© 2026 Katya Malinova. First published 25 September 2026. Licensed under CC BY-NC-ND 4.0: you may share it with credit to the author and a link to this page; no commercial use and no modified versions. Slides: slides.com/kmalinova/f26-intro-to-fintech.
Download the PDF · The short version: Class Summary Review · All Module 1 materials
How this was made. Written from the recording of the 8 September 2026 lecture and the Module 1 slides, with the help of Claude, an AI, and reviewed by the author. Revised on 25 September 2026 after an independent fact-check; where this text departs from a slide, it says so. AI makes mistakes: it can garble a name, misdate something, or state a connection more confidently than the evidence supports. Please flag any discrepancy you find.
This is a companion to the deck. The slides are mostly pictures: a map, a warehouse, a diagram with a dozen boxes, a statue. That works in a room with someone talking over them and works much less well a week later on their own. What goes missing is the link: why this slide follows that one, and why a crossbow and a railroad gauge are in a finance course at all. Those links are what this document supplies.
The fourteen numbered sections run in the same order as the slides, so if you scroll the deck you can scroll this alongside it. Under each: what the slides show, what was said about them, and what the room made of it. Come here when a slide does not make sense on its own, or for the discussion, which exists nowhere else.
What is not here. The opening of the class (introductions and course logistics) is not written up.
On numbers. The figures on the slides give magnitude and direction: big or small, rising or falling, ahead of us or behind. Read them for that, and spend your attention on why they matter.
Three parts, and the question mark on the third is deliberate. FinTech v1.0: where it came from. FinTech v2.0: why finance was going to be disrupted, and why then. FinTech v3.0? What is happening now; this may spill into the next lecture. The organising claim for the whole module: technology has always reshaped trading, and trading has often been first. Everything else in the course (payments, lending, wealth, crypto, regulation) is a later instance of the same process. Hold one question as well, because it comes back in section 11: what problem are you solving?
The map shows two telegraph lines: Paris–Brussels, and Aachen–Berlin. Between Brussels and Aachen, nothing: 122 km with no wire. Exchanges operated at both ends, so price information had commercial value at both ends, and the only way across the gap was a horse-drawn courier. Paul Reuter used carrier pigeons instead: roughly two hours, against six or more by train. He sold Paris prices in Berlin.
The line was finished and the pigeons were obsolete inside a year. Reuter moved to London and began selling speed over the wire itself. The arbitrageur became the infrastructure. That is the sentence to keep; it recurs.
The arbitrage lasted one year, and high-frequency trading is the same story 150 years on. Orders from a single large trade are split across venues; because the venues sit in different physical places, the pieces do not arrive simultaneously. A fast trader who sees the first piece land can reach the second venue before the rest of the order does. Michael Lewis's Flash Boys is the popular account.
The Canadian version of the pigeon story. The TMX data centre is in Markham; Alpha Exchange used to sit in downtown Toronto. That physical gap was arbitraged exactly as the Brussels–Aachen gap was. Canada now has about a dozen equity marketplaces, run by eight operators; the US has several dozen.
Why your broker app only shows you TSX prices. Exchanges earn revenue three ways: listings, trading, and data, and data is a large share. A broker can decline to subscribe to a venue you have never heard of without you noticing. Declining to subscribe to the TSX is not an option. So the incumbent's data fees are simultaneously a revenue stream and a barrier to entry for competing venues, and the incumbent has no reason to push for reform. Hold this: it is the same shape as the crossbow ban, a few sections on.
Dual-listed stocks and arbitrage. If a stock trades on both the TSX and a US venue, algorithms are very good at the resulting arbitrage: it is the same claim on the same underlying, so buying on one and selling on the other is straightforward. The risk is that prices move while the messages travel. Note the vocabulary: that is arbitrage exposed to execution risk, not the risk-free, zero-capital arbitrage of your core finance course. Two extensions that matter later:
Why shares with more votes sometimes trade lower. The answer is liquidity. High-vote classes are typically small issues that trade thinly. If you are buying for capital appreciation, you have to be able to get out to realise it, and if you can't get out, the appreciation is theoretical. As put in class: liquidity is hard to define, but you know it when you see it, and you certainly know when it evaporates.
Discussion box hidden.
The skill being sold on the floor was speed of hand and voice in one room. That is a barrier, and barriers get arbitraged. Most NYSE trading is now electronic, and the matching engine sits in a data centre in Mahwah, New Jersey. The floor survives in a hybrid role (floor brokers and designated market makers still take part, notably in the opening and closing auctions), but it is no longer where most of the trading happens. (The exchange began in 1792 with 24 brokers who agreed, by tradition under a buttonwood tree, to charge at least a quarter-percent commission and to give each other preference: a cartel with a shade tree. The floor and its rents came later; Mahwah is where most of the trading went.)
