Chapter 2 — A Seemingly Perfect Compounding Machine

An arithmetic problem that mesmerized investors

Start with a piece of arithmetic.

Suppose a software company has $100 million in annual recurring revenue (ARR) this year. Now suppose that starting next year it adds not a single new customer — the sales team is disbanded, the marketing budget is zeroed out, and it spends nothing to acquire anyone new.

In most businesses, this would mean the end of growth. A restaurant with no new diners, a carmaker selling no new cars — revenue only falls. But in the world of software, things can be entirely different. If this company’s existing customers not only barely churn but also naturally buy a bit more each year — add a few more seats, use more features, consume more credits — then even with no new customers at all, its revenue still grows.

We measure this phenomenon with a number called net revenue retention (NRR): the revenue a cohort of existing customers contributes a year later, divided by the revenue they contributed a year earlier. If that ratio exceeds 100%, it means “the existing customer base is expanding on net” — the portion they expanded outweighs the portion that churned.

Now suppose this company’s NRR is 120%. Then, adding no new customers at all: in year one, the $100 million existing base becomes $120 million; in year two, that $120 million times 120% becomes $144 million; in year three, $173 million… revenue grows itself at 20% a year, and all of it without a single new customer. If, on top of this self-growing machine, the sales team keeps stacking new customers, total revenue grows even more steeply.

This is the arithmetic problem that mesmerized all of Wall Street. It depicts a business close to perfect: revenue that grows on its own. A traditional business has to win back every dollar of revenue anew, year after year; a SaaS company with NRR above 100% seems to possess a self-powered compounding machine — past success carries forward automatically into the future, and new effort accumulates on top of an ever-rising base.

This chapter does two things. In the first half, we take the compounding machine apart to see exactly what parts it’s made of and what makes it run. In the second half, we throw cold water on it: the machine looks perfect, but it neither runs automatically nor is immune to running in reverse — and telling those two apart is the real difficulty of SaaS investing.

The four parts of the compounding machine

A SaaS compounding machine runs on four interlocking parts.

Part one: recurring revenue — the “stock” character of revenue. As Chapter 1 explained, subscription turns revenue from “a series of rains that must fall anew” into “a reservoir that keeps filling.” This is the base of compounding: so long as customers don’t churn, last year’s revenue automatically recurs this year. Without this stock base, there is no compounding at all — a business that resets to zero every year cannot compound.

Part two: prepaid cash — customers financing your growth interest-free. SaaS customers typically prepay their subscription by the year (sometimes for multiple years). This means the company often receives cash first and recognizes revenue slowly over the coming year. This early-arriving cash forms an extremely favorable “negative working capital” structure: customers are effectively providing interest-free financing, with no principal to repay, for the company’s operations and expansion. A healthily growing SaaS company often has a large stock of deferred revenue on its books, and its operating cash flow can run meaningfully ahead of its reported profit. This is a hidden, enormous advantage of SaaS over traditional businesses — but it also hides a trap: as we’ll see in Chapter 7, the “earliness” of this cash flow is sometimes used to flatter true profitability.

Part three: low marginal cost — the physical basis of high gross margins. Cloud multi-tenant architecture makes the cost of “serving one more customer” extremely low. The same system serving a thousand customers versus a thousand and one adds negligible cost. That is why good SaaS companies can have gross margins of 70% or 80%. High margins mean that of every incremental dollar of revenue, the great majority stays to cover expenses, form profit, or be reinvested in growth. The revenue the compounding machine “throws off” with each turn is high-grade ore — at least until AI inference cost enters at scale, an important exception that is the subject of Chapter 16.

Part four: the expansion engine — getting existing customers to use more. The first three parts only ensure revenue can be “held”; what truly lets it “grow large” is the fourth: expansion. A SaaS company gets existing customers to pay more, usually along four paths — adding seats (more employees using it), adding usage (used more often, more data), cross-selling (bought product A, now buy B), and price increases. These four paths stacked together are the power source that drives NRR above 100%. And systematizing this “enter small, then keep expanding” playbook is the most important growth paradigm in the SaaS world: land and expand.

【Key concept】Land and expand, and net revenue retention (NRR)

Land and expand: a growth strategy common in SaaS. The vendor first gets the customer “landed” and using the product at a smaller scale and lower barrier (say, selling to one department, a few dozen seats), then — after the customer experiences the value and grows dependent — gradually expands to more departments, more seats, more product lines. Its essence: the first contract does not chase size, it chases “getting in the door”; the real value lies in the expansion after entry.

