ARR is not revenue
ARR is a company-defined operating metric, not an accounting one, and the gap between it and recognised revenue is where most confident-sounding design claims quietly stop being true.
You can now point at an income statement and name the line your work reaches. Nobody in the room talks that way. They talk in ARR, gross and net retention, CAC payback and the Rule of 40 — a layer of operating metrics sitting on top of the statement, each one defined by the company reporting it rather than by any standard. This module defines the five precisely enough to argue with. It starts with the one that gets said most and understood least.
Three numbers, one customer, one day
A customer signs. That single event produces at least three numbers, and they are not versions of each other.
Bookings is the contract. a16z defines it as “the value of a contract… a contractual obligation on the part of the customer to pay the company”. It is the largest of the three and the least constrained: a promise, recorded when the promise is made.
ARR is a run rate. The same page scopes it: “ARR (annual recurring revenue) is a measure of revenue components that are recurring in nature. It should exclude one-time (non-recurring) fees and professional service fees”. Read that carefully. It is a scope, not an equation. It tells you what to leave out. It does not tell you how to annualise what is left, and that gap matters more than it looks.
Revenue is the accounting one. a16z again: revenue is “recognized when the service is actually provided or ratably over the life of the subscription agreement,” and “how and when revenue is recognized is governed by GAAP”. This is the only one of the three with a rulebook and a regulator behind it. It is also the smallest and the slowest.
Name the interested party before going further. Andreessen Horowitz is a venture firm, and these are the definitions by which the companies it funds get judged. That is not a reason to discard them — they are the clearest free statement of the distinctions available — but a firm publishing the scoreboard its own portfolio is scored on is not a neutral party, and it is the same conflict you would name instantly if a vendor published a benchmark showing its category winning.
The illustration, with its assumptions stated
No source hands this course a worked example, so the following is the course’s own construction, built to make the three numbers visible at once. The figures are invented. The assumption doing the work is that the subscription is recognised ratably across the term, which is one of the two treatments a16z names.
Contract signed 1 March, 24-month term
Subscription $10,000 / month
Implementation, one time $40,000
Bookings, March $280,000 the whole contract
ARR contribution, from 1 March $120,000 recurring only
Revenue recognised in March $10,000 plus whatever share
of implementation
was deliveredThe last line is deliberately not a number. When implementation revenue gets recognised depends on what was delivered and what the contract says, and this course is not going to invent a treatment for it. The point survives without it: the same signature on the same day produces $280,000, $120,000 and about $10,000, and the largest of those is twenty-eight times the smallest.
Notice too that the implementation fee vanishes from the middle line entirely. That is a16z’s scope rule doing its job, and it is the first place a design claim can quietly break: services work you made faster or unnecessary shows up in revenue and in cost of revenue, and never in ARR at all.
Why you cannot look up the definition
Look at the asymmetry in a16z’s own two sentences. For revenue it points straight at a rulebook: recognition “is governed by GAAP.” For ARR it points at nothing, and offers a scope instead. This course found no GAAP or SEC definition of ARR to point at either. That is not a scandal. It is a category difference: ARR is an operating metric a company defines for itself, in the same way a company defines its own active-user count.
The consequence is that two companies can both report ARR honestly and mean different arithmetic. a16z Growth says as much about its own dataset, describing why it had to normalise before comparing anything: “Different startups use different methods to calculate and report metrics, so we standardized our metrics calculations”. A firm with direct visibility into many companies’ reporting found the raw numbers not comparable. That is stronger evidence than any warning a course could write.
So the first question in any room, before you promise to move anything, is how this company computes the number. Not because you suspect anybody. Because the phrase has no fixed referent and you are about to attach your credibility to it.
What this does to a design claim
Three consequences, in the order they will bite you.
- ARR moves on renewal dates, not on ship dates. A change that makes a customer more likely to stay does nothing to the run rate until that customer’s contract comes up and does not leave. If your hold-period story needs an effect this quarter and your mechanism is a renewal twelve months out, the two do not line up and somebody will say so.
- Work that removes services revenue can look like a loss. Automating an implementation step reduces professional services revenue, which is real revenue on the statement, while touching ARR not at all. That can still be the right trade. It is a much better conversation when you name it first.
- “Protected ARR” is a counterfactual. It claims a customer who did not leave would have left. That is not a measurement, it is a comparison against a world nobody observed. There are honest ways to argue it. Stating it as a figure is not one of them.
Snowflake’s statement from the previous module shows the first distinction structurally: product revenue and professional services are reported on separate lines, $4,472,317 thousand against $211,629 thousand for the twelve months ended 31 January 2026, because they behave differently. Snowflake reports product revenue rather than ARR, so this is the shape of the distinction rather than an example of the metric.
Check your recall
Answer from memory — no scrolling back.
Retrieval check
Someone hands you a deck saying “ARR grew 30% this year.” Name two questions you would ask before repeating that number in a case of your own.
Check your answer
The two that do the most work: how is ARR annualised here — a month multiplied by twelve, a contracted run rate at a date, something else — and what is in it that a16z would exclude, meaning one-time fees, implementation, professional services, anything usage-based that somebody chose to treat as recurring.
Neither question is an accusation, and both are the kind of thing a finance lead answers in one sentence without offence. Asking them costs you nothing. Not asking them, and then building a case on top of the answer you assumed, costs you the case.
Hands on
Put the ARR question in writing on all three cases
Done when: Each entry in VALUE-CASES.md carries one line naming whether it plausibly touches ARR, one line naming whether it touches recognised revenue, and the definitional question you would ask before either claim.
- For each case — the review gate, the advisory recommendation surface, the client work — write Touches ARR: followed by yes, no, or not without a renewal, and a clause saying through what.
- Write Touches recognised revenue: separately. These two answers differ more often than people expect, and a case that answers them identically has usually not distinguished them.
- Add Would ask: and one question about how this company computes ARR. One. A list of five reads as stalling; a single precise question reads as someone who has done this before.
- Look hard at the advisory surface. It recommends and does not execute, which makes almost every recurring-revenue claim about it weak. If the honest answer is “no, and here is what I would need to observe first,” write that. It is a pass, not a failure.
- Bring the three lines into the chat. The one I will push on is any “yes” whose mechanism runs through a renewal you have not dated.
What this does not cover
This lesson stayed on the run rate itself and said nothing about what happens to it as customers leave and expand. That is the retention pair — gross retention, which cannot rise above 100%, and net revenue retention, which can — and it is where a churn-reducing change like the review gate either shows up or hides. That is the next lesson. The two after it are the efficiency metrics, CAC payback and the Rule of 40, and the module closes on the subtractive move: saying out loud which of these five your change cannot touch.
Read this next — primary source
16 Startup MetricsAndreessen Horowitz — free; a venture firm publishing the definitions its own portfolio is judged by
This lesson takes three definitions from it — what ARR is scoped to include, what a booking is, and when revenue is recognised — and stops there. The page itself is longer and more useful than that: it is a list of sixteen ways a number that sounds like money turns out not to be, written by people who see the same misreporting arrive in pitch decks every week. Read it as a catalogue of the specific errors, not as a standard. And read it knowing who wrote it: a venture firm has a direct interest in how the companies it funds are measured, which does not make the definitions wrong but does mean nobody is obliged to use them.
Stuck, curious, or think this lesson is wrong? Ask your teaching agent. The lessons are the scaffold; the conversation is where the learning gets unstuck.