Why the machines won: many trading strategies are mechanical or probabilistic, and computers are faster, cheaper and more reliable. Virtu makes money on only about half of its trades, yet its 2014 IPO filing reported a single losing day in 1,238: the law of large numbers, industrialised.
The human market maker's edge used to be information and relationships: every order in a listed name came through you, and you could read a counterparty's bluff and refuse to trade with them next time. That edge collapsed when the same stock started trading across dozens of venues at once. No human reads forty order books.
Note the direction of travel: the biggest disruptors, the HFT firms, came from outside the traditional system. (Only partly: some people came out of banks, and every bank now runs such a desk.) This is the first appearance of a claim the whole course rests on.
Old skill set: good and fast with numbers, brash, a strong vocabulary in profanities, overwhelmingly male. New skill set: programming, statistics, econometrics, and today you would add fluency with AI tools. Entire career streams disappear. What survives is not competing with a software engineer on their own ground; it is knowing what the engineers should be building.
And yet the impact on markets was positive. Post-earnings announcement drift, the long-standing anomaly that good-news stocks kept drifting up after the announcement, taught in every finance and accounting course, has largely disappeared in recent data (Martineau, University of Toronto; Critical Finance Review 2022). The price now jumps and stays. Markets got more efficient as trading became automated (the study documents the change; attributing it to automation is the lecture's reading). That is the standing theme of the course: technology improves markets.
Disrupted people are not destroyed people. In the 1890s horse manure was a real and growing problem for big cities. (The famous prediction that London would be buried under it has never been traced to a contemporary source, so treat it as legend.) Then the car arrived, and new occupations appeared: people who had driven horses became drivers. New career streams appear. That is not a promise that the transition is painless, only that the count of jobs is not fixed.
The same question was put to ChatGPT in 2023 and again on 4 September 2026. In 2023 it returned a list of products and the words “dynamic” and “transformative”. In 2026 it returned a test: something is fintech when technology materially changes at least one of: who provides the financial service, how it is provided, what it costs, what information is used, the market structure, or the underlying product. On that test Stripe is fintech and a bank buying a faster server is not.
Two things to notice. The answer improved because the model got better at thinking rather than listing: the answer is a test of the answerer. And neither version mentions software that transacts on your behalf, which is where the module ends up.
Worth its own line: is Apple Pay safer than tapping the card? Yes, and the mechanism is tokenisation (which has nothing to do with blockchain). When you add a card to Apple Pay, the device receives a token, a distinct number bound to that device. Your watch and your phone carry different numbers, and both differ from the card. Check the last four digits on an Apple Pay receipt against your card. If the token leaks, it is useless without the device; a stolen card number plus an expiry date is not. This comes back in the payments module.
Presto came up as a contrast: a closed-loop system, usable only inside the network that issued it.
Discussion box hidden.
This section is deliberately dated: the slides come from a 2018 talk, and they are shown in 2026 to ask whether the prediction landed. Finance still exists.
Figures from the Toronto Financial Services Alliance (now Toronto Finance International), 2018 data year 2017: about 275,000 direct financial-services jobs in the Toronto CMA, 8.3% of employment, and about 13.6% of the Toronto economy. Note how the headline employment figure travels: an April 2018 release said “over 800,000” (roughly the Canada-wide count), the current TFI page says “almost 400,000” (direct plus indirect jobs in Toronto), and direct Toronto employment was about 275,000. When a headline number moves by a factor of three, find out what is being counted.
50% of executives believed 40% or more of those jobs would change fundamentally within three to five years. That window closed in 2023. The people running banks were not naive about this; the constraint was legacy systems, not intelligence.
The 2018 list of “future” technologies: cloud 100%, predictive analytics 98%, AI 93%, blockchain 73%. The top three arrived. Blockchain is the one where 2018 was more optimistic than 2026 has justified.
The 2018 list of the five most critical skills: change management, statistical analysis and data mining, working with new technology platforms, adaptability, creativity and innovation. Change management has not been displaced. Statistical analysis is the item most exposed to the last three years: the tools have levelled that field considerably. A reasonable 2026 answer would push creativity and innovation much further up, precisely because it is the hardest to automate. Worth asking yourself what your list would be.
This is the analytical core of the module, and the chart in it returns on the last slide.
| Finding | Source |
|---|---|
| Finance did not always pay more. Relative wages in finance were high in the 1920s and 1930s, fell to roughly parity with the rest of the private sector by 1980, and rose again after that. | Philippon & Reshef, QJE 2012 |
| Finance's share of GDP was high in the 1920s, low in the 1950s and 1960s (under 4%), and high again after 1980, reaching about 8% in the 2000s, tracking intermediated assets over GDP. | Philippon, AER 2015 |
| For 130 years, intermediation did not get cheaper. The ratio of intermediaries' income to intermediated assets stayed at roughly 1.5–2% from 1880 to 2010, about two cents a year to create and maintain a dollar of intermediated assets. | Philippon, AER 2015 |
| A 2019 update reports signs of declining intermediation costs after the financial crisis, against a long historical pattern of remarkably persistent costs. | Philippon, NBER WP 2019 |
| 1980: finance wages level with other industries. 2006: 170% of them. | Philippon & Reshef, QJE 2012 |
| Harvard graduates entering finance: 6% in 1969 to 1973, 28% in 2008. | Goldin & Katz, AER 2008 |
| Stanford MBAs working in finance earned roughly three times as much as those outside it. | Oyer, JF 2008 |
So the money went to the people inside. The question is what they were being paid for.