Net revenue retention (NRR): the revenue a cohort of existing customers contributes a year later ÷ the revenue they contributed a year earlier, including expansion, including churn, but excluding any new customers. NRR > 100% means the existing base is expanding on net; NRR < 100% means it is shrinking on net. It is one of the core metrics for the power of the “compounding machine.” (The strict definition, calculation basis, and cross-company differences of NRR are unpacked in Chapter 3; here we’re just building intuition.)

Note: NRR has no industry-wide standard definition; companies use different bases and it cannot be compared across companies naively — a point that matters greatly, emphasized in Chapter 3.

Why expansion compounds: a numerical illustration

To drive “compounding” home, we use a purely fictional, illustration-only numerical example. Note that this is a model to explain the mechanism; it represents no real company.

Suppose a company whose initial existing customer base contributes $100 of revenue. We push three scenarios forward three years:

Scenario	NRR	Year 1	Year 2	Year 3
A. Strong expansion	120%	120	144	173
B. Just holding	100%	100	100	100
C. Leaking	90%	90	81	73

The difference across the three is only NRR moving from 120% to 90% — seemingly just thirty percentage points. But after three years, A’s existing-customer revenue has grown to 173 while C’s has shrunk to 73 — the same set of existing customers, one up 70% and one down nearly 30%, a gap of more than double. And this is before counting new customers. If company A adds a fresh batch of customers each year on top of the base of 173, its total revenue traces a dazzling growth curve; company C, meanwhile, must desperately use new customers to fill the hole left by churning existing ones, running out of breath while merely marking time.

This example reveals two brutal properties of SaaS compounding:

First, compounding is extremely sensitive to NRR. A few percentage points of difference in NRR, magnified by time, becomes a world of difference in revenue scale. This is also why the market pays a premium for high-NRR companies — it is buying not today’s revenue but the machine’s future capacity to grow itself.

Second, compounding is a double-edged sword. When NRR is above 100%, time is your friend and revenue accumulates on its own; but once NRR drops below 100%, that same machine begins to run in reverse — time becomes the enemy, the existing base shrinks itself like a leaking bucket, and the company must use more and more new customers just to sustain surface-level growth. The compounding machine does not automatically turn in the good direction; which way it turns depends on which side of 100% NRR sits.

Company field notes: when “expansion” is written into the physics of the product

The example above is fictional. Now a real one, to grasp how strong the “expansion engine” can be in practice.

The data-cloud company Snowflake uses a usage-based billing model — customers pay for the compute and storage they consume, not a fixed number of seats. The beauty of this model: when a customer’s business grows, its data multiplies, and its analytics needs rise, its consumption on Snowflake grows naturally, almost automatically, with no need for a Snowflake salesperson to negotiate a new contract. Expansion is written into the physics of the product.

This structural expansion is clearly visible in Snowflake’s revenue curve. Verified against SEC filings, Snowflake’s product revenue grew from about $3.63 billion in fiscal 2025 (ended January 31, 2025) to about $4.68 billion in fiscal 2026 (ended January 31, 2026), up about 29% year over year; in the most recent quarter before this book’s cutoff (ended April 30, 2026), quarterly revenue was about $1.39 billion, up about 34% from about $1.04 billion in the year-ago quarter.^[1] Sustaining roughly 30% growth at that revenue scale is exactly the double stacking of “existing customers’ consumption growing naturally + new customers continually landing.” This is what the compounding machine looks like when it runs smoothly.

But notice — immediately — the other side of this coin. Usage-based billing makes expansion automatic, and it makes contraction automatic too. When the macro turns cold and customers start cutting spending, they need not cancel any contract; they only have to use a little less, and Snowflake’s revenue feels the pressure at once. Seat-based revenue is relatively rigid (seat counts usually don’t change within a contract term), while usage revenue is elastic and procyclical — it surprises you on the upside in booms and on the downside in contractions. This is why, for the same “high NRR,” a usage-based company’s NRR often swings more violently than a seat-based one’s. Where the expansion engine is strongest, the risk of running in reverse is greatest too. This double-sidedness of usage billing is the deep subject of Chapters 10 and 17; here we need only remember: no single part comes with only upside.

Why “seemingly perfect” does not equal “actually perfect”

Now, the cold-water moment. We took the compounding machine apart very neatly above, but an honest investor must press: what guarantees this machine will actually run?

The answer: it doesn’t necessarily. Between “seemingly perfect” and “actually perfect” lie at least three hurdles.

