How to Calculate ARR: Every Method, and How to Choose Yours

Annual Recurring Revenue is the predictable, recurring value of active customer contracts normalized to a one-year period. It is not a GAAP metric, and there is no universal standard for calculating it. As Ben Murray of The SaaS CFO puts it after reviewing hundreds of public disclosures: ARR not being GAAP "makes how you define it more important than the number itself."

That's why "how do I calculate ARR" doesn't have one answer. What it has is four independent choices. Make them in order and the arithmetic falls out; the combinations are what produce the wide range of numbers you see across companies that look similar from the outside.

1. What revenue counts
2. Whether you annualize contracts or revenue
3. Which annualization formula you use
4. How you handle usage and other variable revenue

Then a set of population rules — churn timing, renewal assumptions, thresholds — and, optionally, which related variants you report alongside it.

Choice 1: What revenue counts

The dividing line most companies use is whether revenue continues if the customer takes no further action.

‍

There is variation even here. Commvault includes "maintenance related to perpetual and term licenses, extended maintenance contracts (enterprise support), and managed services" — recurring services revenue that some companies would exclude on principle.

Murray's rule for resolving the edge cases is a useful one: if it appears in the revenue section of your P&L, state explicitly whether it is in or out. Silence is where ambiguity accumulates.

Choice 2: Contracts or revenue

This is the data source your calculation reads from, and it's roughly an even split in practice. Across Murray's dataset of public definitions, the divide between a revenue build and a contract build runs "about 50/50," with hybrids most common among companies selling both subscription and usage.

Contract-based. Annualize contracted value — TCV or ACV. Anchored to what customers have committed to. Reflects signed commitments even when billing lags, and doesn't move when a single month runs hot or cold.

Revenue-based. Annualize the current recurring revenue run-rate. Anchored to what is actually being earned right now. Reflects reality faster, including the reality of a bad month.

Hybrid. Contract values for fixed subscriptions, actual consumption for variable usage. The standard answer for businesses with a committed platform fee plus metered usage on top.

Choice 3: The annualization formula

Murray's observation here is worth internalizing: timing language is not a formula. "Annualized as of period end" describes when you measured, not what you did. The actual formulas in use:

‍

The day-count and month-count variants of contract annualization usually land close together. On a book with many mid-month starts, they don't.

One consequence worth knowing about: annualizing contracts shorter than a year can produce ARR above the contract's own total value. Alteryx disclosed exactly this in a 2023 investor deck, noting that annualizing sub-year contracts "results in amounts being included in our ARR calculation that are in excess of the total contract value for those contracts." Disclosed, and arithmetically unavoidable under that method.

Choice 4: How usage is handled

For usage-heavy businesses this choice moves the number more than the other three combined. There are five positions, and all five are in use:

• Trailing 90 days × 4 — the most popular approach, per Murray's review
• Last month × 12 — fastest to reflect growth, most exposed to a spike
• Trailing twelve months — smoothest, slowest to reflect a genuine step change
• Forward-looking run-rate estimate — explicit management projection, requires the most disclosure
• Excluded entirely — usage sits outside ARR

Datadog and Dynatrace, direct competitors in observability, sit at opposite ends. Datadog's ARR aggregates "monthly revenue from committed contractual amounts, additional usage, usage from subscriptions for a committed contractual amount of usage that is delivered as used, and monthly subscriptions." Dynatrace excludes "any revenues derived from month-to-month agreements and/or product usage overage billings, where customers are billed in arrears based on product usage."

Both are disclosed in SEC filings. Applied to the same book of business, they produce different numbers.

The population rules

Four choices define the calculation. These define who's in it.

Churn timing. CrowdStrike keeps an expired contract in ARR "if we are actively in discussion with such an organization for a new subscription or renewal, or until such organization notifies us that it is not renewing" — notice is the trigger. The SaaS Metrics Standards Board takes the opposite position for CARR: churn "should be deducted in the same period the contract expires," and specifically "should not be deducted from CARR upon learning about known churn in the future." Notice-based removal is more forward-looking; expiry-based is more contractual.

Renewal assumption. Most definitions assume active contracts renew at the measurement date. This keeps ARR a snapshot rather than a forecast, and leaves churn to be analyzed separately.

No assumed expansion or contraction. ARR reflects current commitments or run-rate, not expected growth.

Month-to-month treatment. Zoom includes monthly subscribers "who have not provided any indication that they intend to cancel their subscriptions." Dynatrace excludes month-to-month outright.

Minimum thresholds. Some companies count only customers above a floor — $10k ARR, for instance — to reduce noise from very small accounts.