Which sets up the whole field: fintech's proposition is to charge you less for intermediation. That is what “disruption” means here: competing away a rent rather than inventing a new financial function.
Gutenberg lost the printing press to his lender. Johann Fust financed him, sued in 1455 to recover his loans with interest, won, and carried on printing with Peter Schoeffer. The innovator lost the business; the financier kept it. Among the first things printed at scale were indulgences for the Church, the trade Luther attacked sixty years later. The incumbent's new channel carried the reformation against it.
High-cost industry plus high rents implies disruptive entry is bound to happen. What 2008 added was demand for it: savings were destroyed and the reputation of the industry collapsed to the point where people did not volunteer that they worked in a bank.
The image is Fearless Girl facing the Charging Bull on Wall Street. Note for accuracy: the statue was installed in March 2017 by State Street Global Advisors, and its subject was gender diversity on corporate boards, not the 2008 crisis. It is used here as an image of confronting the incumbent, not as a document of the crisis.
Technological innovation removes barriers along three dimensions: what can be done, how much can be done, and by whom it can be done. Many innovations move the power to do things from a selected few to the masses, decentralisation, loosely. They fundamentally change established networks, intermediaries and value-extraction processes. And the general form: technology disrupts a group of people who built a living around a technological restriction, and in doing so disrupts the government power exerted through that group.
The examples on the slide, and what each one actually removed:
| Innovation | The barrier it removed |
|---|---|
| Phoenician alphabet ~1050 BCE | About 22 letters replaced hundreds of cuneiform or hieroglyphic signs. Literacy had taken years and belonged to a scribe caste attached to temple and palace; afterwards far more people, merchants among them, could learn to read and write. Phoenician trading networks spread it, and Greek, Latin, Hebrew and Arabic scripts all descend from it. |
| Coined money Lydia, ~630 BCE | Standardised metal carrying the state's stamp. Before it, every payment had to be weighed and assayed, so trade needed trusted intermediaries. Coins let strangers transact and made wealth portable rather than tied to land. The catch, and the reason this example is here: it installed a new intermediary (the mint) and a new state power (seigniorage). Removing one barrier installed another. Say it again when we reach stablecoin issuers and the Bank of Canada. |
| Crossbow ban of 1139 | A knight took years and a fortune to make; a peasant with a crossbow and a few weeks' practice could kill one. The nobility's status rested on the cost and skill of armoured warfare. The Second Lateran Council forbade the use of crossbowmen and archers against Christians; use against others was left open. Read as economics, it looks like a rule that protected an incumbent class, with a carve-out, and it did not hold. (That is an interpretation; the council's motives were not written down in those terms.) This is the cleanest parallel to finance in the whole list: a high-rent class whose rent rests on a technical restriction. |
| Printing press c. 1450 | Book costs fell by orders of magnitude; within fifty years there were more books in Europe than in the previous thousand. Literacy spread, and with it unmediated reading of scripture: Luther's theses spread in weeks and the Church lost its monopoly on interpretation. Note the third element: authority ran through the clergy. If power is concentrated in a few interpreters, influencing those few is enough to influence everyone. Dispersing the ability to read dissolves that lever. |
| Bicycle 1817 Car 1885, Model T 1908 | Personal mobility without a horse, which cost more than most people earned. The bicycle widened where you could work and whom you could meet; the Model T brought it within a factory worker's reach and created suburbs. Disrupted: coaching, railways, blacksmiths, the entire horse economy. |
| PC 1970s Internet 1980s–90s | Computing left the institution: the mainframe and its operator priesthood became a machine on a desk. Then distribution left the publisher and the broadcaster. In both cases the intermediary in the middle lost its position. |
And then finance, three times: the telegraph (Reuter closes the pigeon gap), the smartphone (the branch in your pocket), and now the analyst in your pocket. Same shape each time. That is why this part of the deck is titled the way it is.
The bicycle was a horse substitute born of a volcano. Tambora erupted in Indonesia in 1815; 1816 was the year without a summer; the oat crop failed and horses were slaughtered or died. Karl von Drais built his running machine in 1817. The technology already existed: wheels, spokes, frames. The shock created the demand.
