Tech (https://www.insidedeeptech.com/stablecoin-payments-report-q3-2026/); TokenPost on stablecoin market capitalization (https://www.tokenpost.com/news/business/25481); Federal Register, OCC GENIUS Act proposed rule (https://www.federalregister.gov/documents/2026/03/02/2026-04089/implementing-the-guiding-and-establishing-national-innovation-for-us-stablecoins-act-for-the); American Bar Association on the FinCEN and OFAC proposal (https://www.americanbar.org/groups/business_law/resources/business-law-today/2026-june/treasurys-proposed-stablecoin-compliance-framework/); Insurance Times on GlobalData underwriting research (https://www.insurancetimes.co.uk/news/underwriting-remains-insurers-top-ai-use-case-despite-broader-adoption/1459251.article); Outcome Catalyst on Conning insurance AI adoption (https://www.outcomecatalyst.com/blog/insurance-ai-data-trends-2026); Pacific Life 2026 Underwriting Outlook Survey (https://www.businesswire.com/news/home/20260325915067/en); Transparency International Bangladesh, governance in mobile financial services (https://www.ti-bangladesh.org/images/2025/report/mfs/Executive-Summary-Mobile-Financial-Services-Sector-En.pdf); PaymentBrief on Bangladesh payments (https://paymentbrief.com/markets/bangladesh/); Bitget Wallet press release, March 2026 (https://www.globenewswire.com/news-release/2026/03/18/3258383/0/en/Bitget-Wallet-Launches-Bank-Transfer-in-Bangladesh-Enabling-Stablecoin-Payouts-to-bKash-and-Nagad.html).