ARR is not revenue

Every subscription business leads with ARR. It’s the first slide of the board pack, the headline of the funding announcement, the number the CEO says out loud at the all-hands. And it appears nowhere in the financial statements. It isn’t audited, no accounting standard governs it, and it is not revenue: not this year’s, not next year’s, not any year’s.

That’s not a complaint. ARR is a good metric doing a job revenue genuinely cannot do. The trouble starts when the two are treated as the same number wearing different clothes, which happens constantly: in board meetings, in models, and in the gap between what sales is paid on and what the auditor signs.

Two different kinds of number

Revenue is a flow: an amount earned across a period. Under ASC 606 and IFRS 15 (the same five-step model, one for US GAAP and one for everyone else) revenue is recognized as performance obligations are satisfied. For a subscription, that means ratably over the service period. A $12,000 annual contract signed on 1 July produces $6,000 of revenue this year and $6,000 next, no matter when the cash arrived or when the ink dried.

ARR is a stock: a value at an instant. It’s the recurring subscription value on the books right now, annualized, or what you would earn over the next twelve months if absolutely nothing changed. That same $12,000 contract adds $12,000 to ARR on the day it’s signed, and sits there until it churns.

Stock and flow. Balance and movement. Photograph and film. Once you hold the distinction properly, most of the confusion dissolves and the rest becomes predictable.

Revenue answers “what did we earn?” ARR answers “what are we currently entitled to earn, if the world stops moving?” One is a fact about the past, the other a conditional about the future. They are never the same number, and they were never meant to be.

The four places they diverge

Putting this year’s ARR growth next to this year’s revenue growth and expecting them to agree is the single most common way the confusion surfaces in a board meeting. A usage-priced business often has the opposite scope problem to a services-heavy one, with real revenue that no definition of “recurring” will hold still long enough to count. And the SEC has pushed companies to say the direction point plainly in their own filings: actual revenue over the next twelve months is likely to differ from opening ARR, sometimes significantly.

Nobody’s ARR is quite like anyone else’s

That last point deserves its own section, because it’s the one that costs money.

A study of 136 listed SaaS, cloud and fintech companies found that 86 of them, or 63%, referenced ARR or a variant somewhere in their annual filings, investor materials or earnings calls, while only about a quarter reported current and prior-period ARR as a stated metric. More striking than the adoption is the divergence in what the word means. The calculations in use include monthly recurring revenue times twelve, last month’s GAAP revenue annualized with adjustments, contracted monthly recurring revenue plus trailing-twelve-month variable revenue, customer count multiplied by annualized subscription price, and invoiced subscription and maintenance obligations. Some companies exclude professional services entirely; others fold recurring services in. One counts contracted usage minimums but not the overages above them.

Every one of those is defensible. None of them is wrong. But it means two companies with identical financial statements can publish ARR figures that differ by double digits.

ARR is a word, not a measurement.

The companies themselves know it, and say so in writing. Adobe states that ARR “should be viewed independently of revenue, deferred revenue and remaining performance obligations as ARR is a performance metric.” CrowdStrike defines it as the annualized value of subscription contracts “assuming any contract that expires during the next 12 months is renewed on its existing terms,” an assumption doing an enormous amount of quiet work.

The SEC’s 2020 guidance on metrics in MD&A set a sensible bar for numbers like this: define it clearly, disclose how it’s calculated, say why it’s useful to a reader, and say how management actually uses it. That’s good practice whether or not anyone is filing anything. If your ARR definition isn’t written down as one paragraph a new analyst could apply without asking a colleague, you don’t have a metric. You have a habit.

Where the confusion does damage

Planning. You cannot build a P&L from ARR. Converting the stock into the flow means handling recognition timing, mid-period starts, and every non-recurring line ARR deliberately drops. Teams that plan the ARR bridge and then “annualize to revenue” discover the gap in Q1, usually in front of an audience.

Targets and pay. A sales team compensated on ARR and a board tracking revenue are watching two different clocks. Both can hit their number in the same quarter and disagree about whether it was a good one. That’s a design problem, not a communication problem.

Diligence. The ARR definition is the first thing an acquirer’s quality-of-earnings team pulls apart. They rebuild it from the contracts, and the number that survives is frequently not the one on the slide. The worst moment to learn how much of your ARR depends on a renewal assumption is halfway through a data-room request list.

The annualization trap. Multiplying one good month by twelve is how a seasonal spike becomes a growth story, a three-month pilot becomes a full-value customer, and consumption revenue that nobody committed to becomes “recurring.” Every one of those unwinds eventually, and it unwinds in the revenue line, where it’s audited.

Holding both honestly

The fix isn’t picking a side. It’s keeping both numbers, using each for the job it’s built for, and being able to explain the distance between them.

That means three things: the definition is written down and applied the same way every quarter; the ARR bridge (opening, new, expansion, contraction, churn, closing) sits beside the revenue waterfall rather than substituting for it; and there’s a standing reconciliation from closing ARR to the following period’s revenue. Not to make them equal, which they never will be, but to make the difference itemized: recognition timing, services and one-offs, usage above or below commitment, churn assumed against churn observed.

Explaining the gap is the deliverable. A finance team that can produce that reconciliation on request has two metrics. One that can’t has one number and a rumour.

In Novi, they’re two views of the same contracts

In a spreadsheet, these numbers live apart. ARR is maintained in a revenue-operations workbook by the people closest to the customers; recognized revenue is built in the accounting model by the people closest to the auditor. Two sources, two definitions, two sets of hands, so they drift, and the reconciliation becomes a quarterly archaeology exercise that produces a number nobody fully trusts and everyone stops questioning.

In Novi, the contracts are one structure. A subscription is an element with a start date, an end date, a value, a customer and a product. ARR is a point-in-time aggregation over that structure; recognized revenue is a period aggregation over the very same one. Because time is a real dimension of the model rather than a column layout someone maintains, “ARR at 31 December” and “revenue for the year” are two reads of one model, not two models kept politely in sync. Change a contract’s start date and both move together. They can’t disagree, because there is nothing to reconcile.

And the definition stops being folklore. The rule that decides what counts as recurring is a formula sitting in the model: legible to the analyst who inherits it in two years, and legible to the machine that has to answer why ARR is up while revenue is flat.