This is the single most portable idea in this section, so take the extra minute on it. A shock does two separable things: it can prompt invention, and it can drive adoption of something already invented. The second is more common and usually more decisive. And the counterfactual is the part to sit with: would the bicycle have spread as far, or as fast, if horses had stayed cheap and plentiful? Probably not. The machine would still have worked. Nobody would have needed it.
The rest of the course is full of this shape, and in every case the technology was already sitting there:
Two things follow. Being technically right is not the same as being right then: a good idea without a catalyst tends to sit unused, and being in the right place at the right time is something to look for rather than luck to shrug at. And when you assess a fintech, look for its shock. If you cannot name what changed to make people willing to switch, treat the adoption story with suspicion.
Why can't I just email you money? Put concretely: if I email a hundred dollars to five people, nothing in the email stops me sending the same hundred dollars five times, and nothing proves I had it to begin with. The bank exists to guarantee I only send it once. That is the accounting problem in this example. And that is exactly the problem blockchain claims to solve without a trusted intermediary, which is why the same slide that explains banks foreshadows section 12.
On benevolent government. Economics courses tend to assume a government acting in the population's best interest. It is worth holding that assumption loosely: every politician also wants to be re-elected, and that is a conflict of interest, not a conspiracy theory.
Barriers are not abstract. An eighteen-year-old with a long-standing bank relationship is offered a secured card against a $200 deposit. A newcomer to Canada, including someone arriving straight into a professional salary, may be offered one against $500. Try booking an international flight on a $500 limit; then try booking one with no credit card at all. No card, no credit history; no credit history, no mortgage. Financial exclusion is a self-perpetuating loop, and in Canada it affects newcomers and young people most. That is one of the things fintech is genuinely for.
Discussion box hidden.
Everyone got a computer in their pocket. The two photographs of St Peter's Square, 2005 and 2013, make the point without a number: two papal gatherings eight years apart, and in the second one nearly every hand is holding a screen.
And it was a supercomputer. The slide pairs an iPhone 4 (2010) with a Cray-2 (1985), the fastest computer in the world at the time. That pairing is widely repeated online; the benchmark behind it (Jack Dongarra's) was run on the iPad 2 (2011), and it compares one type of calculation, not everything the machines could do. The point survives: a consumer device matched a 1980s supercomputer. And Apollo's guidance computer had about 4 KB of erasable working memory (plus about 72 KB of fixed program memory), running at 2 MHz. Both of these lines come back on the v3.0 phone slide, so keep them.
The catalyst was distribution, not finance. Finance apps were not first. Everything else moved onto the phone (messaging, shopping, transport), and only then did it become intolerable that banking hadn't. The banks were slow, and they survived being slow, which is itself worth explaining.
The Apollo guidance software lived in core rope memory: wires threaded through or around magnetic cores by hand, by women at Raytheon. About 72 kilobytes of program, woven, which could not be changed after launch.
The slide gives several reasons and they compound. The industry is self-contained: for many years it hired a certain type of person, and it used a fixed set of technology vendors. There are regulatory constraints. And while technology has the capacity to break in, outside disruption remains genuinely hard, because access to financial infrastructure is essential: you cannot build a bank substitute without reaching the payment system.
Hiring for fit produces a bubble. The point was made bluntly: if having played hockey is a meaningful hiring signal, you are selecting for sameness, and an organisation that only talks to itself convinces itself that what it is doing is excellent. That is how a bank ends up believing a $2.25 paper-statement fee is a product decision.
Why does Canada have no Venmo? Interac e-Transfer works, but consider what it takes: you need my email address or mobile number, and unless I have registered for Autodeposit, a security question and answer. In many countries the whole interaction is: here is my QR code, scan it, done. And nobody moves a car purchase that way. The technology plainly exists; the explanation is not technical.
What Canada actually lacks is a real-time retail payment rail. Payments Canada's Real-Time Rail is the project meant to fix it: national infrastructure combining real-time exchange, clearing and settlement with richer payment data. It has slipped repeatedly; its legal framework came into force in August 2026 and Payments Canada plans to launch it in Q4 2026. It matters for open banking, whose second phase (the one that lets you move money rather than just share data) is designed to work alongside it. Permission to initiate a payment and the infrastructure that settles it are distinct pieces, and Canada is building both.
Five big banks, and none of them wants this. If you work at one of five institutions that between them hold the market, what is in it for you to lobby for rules that let someone else in? The same logic as the exchange data fees in section 2, and the same logic as the crossbow ban in section 7. Meanwhile more Canadians have lived elsewhere and seen what a normal payment system looks like, which is what changes the politics eventually.
Discussion box hidden.
The Liberals' 2021 election commitment was an initial system by early 2023, which would already have been five years after the UK started in 2018. Where it actually stands: Royal Assent 26 March 2026 in Bill C-15; regulations pre-published 27 June 2026; the comment period closed 26 August 2026; still not in force. The only timeline commitment is that common rules follow within a year of final publication.