Hurdle one: the truth about retention hides in the gap between “gross” and “net.” NRR is a net concept — it mixes the two opposing forces of “churn” and “expansion” and shows you only the final resultant. A company with 120% NRR could be “barely any churn + moderate expansion,” or “heavy churn + a few large customers expanding wildly.” The latter looks just as glossy but is in fact fragile: its growth depends heavily on a few large customers keeping up the pace, and once those customers peak or leave, the prosperity reverses fast. To see through this, you cannot look only at net revenue retention — you must also look at gross retention (the part that counts only churn, not expansion). Gross retention tells you how fast the bucket leaks; net retention tells you whether you’re pouring water in fast enough. The two must be read together — this is the key discernment Chapter 3 builds.

Hurdle two: the fuel for expansion may come from a one-time, unsustainable source like price increases. The “expansion” inside NRR may come from customers’ real, sustainable usage growth (healthy), or mainly from the vendor’s price increases (suspect). Price increases can prettily lift NRR in the short run, but they have a ceiling: there’s a limit to how much a customer will tolerate, and retention propped up by price increases is essentially borrowing from future pricing room. Distinguishing whether a company’s expansion comes from “customers using more” or “the vendor charging more” is a key part of judging its growth quality, which we dissect specifically in Chapter 5, “Where Does Revenue Growth Actually Come From.”

Hurdle three: getting in the door has a cost, and that cost is often masked by the halo of growth. The “land” in land and expand is not free. Acquiring a customer requires enormous up-front sales and marketing outlay, while the revenue that customer brings is recovered only slowly over many future years. This creates a counterintuitive phenomenon: the faster a SaaS company grows, the more it may lose on paper — because it is using today’s certain, one-time acquisition loss to buy tomorrow’s uncertain recurring revenue, which must be realized through retention. Whether that trade is worth it depends entirely on how long customers stay and how much they expand. With excellent retention, it’s the best business in the world; with poor retention, it’s a shredder that keeps devouring cash. Distinguishing the two requires a whole toolkit of unit economics — customer acquisition cost, payback period, lifetime value — which is the core content of Chapters 3 and 4.

【Mistakes investors make】Extrapolating today’s high NRR linearly into the future

A 130% NRR easily tempts investors into linear extrapolation: “It grows 30% a year on existing customers alone, and will keep doing so.” This is dangerous.

There are three reasons. First, NRR naturally recedes with scale. When a company is still small and rapidly rolling out customers, the room to expand is vast; when it grows large and its big customers’ seats and usage gradually saturate, the slope of expansion inevitably flattens and NRR reverts toward 100%. Extrapolating maturity from the NRR of a high-growth phase almost certainly overstates. Second, NRR is procyclical, especially for usage-based companies — overstated in good times, understated in contractions. Third, the basis of NRR can be “prettified”: adjusting how the customer cohort is defined or how the time window is chosen can make the number look better. Seeing a beautiful NRR, the right reaction is not to extrapolate but to press: what is it made of (churn vs. expansion)? Does the expansion come from usage or price increases? Where in the lifecycle is it? Chapter 3 puts all the tools for these questions in your hands.

Reverse view: maybe there’s no “compounding machine” at all

To put it bluntly: there is also a sharp counterview that “SaaS is a compounding machine” is itself a marketing narrative inflated by a bull market and low rates together, not a law of business.

This view would point out that SaaS companies that can truly sustain NRR above 100%, retain excellently, and grow without burning cash are a tiny minority. The vast majority of SaaS companies — especially those serving small businesses, with heavily commoditized products and no switching costs — do not retain well, and their growth depends almost entirely on continuous marketing spend. For these companies, “recurring revenue” is more a form of billing than a compounding machine — stop burning cash and growth stalls, even reverses. Generously pinning the “compounding machine” quality — which belongs to a few exceptional companies — onto the entire industry is precisely the cognitive root of the 2021 valuation bubble.

This book partly agrees with this counterview. Indeed, the compounding machine is the privilege of a minority, not a trait of the industry; indeed, many companies called “SaaS” do not truly compound. But this is precisely not a reason to give up on research — it is the whole point of research: if compounding machines are so rare, then whether one can tell the real compounding machines apart from the pile of companies that merely “bill like SaaS” becomes the highest-return skill in SaaS investing. Every chapter ahead adds a brick to that discernment.

【Reading the financials】To judge a compounding machine, find these three signals first

To quickly judge whether a company is a true compounding machine, look first at these three places in the financials and investor materials:

Net revenue retention (NRR/NDR) and gross retention: find both. NRR reflects net expansion capacity; gross retention reflects the leak rate. A company that discloses a pretty NRR but dodges gross retention deserves extra wariness. Note the company’s custom basis for NRR.

