Death of ARR
5 Reasons why ARR is a flawed metric
Cursor went from $1-100m in ARR in 12 months — one of the fastest companies to ever do so. This is great!!
Rapid growth, less than 20 people, AI. This checks off all the boxes for any VC. However, as investors, we should stray away from looking at the headline ARR number as it can be extremely misleading.
In this essay, I’ll write about the death of ARR and why I believe the metric is extremely flawed especially in the age of AI. There’ll be overlapping themes, but churn is the common thread that underpins my argument.
Here are the 5 key reasons why I believe ARR is flawed:
Fast growth = Not Sticky
MRR x 12 is flawed
Experimental budget
Misrepresenting ARR
Doesn’t Capture Retention + Acquisition
1. Fast Growth = Not Sticky
If a company can grow from $1-100m in ARR in less than a year, then this means that people are signing up in waves. However, if people are this quick to sign up, then what does this tell us? Perhaps it tells us that the switch cost in this industry is low.
If I can save down my data, cancel my subscription, and move over to a new platform — then the product is not sticky at all. With that said, if a company is rapidly growing, we should ask ourselves: “Is this even a sticky product in the first place?”
We should first understand why customers churned from the old business, and moved to the new business. Then, via customer calls, we should aim to understand if the business has resolved that pain point that the customer faced in the old business.
It’s a positive sign if the company has resolved that issues and are setting up systems to lock-in the customer for longer-term. Last month on X, we saw waves of people leaving Cursor for Windsurf — calling into question how sticky this revenue truly is.
2. MRR x 12 is flawed
Most founders calculate ARR by taking their MRR and multiplying it by 12. I do it all the time myself, so I’m guilty of this too. However, this makes one huge assumption - no churn. The best way to represent this is by running a hypothetical using an AI study tool for college students. Let’s say in April the company reaches $100k in MRR = $1.2m in ARR. Amazing right?
But May comes around and college users decides to churn (since it’s summer break)… now, the MRR/ARR figure doesn’t look too hot right?
ARR inherently does a bad job capturing churn for businesses, especially while looking at it as a snapshot number.
3. Experimental Budget
How many people are actually using Hebbia? AI has gotten to a point where we’re seeing hundreds of companies conducting paid pilot to test softwares. Budget for AI has also increased, and will only continue to grow.
I don’t consider this as actual revenue though. I call this experimental revenue. Companies, however, are reporting this figure as actual, recurring revenue and considering it ARR.
This is slightly disingenuous in my opinion. It’s not truly “annual recurring” revenue if we’re not sure if they’ll continue pass their paid pilot. Moreover, per conversations with investors in the space, most companies have ran pilots, but often see little to no engagement or adoption.
Scott Kupor, a GP at a16z, guest spoke in a Wharton classroom and mentioned how most of the revenue captured in the AI space are experimental, and we haven’t seen a full cycle to see if these contracts will be renewed. However, we’re nearing the time, where we’ll be provided with more data and engagement metrics, allowing us to be more scientific about our investments.
4. Misrepresenting ARR
There’s also a huge misrepresentation of the definition of ARR and conflating run-rate with ARR. By definition, ARR is Annual Recurring Revenue. The key word here is recurring. A lot of service or project-based companies are actually talking about run-rate revenue and not recurring revenue.
ARR is contractual, run-rate revenue is not. And this differentiation is quite important. I’d argue the recurring revenue is worth more than run-rate revenue. As such, a $100m in ARR and $100m in run-rate revenue have wildly different meanings.
5. Doesn’t Capture Retention + Acquisition
This is the section where I just dump everything else about a business that ARR doesn’t capture, but I can write pages upon pages if I really had to dig deep. For me, it’s all about retention and acquisition. Having been an operator, there’s a couple things I care about:
Is retention improving over time? If I were to compare two cohorts: let’s say a cohort from a 6 months ago and a cohort from a year ago. Hopefully, I see a huge lift in M2 retention from the more recent cohort as we improve the product.
Is CAC going down? Is sales efficiency improving? Acquiring customers should ideally become easier, not harder overtime for a startup. If it is becoming harder over time, then that’s a sign that the company might not scale as smoothly as we would like it to be.
What metric should we use instead?
Ok, I admit… I was a bit clickbait.
There’s nothing wrong about ARR, but we need to be a bit more smart about how we think about this metric. Almost every company is setting new records for reaching $10m, $100m, $1000m in ARR — but let’s dig just a tad bit deeper before writing a check.
Interesting read. I wonder what other metrics we repeatedly use without analyzing the nuance behind it?