Point-in-time measurement. ARR is measured as of a date. It is not an average, not cumulative, and not a forecast.

Related metrics worth separating

CARR (Contracted or Committed ARR). A distinct metric, not a synonym. The SaaS Metrics Standards Board defines it as "contracted annual recurring revenue, whether in production or not yet in production," calculated as (MRR × 12) + contracted ARR not yet in production. It captures signed deals that haven't gone live. Three of the Board's nuances matter in practice: multi-year agreements contribute "only the contracted subscription revenue for one year forward" at that year's specific rate; usage above a contracted minimum is excluded regardless of invoicing; and where there's no minimum commitment at all, "no CARR is present" — which is why consumption-first businesses get little from the metric.

Segment ARR. Cloud ARR versus total ARR, subscription versus usage ARR, or product-level splits. Atlassian reports Cloud ARR specifically, calculated as Cloud MRR × 12.

Constant currency ARR. Strips FX movement to isolate operational performance.

The same customers, calculated different ways

A company on 31 December with six customers:

• A — $120k/year annual contract
• B — $5k/month, month-to-month, no contract
• C — three-year deal, $450k total, ramped $90k / $150k / $210k, currently in year one
• D — pure usage; $8k consumed in December, $60k over the trailing twelve months
• E — $120k/year contract, gave notice 15 December, term ends 31 March
• F — $200k first-year contract including $50k of one-time implementation

Four customers drive nearly all the variance:

C — multi-year ramps. Annualizing TCV over the term reports $150k; the customer pays $90k this year and is committed to $450k overall. The Standards Board's guidance for CARR is to use the year-specific contracted rate, which resolves it at $90k.

F — bundled services. Annualizing the contract total pulls in $50k that will not recur. Working from recurring lines instead gives $150k.

D — usage timing. December × 12 gives $96k; trailing twelve months gives $60k. Trailing 90 days × 4 would give something in between. This single choice is a 60% swing on that customer.

E — churn timing. Only the notice-based approach removes them before the term ends.

Choosing for your business

The four choices aren't independent of your model. A rough mapping:

‍

Two practical notes. Pricing model does not dictate ARR model — plenty of usage-priced companies run subscription-based ARR, and the disclosure is what tells a reader which you've done. And running your book through more than one method before committing is worth the hour: if the results cluster, the choice is low-stakes; if they spread as widely as the example above, it's a decision worth making deliberately rather than inheriting.

The movement components

ARR is reconciled across five components:

• New ARR — from customers who weren't customers at period start
• Renewal ARR — the baseline retained from existing customers renewing
• Expansion ARR — upsells, tier upgrades, added seats, higher committed usage
• Contraction ARR — downgrades and reduced seats
• Churned ARR — fully cancelled subscriptions

The bridge runs: beginning ARR, plus new, plus expansion, less contraction, less churn, equals ending ARR. Renewal sits inside beginning ARR rather than appearing as a separate movement, since a customer renewing at the same price produces no change — it's tracked alongside the bridge as a measure of what was at risk and what held.

Two ratios come from the same components. Gross revenue retention takes beginning ARR net of contraction and churn only. Net revenue retention adds expansion, which is why it can exceed 100%.

Writing the definition down

Murray's disclosure template covers the four choices and the population rules in one paragraph, and adapts cleanly to a private company's board reporting:

"Annual Recurring Revenue (ARR) represents the annualized value of recurring revenue from active customer contracts as of period end. ARR includes recurring subscription fees and usage-based revenue. Professional services and other non-recurring revenue are excluded. Subscription ARR is calculated by dividing total contract value by contract duration and annualizing to 365 days. Usage-based ARR is calculated by annualizing the prior 90 days of their actual consumption, assuming no increases or reductions in their subscriptions or usage. ARR is a non-GAAP operating metric and may differ from GAAP revenue and from similarly titled metrics used by other companies."

Consistency is the part that compounds. Definitions that change between periods make growth rates unreadable unless prior periods are restated — and restatement, disclosed as such, is increasingly common as companies tighten their definitions.

Reporting across multiple companies

For fractional CFOs, advisors, and anyone reporting on a portfolio, the four choices are inherited rather than made: each client may already have a different definition. A roll-up reflects those differences unless the underlying figures are normalized to a common method first. The usual approach is to keep each client's own method for their own board reporting, maintain a separate normalized method for cross-client comparison, and label which is which on every deliverable.

Nockpoint's semantic layer lets a metric definition live in one governed place, so a figure like ARR resolves the same way across every dashboard and report that references it.