Growth in deferred revenue and RPO: these are “signed but not yet recognized” future revenue, and their growth rate often reveals the temperature of the compounding machine earlier than current-period revenue.

The relationship between sales-and-marketing spend and net new revenue: how much S&M did the company spend, and how much net new revenue did it buy? This ratio (the “magic number” we’ll cover in Chapter 3) reveals whether growth is “efficient self-turning” or “propped up by burning cash.”

How to compute these three and cross-check them, Chapter 3, “The Economic Engine of a SaaS Company,” demonstrates step by step with a complete case built on a fictional company.

Chapter summary

SaaS is called a “compounding machine” because of one piece of arithmetic: when net revenue retention (NRR) is above 100%, a company’s revenue grows itself even with no new customers added; new customers then stack on top of an ever-rising base.
The machine is made of four interlocking parts: recurring revenue (the stock base of revenue), prepaid cash (interest-free financing and negative working capital from customers), low marginal cost (the physical basis of high gross margins), and the expansion engine (land and expand, getting existing customers to use more).
Compounding is extremely sensitive to NRR, and is a double-edged sword: above 100%, time is your friend and revenue accumulates; below 100%, the machine runs in reverse and the existing base shrinks itself. The stronger the expansion engine (as with usage billing), the greater the procyclical reversal risk.
“Seemingly perfect” does not equal “actually perfect” — three hurdles lie between: the truth about retention hides in the gap between gross and net retention; the fuel for expansion may be unsustainable price increases; and “getting in the door” has a cost — acquisition cost up front, revenue deferred — which leaves high-growth companies deep in the red on paper.
The most crucial judgment: true compounding machines are a minority. Telling them apart from the pile of companies that merely “bill like SaaS” is the highest-return skill in SaaS investing, and the goal of this book’s entire framework.

Notes and sources

[1] Snowflake product revenue: about $3.63 billion in fiscal 2025 (period end 2025-01-31), about $4.68 billion in fiscal 2026 (period end 2026-01-31); most recent quarter (period end 2026-04-30) about $1.39 billion, year-ago quarter (period end 2025-04-30) about $1.04 billion. Year-over-year growth rates are the author’s calculation from the above (about 29% and about 34%). Source: Snowflake Inc. 10-K/10-Q (SEC EDGAR, CIK 1640147, concept tag RevenueFromContractWithCustomerExcludingAssessedTax), retrieval date 2026-07-15.
The “three NRR scenarios” numerical example in this chapter is a purely fictional illustration model, representing no real company, as explicitly noted in the body.
For the strict definitions and basis differences of NRR, gross retention, and land and expand, see Chapter 3; this chapter only builds intuition.

Chapter fact-check list

Statement	Status	Handling
Snowflake uses usage-based billing	✅ Public fact (10-K business-model disclosure)	Kept; detailed in Chapter 10
Snowflake FY2025/FY2026 product revenue $3.63B/$4.68B	✅ Verified (EDGAR, 2026-07-15)	Kept
Snowflake latest quarter $1.39B, year-ago quarter $1.04B	✅ Verified (EDGAR, 2026-07-15)	Kept
YoY growth ~29%, ~34%	✅ Author’s calculation from verified data	Marked “author’s calculation”
NRR > 100% allows growth with no new customers (arithmetic)	✅ Definitional inference	Kept
The “three scenarios” compounding example	✅ Explicitly marked as fictional illustration	Kept
NRR has no standard definition, bases differ	✅ Industry common knowledge	Detailed in Chapter 3

Connections to adjacent chapters

Following Chapter 1: Chapter 1 explained how the compounding machine “was built” (the three-layer restructuring from license to subscription); this chapter explains how it “runs” (four parts + NRR compounding), and formally delivers on the foreshadowing Chapter 1 planted — “recurring revenue does not equal high-quality revenue.”
Leading to Chapter 3: this chapter repeatedly mentions NRR, gross retention, acquisition cost, payback period, magic number, but gives only intuition, not strict definitions. Chapter 3, “The Economic Engine of a SaaS Company,” uses a fictional company running through the whole chapter to explain, in one pass, the definitions, calculations, and mutual reconciliation of the entire metric set — ARR, MRR, RPO, gross and net retention, CAC, LTV, CAC payback, sales efficiency — upgrading this chapter’s “intuition” into “tools.”
Echoing the main thread: the “compounding is a minority privilege that must be discerned” established here is the first full unfolding of one of the book’s main threads, “revenue quality > revenue growth”; the double-sidedness of usage billing reserves an interface for Chapters 10 and 17.