ARR Doesn’t Need a Redefinition

It’s still recurring. Just a little messier.

With the explosion of AI products, there’s been a wave of panic about how to measure ARR. Investors and operators are all asking the same thing: if everything is usage-based now, what even counts as recurring revenue?

You’ll see headlines that say things like “ARR is dead” or “SaaS is over.” Founders are rewriting pitch decks, finance teams are scrambling to redefine metrics, and everyone seems to be acting like we’re in a whole new world.

But the truth is… not much has actually changed.

Yes, pricing models are evolving. Yes, it’s getting harder to slap a simple definition on ARR. But the messiness isn’t new. We’ve been dealing with it for years, we just didn’t talk about it as much.

ARR Was Never That Clean to Begin With

ARR sounds like a clean metric. Annualized. Recurring. Revenue. It implies predictability and precision. But in practice, ARR has always involved judgment calls.

Is it contract value or trailing revenue times 12? Do you include usage overages? Do credit packs count? What about outcomes-based fees?

You could line up 10 finance teams and get 10 different definitions. And that’s fine, as long as each one can clearly explain what’s included and why.

What matters isn’t that everyone uses the same formula. What matters is that you’re consistent, honest, and aligned internally. ARR has always been a directional metric, a way to summarize your revenue base and project where it’s headed. It’s not GAAP. It never will be.

Examples of ‘Non-Recurring’ ARR in Classic SaaS

This isn’t just a new AI-era phenomenon. Some of the most iconic SaaS companies have had messy, variable revenue for years.

Take Snowflake. It’s entirely usage-based. Customers pre-purchase credits and burn through them as they consume compute and storage. There’s no fixed monthly fee. No guaranteed spend. Yet, it’s one of the most highly valued “recurring revenue” companies in the world. Why? Because usage is sticky, and usage follows patterns.

Same with Datadog. They charge a mix of flat fees (for core platform modules) and usage-based pricing (for metrics, logs, traces). Customers scale spend as their infrastructure scales. It’s variable, but it’s also predictable.

Even at Intercom, we had all sorts of non-flat pricing. Some customers paid by seat. Others paid based on the number of monthly active end users. Some components had usage tiers. All of it fluctuated month to month. But we had years of historical data to prove that it followed a pattern. That we could model it. That it behaved like recurring revenue.

ARR didn’t mean “fixed.” It meant “forecastable.”

What Actually Makes Revenue Recurring

So let’s get to the real question: what actually makes revenue recurring?

It’s not whether the customer signed an annual contract. It’s not whether they pay monthly. It’s not even whether they use the product consistently.

It’s whether you can predict their revenue contribution with some level of confidence over time.

That’s the job. And that’s why ARR is so often a judgment call. You’re making a case that this revenue will stick. That it will grow or shrink within a known range. And that you have data to back up that case.

Sometimes that’s a clean MRR x 12. Sometimes it’s a more complex model that averages usage, normalizes spikes, or buckets customer cohorts by volatility.

But the goal is the same: show the recurring nature of your revenue as it exists in your business.

The Newer Edge Cases

It’s true that we’re seeing even more creative pricing models in the AI era.

Clay, for example, sells credit packs up front. You might spend $5K on credits in January and burn through them over five months. Is that recurring? Not exactly. But if a customer has a history of rebuying credits every few months, you might reasonably argue that their usage is recurring in practice.

At Intercom, we now offer FIN, our AI chatbot, on an outcome-based model. You pay for successful resolutions. That’s new. That’s different. But again, it follows a pattern.

The question isn’t whether it’s “traditional.” The question is whether it’s predictable.

The line is blurrier, yes. And we’ll likely see more examples where founders stretch the definition of ARR a little too far. That’s where trust breaks down. And it’s why transparency matters more than ever.

The Role of the First Operator

This is where the First Operator comes in.

You are the one responsible for defining ARR in your company—and defending it. That doesn’t mean gaming it. It means building a definition that makes sense for your business model, aligning with your CEO and board, and communicating it clearly.

It means modeling your revenue in a way that reflects reality, not fantasy. It means tracking variability, explaining seasonality, and proactively flagging risks.

You don’t get to hide behind a blanket definition anymore. You have to own it.

Know What’s Truly Recurring

ARR doesn’t need a redefinition. It just needs a little more honesty.

The most important thing in all of this is understanding for yourself what, in your business, is actually recurring. What is truly predictable. What you can defend.

And then helping your executive team, your board, your investors understand that too. That’s always been the game.