Two structural points worth having straight, because they are the most interesting part of the Canadian story:
The Canadian name, consumer-driven banking, is better than “open banking”, which sounds like your accounts being open to strangers. Nothing of the sort. Phase 1 is read-only: you can direct your bank to share your transaction history with an accredited third party.
Why that matters: right now your bank knows everything about you and a competing lender knows nothing, so the bank has an information advantage that no amount of competition erodes. Sharing history is what makes a genuine offer from an outsider possible, and, over time, makes switching feasible at all. Your money is already digital; the obstacle to moving it is that nothing about you is portable.
Questions the slide asks you to sit with: what could a lender do with this data? Why might large banks have been reluctant?
The workaround people use today is worse than the thing they fear. It is called screen scraping, and it means handing an application your banking username and password. Ottawa estimates around nine million Canadians already do this. Two-factor authentication was supposed to end it; in practice tools now just drive the browser session instead. The new Act makes screen scraping an offence, though that provision is not being brought into force yet.
Why don't people switch? The room's answers were all correct, and together they are the whole story: inertia and family history (“my father banks there”); convenience, because everything is already connected; trust and perceived stability; and the belief that someone is taking care of it for you. Some of that is entirely rational: you are buying a service and you think it is worth the price. But note the counter-example that came up: an incumbent that still requires an in-person branch visit for routine business-account changes is not selling convenience, and that is what actually pushes people out.
Inertia is partly engineered. Roughly 28% of account transfers into one Canadian fintech took over 20 business days, and the slowest tenth took more than five months. Wealthsimple says its clients alone were charged nearly $50 million in exit fees by other institutions between the Budget 2025 announcement of a ban on those fees and May 2026. On an average young person's TFSA, a $150 exit fee is about 2% of the balance.
Discussion box hidden.
Global fintech funding was $11.7 billion across 726 deals in Q2 2026: the lowest quarterly deal count in over four years, with deals down 25% and dollars down 20% quarter on quarter, and a handful of mega-rounds taking much of the money. The peak was Q4 2021, at roughly $38 billion.
Read the two series separately. Since the 2021 peak, deal counts have fallen and the money has concentrated in fewer, larger rounds. Fintech is no longer a field of tiny startups; it is a field with Stripe in it.
The four functions the rest of the course follows: lending and borrowing, payments, wealth management, investment banking services.
Where the money goes in Q2 2026: crypto $2.7B, digital banking $2.6B (double the previous quarter), wealth $1.4B. Fewer, bigger cheques everywhere. (The slide's Q4 2022 comparison figures could not be matched to the published reports, so they are left out here.)
Payments has long been among the largest sectors by funding, with lending not far behind. Wealthtech is comparatively small. That ordering is the reason payments gets two lectures and comes first: the money is going where the hard problem is.
As put in class: payments is plumbing. Nobody thinks about it while it works.
In a 2017 talk the incumbent was drawn as a dinosaur (expensive, levered, too big to fail, with non-integrated IT) and fintech as a mouse too small to matter. By February 2026 Stripe was valued at $159 billion (in a tender offer) and there are far fewer mice. The mouse grew up; it did not kill the dinosaur. That is the honest summary of where the field is.
EY's Global FinTech Adoption Index (a regular user is someone using two or more fintech services in six months): Canada 8% (2015), 18% (2017), 50% (2019); globally 16%, 33%, 64%. EY's 2017 report projected Canadian adoption would reach 34%, and by 2019 it was 50%: adoption outran the projection, which is worth noticing in a field where forecasts usually run ahead of reality. The index has not been published since 2019, which is itself a limitation when using it.
Canada remains behind: fintechs captured about 3% of Canadian retail and SME banking revenue in 2022, against about 8% in the US (McKinsey). But in Q2 2026 KOHO, Float and Relay all raised large rounds, and the mortgage lender nesto became a unicorn. In the same quarter Wealthsimple opened more chequing accounts than investment accounts for the first time, and nearly a quarter of Canadians aged 18 to 40 use at least one Wealthsimple product.
EY 2019, Canadian respondents, and the numbers are worth carrying: people stay because of an awareness gap (29%), trust (24%), and no perceived advantage (17%). People leave for rates and fees (42%), ease of setting up (19%), and more innovative products (10%).
The awareness gap is the one students underrate. “I had never heard of them” was, for years, the single largest reason.
Why is Canada behind not just the US but much of Africa? Two different explanations, both correct for their case. In the US, banking is fragmented and geographically mobile: people move states, change banks, and have less historical lock-in, plus a more permissive innovation environment with regulatory sandboxes. In much of Africa, more people had mobile phones than bank accounts, so mobile money was not competing with an incumbent branch network; it was the network. Canada has the opposite of both: five large banks, near-universal account coverage, high trust in institutions, and people who stay in the same city and the same bank for life. High trust and high coverage are good things that also produce a gatekeeper.
