Revenue Model Examples: Where One Dollar Actually Goes

A company's revenue model fits into one sentence: who pays it, for what, and what has to happen for revenue to grow. The fastest way to write that sentence is to follow one dollar from the customer's wallet to the last line of the income statement. The same dollar leaves a very different residue at Visa, at Spotify and at a game studio, and the size of that residue decides everything you will calculate afterwards.

If you cannot write the sentence, you know a logo, not an engine.

Start with a revenue model you can argue with#

Use a shape that is easy to test: the company sells X to customer Y and grows when Z happens. "Adobe sells subscription software to businesses and creators, and grows when it adds customers, sells more inside existing accounts and keeps churn down." The analysis comes later. What you have so far is a map to navigate by, and a hypothesis the filing can knock down.

Trouble is, the brain substitutes a familiar logo for the answer. You use the app, you pass the sign on the way to work, you recognise the founder, and that feels like understanding. An engine is more specific than a brand: what exactly the customer pays for, what triggers the payment, which costs rise with sales, and which unassuming corner of the company quietly does most of the earning.

The usual guides sort models into drawers: subscription, commission, advertising, one-off sales, freemium. The drawers are honest and genuinely useful, because a business billed per transaction behaves differently from one billed per month of access. Two businesses from the same drawer tend to share a revenue rhythm and the same questions about growth, so the label saves real time at the start.

The label just never answers the one question an investor has: how much of the customer's money reaches the company at all.

"Online retail" describes a firm that runs a cloud business, an advertising business, subscriptions and fees charged to third-party sellers alongside the store. "Phone maker" describes a firm where services and app-store commissions produce recurring revenue from devices bought years ago. A label puts you at the start of the route and stops being useful there.

So before you drive any further, answer five questions about one company:

1. What unit does the customer pay for: an item, a month of access, a transaction, a licence?
2. What triggers the payment: an order, usage, or a commission collected by an intermediary?
3. Who actually sends the money, and is that the same person who uses the product?
4. Which costs rise together with sales?
5. Which costs had to be paid before the first customer showed up?

At Visa the dollar pays a toll#

Who actually pays Visa? Most people look at the card in their wallet and assume they do. They do not. That dollar starts as a cup of coffee, passes through the merchant, the merchant's bank and the bank that issued the card, and then reaches Visa as a toll for crossing its rails. Credit comes from the issuing bank, and so does the risk if the cardholder never pays it back.

The size of that toll explains the rest. In fiscal 2025 Visa reported $40.0 billion of net revenue against $17 trillion of total payments and cash volume, which means roughly 24 cents for every $100 that crosses the network. The gate is narrow, and that is precisely why it works.

That single fraction sets up every other question. Visa grows with the number and size of transactions, while the credit risk stays with the bank that issued the card. A recession reaches Visa through a smaller basket at the till, and that is where it stops. A bank with similar revenue reads the opposite way: there a downturn arrives through loan quality and can eat a year of profit in two quarters. Whether that same dollar still reaches the company in five years is settled by an advantage a rival cannot rebuild, not by the shape of the model.

Notice what is missing from the description as well. No inventory, no factories, no receivables from retail customers, because Visa sells you nothing directly. An engine that takes a fraction of somebody else's turnover needs very little capital of its own to handle that turnover, and that is where returns come from that the size of the fee alone could never explain.

At Spotify almost all of the dollar leaves#

A subscription looks like the simplest model on earth until you ask where the money goes next. Before Spotify keeps anything, it pays rights holders for every stream, at rates set in contracts the company does not write on its own.

The size of that first stop shows up in the filing. In 2025, on revenue of €17,186 million, cost of revenue came to €11,690 million, which means 68 cents of every euro left the building, against roughly 70 cents the year before. What remains is 32 cents to cover technology, marketing, people and profit.

So the Spotify engine reads like this: collect subscriptions, hand most of them straight to the rights holders, and earn on the remainder plus advertising on the free tier. Two percentage points of improvement at that first stop matter more to the result than a solid increase in listeners, and that is the question I start with every time I open this company.

What follows for reading the results? Subscriber growth stops being good news on its own, because every new subscriber brings their share of the royalty along with them. Improving the result requires a higher subscription price, cheaper access to the catalogue, or a bigger share of revenue that never passes through that first stop. Which is why a price increase in this model is a fundamental event.

At a game studio the dollar arrives in waves#

A player buys a game, and the money does not reach the publisher whole: the store where the purchase happened takes its commission first. Whatever is left has to cover years of production that finished before anyone paid a cent.

Say a studio spends $400 million over four years on one title, and the store takes 30% of every sale. At $70 a copy the studio keeps $49, so the costs are only covered somewhere past eight million copies sold. From that point on the title earns. The numbers are invented and round, but the shape is real: the spending is spread evenly over years, the revenue turns up in a few weeks.

That is why a quarter without a release tells you almost nothing about such a publisher, and a quarter with one tells you nothing about next year. This engine gets judged over a whole release cycle, and here a year-on-year comparison is simply misleading.

The same mechanism changes what the balance sheet means. Production costs for a game that has not shipped sit in assets as capitalised software development and only become an expense once the title starts selling. A rising balance there with no releases means the company is feeding a machine that has not reached the income statement yet. A cancelled project ends in a write-down, and that is one of the few places a publisher states plainly that four years of work went nowhere.

