Annualized Recurring Revenue

Many software companies don't sell their product as a one-off, but as a subscription: customers keep paying every month or every year. Annualized recurring revenue answers the question of how much money would come in from these subscriptions over a full year if nothing changed. So you take the subscriptions currently running and project them out to twelve months. If a thousand customers pay 100 euros a month each, that's 100,000 euros a month and therefore 1.2 million euros a year. That 1.2 million is the ARR, the English abbreviation for Annual Recurring Revenue. Importantly, it is not a measurement of the past, but a projection of the current state into the future.

Why investors look at ARR first

Classic annual revenue shows what has actually come in over the last twelve months. For fast-growing companies, this backward-looking view is almost worthless. A company that makes 10 million in revenue in January and 100 million in December has a low annual revenue figure – but enormous momentum. ARR captures exactly that, because it only projects today's state forward.

On top of that comes reliability. A subscription keeps running as long as no one cancels. That's why recurring revenue is considered more predictable than one-off sales. A company with 50 million in ARR has a rough idea of what it can expect next year. A company that earned 50 million from one-off projects has to fight for that money all over again every year. For investors, the first case is worth considerably more.

That's why ARR is also the basis for many valuations. For start-ups, a multiple of ARR is often used as the company's valuation, say ten to twenty times. If a company doubles its ARR, its valuation doubles too under this logic. That explains why founders so like to put this one number front and center in press releases.

What may go into the calculation – and what may not

ARR should only include revenue that recurs on its own. So subscription fees, licenses with a fixed term, fixed monthly service flat rates. It should not include one-off payments: a setup fee, a consulting project, a piece of hardware sold. Anyone who includes such items artificially inflates the number.

In practice, many companies calculate the metric by projecting the last month forward: current monthly subscription revenue times twelve. Some even take just the last week. That makes the number very current, but also sensitive. A single large customer who signs in December can raise ARR by millions in one stroke, even though in reality only a fraction of that has actually come in yet.

ARR is also not profit. It says nothing about what the company spends on servers, salaries, or advertising. Many AI companies with high ARR simultaneously post high losses, because running the models is expensive. You also have to factor in churn: the share of customers who cancel. An ARR of 20 million with 40 percent annual churn is something completely different from the same figure with 3 percent churn.

The number in headlines about OpenAI and co.

Anyone reading news about AI companies keeps stumbling across this metric. Sentences like "the start-up has reached 100 million dollars in ARR" have become standard. What's almost always meant is the projection of a single good month, not a completed fiscal year. At OpenAI, Anthropic, or Cursor, such figures are often leaked by investors rather than officially audited and published.

As a reader, it's therefore worth asking a quick follow-up question: What time period does the number refer to, and has it been confirmed by an auditor? Trial periods, first-year discounts, and non-binding letters of intent show up in some calculations. Reputable companies additionally state the churn rate and the number of customers.

Even with services you know yourself, this metric lurks in the background. A Spotify subscription, cloud storage, or a ChatGPT subscription at 20 euros a month each amount to 240 euros in ARR. Millions of such small amounts add up to the billion-figure sums reported on in business news.