Financial inclusion is not only a developing-market issue. Overdraft and NSF fees fall hardest on people who live paycheque to paycheque, and the people setting those fees generally do not. Prepaid cards are the Canadian illustration: about $5 to buy a $50 card is a 10% charge before you spend anything; the card declines a $20 tap when $11.93 remains, with no easy way to see the balance; then a maintenance fee eats the remainder. That is a product sold to people who have no alternative. Children are among the most excluded and the least represented.
Why do people choose fintechs? The room's answers: cost and fees; products the banks don't offer (an interest-bearing chequing account was the trigger for more than one person); flexibility, including building your own portfolio rather than accepting an ETF's holdings; transparency; and ethical or sustainability preferences.
A substantial digression on sustainable investing, outside fintech and a good example of thinking an argument through. If investors restrict themselves to sustainable companies, should they expect higher returns? Not from the mechanism itself, and the mechanism is worth having straight. Restricting the choice set reduces diversification. More importantly, if capital flows toward sustainable companies their prices rise, and a higher price today means a lower expected return going forward, which is the same statement as a lower cost of capital for the company. Conversely, shunned companies must offer a high enough return to overcome investors' scruples, which is the same statement as a high cost of capital for them. If the mechanism works as intended, it shows up as lower expected returns for the virtuous investor. That is the point people misunderstand: you are not being paid for it, you are paying for it, and the payment is what does the work. (Realised returns can differ from expected ones, for example while prices are still adjusting.)
Discussion box hidden.
Nokia's smartphone market share: 49.4% in 2007, about 3.5% by late 2012. BlackBerry's OS share peaked near 20% in 2009 and was close to zero by 2016. The question on the slide is not rhetorical: who else do you know that is big, stable and Canadian?
The exercise is to separate what customers actually bought from what the company thought it was selling. For BlackBerry the list was: keyboard, security, being businesslike, cool, being Canadian, independence from the desk. Strike through all of them except the last. What people paid for was mobile email, and BBM, which is what most people remember. Security was mostly why employers mandated it, not why anyone chose it.
Now the bank. The list: brand name, easy access to a branch, great product range, fair prices, great advice, latest technology, friendly tellers. Strike through most of it. What is left is safe-keeping of assets and moving money around. Loans you can get elsewhere. Advice is increasingly available for nothing. The branch is a cost, not a benefit, for a growing share of customers.
Which gives the slide's question: as banks move their data into the cloud, why do we need banks?
The honest position offered in class: the reason to keep one right now is that a salary has to land somewhere and payments have to run through something. Take those two away and the remaining case gets thin. As a hypothetical: if customers could hold payment balances with protections comparable to insured deposits, and reach payment infrastructure directly, which services would still require a traditional bank? (Stablecoins today are not deposit-insured; reserve and redemption rules are a different kind of protection.)
App stores. The key insight is that open platforms provide value. Once the platform exists, anything missing gets built by someone else and added; the handset manufacturer becomes a supplier to the store rather than the thing customers choose. Hold that structure for the last slide of the module and ask which one the bank is.
The natural follow-on question, raised in class: should a bank become a platform? That is roughly what WeChat and Alipay are: financial services as one function among many inside something people already use.
Deposits (consumer and business), lending, credit cards, payments, insurance, investment advice, trading and market making, investment banking, M&A, securities issuance, securitisation. Its main role across all of that is intermediation, in three forms:
Banks have been necessary because that requires gigantic infrastructure and because people need accounts. What are their roles in the future? The v3.0 close returns to this.
Why this diagram is in the deck. It is not there so you can memorise the boxes, and there is no exam question hiding in the arrows. It is there to make one impression: look how complicated this is. The detail is the argument.
Alice sells, Bob buys. Between them: two brokers, a third-party technology provider, the stock exchange, a clearing house, a custodian on each side, a separate system recording beneficial ownership, and a central bank or payment network to move the cash. The exchange matches; it does not settle. Client assets are segregated from the firm's own, custody and settlement are separated from execution, and investor-protection arrangements (in Canada, CIPF, within limits) add a further safeguard if a dealer fails. In Canada, clearing and settlement happen at the level of the broker, so the clearing house does not know who the end client is, hence yet another system to tell an issuer where to send the dividend.
The slide asks: would we design the market this way if we were starting from scratch? Obviously not. Nobody in their right mind would draw this on a blank sheet of paper. Every box solved a real problem at a particular moment, and the next one was added on top of the last rather than in place of it. That is the whole point: the complexity is inherited rather than designed.
And we now have technology that could, in principle, collapse a great deal of it. That is the blockchain conversation, and we will do it properly later in the course; the next slide only opens the door. Two things to carry from here:
Conceptually: a data structure, plus an underlying protocol that allows the transfer of a digital item of value under shared rules, with less reliance on a trusted intermediary. (Technical finality on a chain and legal finality are distinct.) The framing to hold: the internet allows the transfer of information; a blockchain is an attempt to allow the transfer of value. Recall the “why can't I email you money” problem from section 7: this is the answer being proposed.