Revenue is not profit#

Those three drives lead to the same conclusion: what matters is not how much money passes through a company but how much stays. The cleanest example sits in Amazon's filing. In 2025 the cloud accounted for $128.7 billion of the group's $716.9 billion of net sales, or 18%, and at the same time for $45.6 billion of $80.0 billion in operating income, or 57%.

Everyone sees the store. The cloud does the earning.

The same filing holds a second quiet earner. Advertising produced $68.6 billion of revenue in 2025 against $49.6 billion from subscriptions, close to 40% more than the whole subscription business brings in. Nobody calls Amazon an advertising company, and that is exactly the distance between a sign and an engine.

The trap works in reverse too. A retail chain with a thin net margin looks like an industrial accident in a screener and can still be an excellent business, because it turns its inventory over more than a dozen times a year. Margin multiplied by turnover gives you the return on capital, so a thin margin at high speed can beat a fat margin at low speed. That whole route, from what the customer pays for to what is left at the end, fits on one page about one retail chain.

So before you judge revenue growth, check which segments it came from and which of them actually adds to the result. Often the fastest-growing segment is the lowest-margin one, and the press release leads on the growth rate while the margin of that segment goes unmentioned.

Who uses it and who pays for it#

The dollar's route has one more fork that is easy to forget: the user and the payer are frequently not the same person. At Google and Meta, users supply the attention and advertisers settle the bill. A corporate gym benefit is bought by your employer, not by you. In enterprise software an employee uses the product, a manager approves it, procurement argues about price, and the finance director signs the transfer.

That has hard consequences, because a company builds its product for whoever pays. If advertisers pay, the product gets tuned for time spent in the app. If the employer pays, what matters is the share of staff who use the benefit at all. Find the payer and you have found the direction the product will drift in.

Separating the roles also explains things that look absurd from outside. A system wired into a company's warehouse, invoicing and workflow stops being software and becomes the fixed installation the whole operation runs on. When it stalls at a bad moment, contractual penalties start running. The licence price fades into the background, because switching vendors is measured in years, which is how such a supplier raises prices annually without losing customers.

There is also a test that costs one question. What happens if the customer does nothing? When doing nothing is cheap and painless, the company has to earn its keep again every year. When it means downtime, lost revenue or a letter from the regulator, the promise is strong, and that is usually where a durable economic moat comes from.

Which costs rise with sales#

The model sentence is incomplete until you say what happens to costs when sales grow. Say two companies each have $100 million of revenue and both grow 20% this year. The first sells software: the cost of writing it was paid earlier, each extra subscriber costs pennies, so nearly the whole $20 million of new revenue reaches the bottom line. The second mines copper: every extra tonne needs energy, people and equipment, so the same $20 million leaves most of itself behind on the way.

Same growth, two different results. Without separating fixed from variable costs, a revenue growth rate is a number with no meaning attached.

There is a separate category for costs that had to be paid before the first customer existed. A semiconductor fabrication plant stands ready and paid for before anyone places an order, so a question about its utilisation is a question about the entire result. An aircraft engine can sell at a margin close to the cost of building it, because the real money sits in the decades of servicing it drags behind. Two identical-looking "engine sales" lines then describe two completely different businesses.

The second cost question is: who sets the price? If the company can raise it a few per cent and keep its customers, the margin belongs to it. If the price is set by a commodity index, an exchange rate or the largest competitor's next move, the margin is borrowed and will have to be handed back eventually. That question has its own pricing-power test, and it is worth asking before you believe any margin forecast.

Where this method breaks down#

Tracing one dollar shows you the state of the route on the day you drove it. It has three weak points.

First, one dollar is an average. A company with three segments has three different dollars, and the average describes none of them faithfully. If the mix between segments shifts, the result changes even when nothing happened inside any of them. So the trace has to be repeated separately for every segment above roughly a tenth of revenue.

Second, engines drift, sometimes deliberately. A software publisher moving its customers off boxed licences and onto subscriptions turns one-off revenue into recurring revenue, and a few years later the model sentence reads completely differently. A trace run three years ago would then describe a company that no longer exists, and management announcing a "model transformation" rarely mentions which stream is shrinking. You have to ask that one yourself.

Third, revenue recognition can drift away from cash. The money arrives today while the revenue is spread across twelve months of a subscription, so deferred revenue sometimes tells you more than the top line does. A marketplace books the commission in revenue; a retailer selling on its own account books the value of the whole basket. Setting those two top lines side by side compares two different things wearing the same name.

I do not know how much of Spotify's current margin survives the next round of negotiations with the rights holders. I do know where to check and when: in the next annual filing, on the same cost-of-revenue line.

The verb that gives away a missing model#

When a description says a company "monetizes users", "leverages data" or "captures value across the ecosystem", you are looking at a sentence from a slide deck. The rule is short: if the verb is vague, you do not have the model yet. Replace it with something concrete, namely who sends the payment, for which unit and how often, and either the sentence closes or it turns out there is nothing to close it with.

I think this is the cheapest filter in fundamental research. It costs one sentence and it removes most of the companies about which the only known fact is that they are well known.

In Taufolio this part of the work sits in the Full report, in the business-model and segment sections, where every claim carries a link back to the document it came from. Read them, then write your own sentence in your own words and underline the part you are guessing at. If you would rather see how such a report is put together first, start with the product page and take one company through it as a trial run.