Will AI Mess Up Revenue Multiples?

Mostly metrics is proudly sponsored by Runway.

“Hey, metrics man! Will you help me count my candy?”

Will AI Mess Up Revenue Multiples?

AI is shaking up a lot of industries, but one question I keep coming back to: what does it mean for how we value companies? Specifically, will AI revenue kill the beloved ARR multiples that SaaS companies rely upon?

As we move to a pricing model based on outcomes rather than access, AI revenue feels more transactional, and less predictable.

Am I wrong?

I asked Kyle Poyar from Kyle Poyar’s Growth Unhinged this very question, and his response gave me a lot to think about.

Recurring vs. Reoccurring Revenue – What's the Difference?

Kyle pointed out something that’s easy to gloss over—there’s a big difference between recurring and reoccurring revenue. Let’s break it down:

Recurring revenue is predictable, happening at regular intervals for the same amount. Think about my monthly Spotify subscription of $11.99, billed like clockwork.
Reoccurring revenue happens more than once, but it’s irregular, both in timing and amount. Think Uber rides or payments revenue for a company like Toast—restaurants continue to use the service, but the amounts vary.

A lot of AI products, especially those linked to a successful outcome, like support tickets solved or meetings booked, fall into this "reoccurring" category. You might keep using the product, but it’s not as predictable or consistent as a standard SaaS subscription.

What About Valuations?

Here’s where Kyle's insight really hit home:

❝

"If you think about valuations, they're very tied to ARR multiples in software, not like other industries where it's maybe tied to EBITDA multiples.

We're very attached to ARR multiples because the revenue is high quality. It's usually high margin, at least 70%, often 80%. So where ARR is kind of a proxy for gross profitability, it’s highly sticky, with net dollar retention over 100%. You can feel confident that ARR will stay consistent or increase over time.

But with AI products, margins are often worse than classic SaaS. Some AI products even have negative or highly variable margins. Plus, AI revenue isn’t as predictable, with high churn from early adopters who might not have a mission-critical use case yet.

ARR won’t be the best predictor of valuations for AI companies. We’ll need alternative metrics to help us assess the quality of these businesses."

-Kyle Poyar on the RTN Podcast

Is ARR Still King?

In SaaS, we’ve long relied on ARR multiples because, as Dave Kellogg joked, "If we don’t mess up, this is what we get next year."

But with AI, things get a little messier. AI products don’t have the same margins, and churn is higher because many early users are still just "kicking the tires." So, does this really count as ARR?

ARR assumes steady, predictable revenue streams, and AI doesn’t always fit that mold.

Time for a New Metric?

If ARR multiples don’t work as well for AI companies, what’s next? Kyle suggests we start thinking about annual revenue run rate (ARRR) instead. This would capture all types of revenue—whether it’s from software, AI, or payments—and then assess the quality of that revenue based on mix. We should be asking: what’s the margin? How predictable is it? How sticky is the customer base?

Perhaps we stop looking forward, and start looking backwards at the trailing twelve months of revenue? At least that way we’ll know what actually occurred. But that, of course, would mess up discounted cash flow models, which are predicated on forecasting the future.

Maybe, as someone commented on one of my LinkedIn posts: "We just go back to GAAP revenue and cut out all the fluff?" He is annoying. But he’s got a point.

Another issue: a lot of public companies don’t even give you a proper ARR number. Some just multiply quarterly revenue by four, which isn’t really what ARR is meant to reflect.

There’s hope that AI margins will improve as products scale, but until then, we should be cautious about blindly applying ARR multiples to AI businesses without deconstructing revenues. Kyle summed it up well: AI isn’t taking CFO jobs anytime soon—in fact, it’s giving them even more to think about.