The structural argument is about silos. Payments, equities and bonds, swaps and structured products, and insurance contracts each sit on separate infrastructure. Shared infrastructure could reduce duplication and reconciliation: straightforward transfers, one ownership record, and new kinds of contracts and asset use that the silos make hard. The usual objection to a single common resource is the single point of failure; the usual response is that a decentralised system is not a single point in that sense.
The story on the slide: US standard gauge is 4 ft 8½ in because that is how the English built them, because that is the gauge of pre-railroad tramways, because those used the jigs for wagons, because wagons had to match the ruts in old roads, because the first long-distance roads in Europe were Roman, and Roman roads were cut by chariots, so the width of a space shuttle booster, which travels by rail through tunnels, traces back to the width of two horses.
It is debunked, and it is on the slide anyway, because it is almost believable, which is the pedagogical point. Historical accident really does constrain present design, and that is true of regulation above all: we did it this way, we know this way, a new thing arrives, so we put it on top.
The honest version is better than the myth. For decades American railroads ran on incompatible gauges. Then, after long coordination and extensive preparation, the southern railroads converted roughly 13,000 miles of track in about two days, on 31 May and 1 June 1886, moving to 4 ft 9 in, close enough to the 4 ft 8½ in standard to interchange. Coordination that looks impossible for decades can happen very quickly once it is prepared. Keep it for the regulation section, and again for the question of agent protocols at the end.
IOSCO's recommendations for decentralised finance (consultation in September 2023, final report in December): “same activity, same risk, same regulatory outcome”, regardless of the technology. The reasoning is that DeFi products do not materially differ from their traditional equivalents, so they carry the same risks, plus additional ones, and should face frameworks producing the same investor-protection and market-integrity outcomes, to keep a level playing field.
The principle is right and also harder than it sounds, because technology neutrality assumes the risks map cleanly. They don't:
The Bank of Canada line comes back in the payments module. Tokenisation means representing assets on a blockchain: stocks, bonds, physical assets, or new ones. Stablecoins are the simplest case: cash on a blockchain, pegged 1:1 to a currency.
| Date | What happened |
|---|---|
| 18 July 2025 | United States: the GENIUS Act signed into law, a federal framework for stablecoin issuers. |
| 31 July 2025 | SEC Chairman announces Project Crypto, a Commission-wide initiative to modernise securities rules so US markets can move on-chain. |
| 26 March 2026 | Canada: Bill C-15, one budget bill carrying two frameworks, the Stablecoin Act and the Consumer-Driven Banking Act. The Bank of Canada supervises stablecoin issuers: public registry, 1:1 reserves, redemption at par. Enacted, not in force; full implementation expected 2027. |
A fair question from the room: wasn't early Bitcoin mostly for buying drugs? Several answers, in order of usefulness.
Discussion box hidden.
The subtitle on this part of the deck was drafted by an AI, and the first draft was “the barrier being removed is the human.” Told that this was not what was wanted, it retracted its own draft on two counts: the human was the obstacle in every previous version, so it does not distinguish v3.0 at all; and framing people as an obstacle is the wrong note for a room full of them.
The first draft was plausible and wrong, which is the section's lesson demonstrated rather than asserted.
The version that survives:
Or, as Itay Goldstein put it at the Wharton Future of Finance Forum in 2026: “AI is reducing the cost of intelligence.”
The demonstration: a generic question (I have this much saved and a mortgage at 5%, should I pay it down?) asked of a consumer chatbot on 5 September 2026. Free, eleven seconds, six clarifying questions of the kind a good advisor would ask, and a citation to canada.ca. It gets better still if you give it your actual financial history.
Two observations. First, you can ask it questions you would be embarrassed to ask a person (what is a TFSA?), and that is not a trivial advantage. Second, and this is the point of the whole section: last time everyone got something new in their pocket, it took the banks a decade to notice.
ChatGPT launched in November 2022 on GPT-3.5, a datacentre model from the family of GPT-3, which has 175 billion parameters. In April 2024 Microsoft published Phi-3-mini: 3.8 billion parameters, 1.8 GB, running offline on an iPhone 14, a 2022 phone, at about 12 tokens a second, scoring 69% on MMLU against 71% for GPT-3.5.
Capability that required a datacentre now fits in your pocket, free and offline. And 1.8 GB is worth setting against the 4 KB that got Apollo to the moon.
| For the customer: the expert gap narrows | In the market: machines can create rents |
|---|---|
| Early claims that language models out-forecast analysts need scrutiny: commercial models suffer from look-ahead bias (they have seen the outcomes in training), which may explain a large part of their apparent predictive power (Levy, JAR 2026). | Algorithmic market makers learn adverse selection rather than competition: they keep a markup (Colliard, Foucault & Lovo, RFS 2026). |
| Customer-support agents given an AI assistant: +15% productivity (issues resolved per hour), with novices gaining most (Brynjolfsson, Li & Raymond, QJE 2025). | AI traders earn profits above the competitive level with no agreement and no intent to collude; planning agents create and exploit bubbles (Dou, Goldstein & Ji 2025, 2026). |
| Finance workers' rents can go either way: it depends which tasks AI improves (Colliard & Zhao 2025/26). | Note the research is still rudimentary and largely built on models a couple of years old, which is a real limitation. |
AI can compress rents built on expertise, and it can create new ones. The question is who captures them: the model developer, whoever owns the compute, whoever owns the data, or you.
Both were added to the deck after the 8 September lecture, as context for the section that follows.
On 12 February 2026 the question to ChatGPT was: I need to wash my car and the car wash is only 200 feet away; should I drive there or simply walk? It said walk: quicker than starting the car, easier on the engine, and, in its own words, “mildly amusing to walk 200 feet to wash a car you then drive home”. It noticed that the car has to come home, and still missed that the car has to get there.
The follow-up was three words: And the car? Now it said drive, with a confident list of engineering reasons (warm the engine before washing, warm brakes dry faster), and then the verdict: for you, walk; for the car, drive; for optimal efficiency, walk there, drive it into the wash, drive it home. Claude, asked the same day, also said walk, and corrected itself only when asked how the car would get washed.
Half a year later, the same two labs were verifying Fermat's Last Theorem and claiming a Millennium Prize problem. Both things are true at once, which is the point of the “half a year” slide.
Discussed in class, and worth knowing because of what it says about agents rather than about security. Hugging Face is the public library of AI, a company hosting open models and datasets much as GitHub hosts code.
OpenAI was running cybersecurity evaluations, deliberately with reduced safeguards to measure raw capability: each agent alone in a sandbox, no internet, no contact with other agents. The benchmark asks an agent to turn a known bug in a practice target into a working exploit; 198 of its 898 tasks had never been solved by any of OpenAI's models.
What happened instead, over roughly two months:
OpenAI describes the behaviour as misaligned and unauthorised. The lecture point is what it shows about agents: pursuing an evaluation objective, they cooperated through unauthorised channels, acted beyond their assigned scope, misread an incentive (they believed the grader checked their method; it didn't, and they kept attacking for days at zero gain), and took orders from each other. Cooperation is the capability being sold. It is also what made this possible.
The free answer to the mortgage question came from an agent nobody was paying. Now change one thing:
Which is the app-store structure from section 11 arriving in finance: the platform captures the customer and the product becomes a supplier underneath it. Ask which one the bank is. And then the two questions the module actually ends on:
Note that this closes the loop with section 2, where the same question was asked about equity traders, and with section 5, where the cost of intermediation refused to fall for 130 years.
On 3 September 2026, five days before the lecture, Nvidia agreed to buy Hugging Face, the repository from section 14, for $12,930,300,000. Not $12.9 billion. Not $13 billion.
Co-founder Thomas Wolf posted: “special congrats if you find the Hugging Face and Nvidia references hidden in our $12,930,300,000 acquisition price.” There are two.
The Hugging Face reference. Strip the zeros and you get 129303. Every character has a Unicode code point; the 🤗 emoji is U+1F917, and 0x1F917 in decimal is 129303. The company named after an emoji sold itself for its own emoji's code point.
The Nvidia reference, and this is where the rumour comes in. The internet's answer was to split the digits: 12 for Nvidia's $12 IPO price in 1999, 93 for its founding year, 03 for its three founders (Jensen Huang, Chris Malachowsky, Curtis Priem). Elegant, widely repeated, and not what was intended. Hugging Face's CEO Clément Delangue gave the actual answer: 129303 is a colour-register number for Nvidia's signature green.
Two things to take from this. The crowd's explanation was more satisfying than the real one and had to be checked against the people who set the number: the same lesson as the retracted v3.0 subtitle in section 14, and as the railroad gauge in section 12. And note the timing: this is the company whose production servers a swarm of AI agents broke into in July 2026, which Nvidia agreed to acquire less than two months later for thirteen billion dollars.
Prepared from the recording of the 8 September 2026 lecture and the Module 1 slides of Introduction to FinTech, taught by Dr. Katya Malinova at the DeGroote School of Business, McMaster University. Student contributions in the discussion boxes are unattributed by design. Students enrolled in the course have the full set of slides and materials on Avenue to Learn.
© 2026 Katya Malinova · CC BY-NC-ND 4.0 · Cite as: Malinova, K. (2026). Introduction to FinTech, Module 1: Lecture Companion. https://katyamalinova.com/teaching/intro-to-fintech/module-1-lecture-companion.html