Calculating NRR in Usage- and Outcome-based Pricing

Live from Schenectady, New York, it's the Metrics Brothers, Growth and CAC.

And I'm Growth, better known as Ray Reich, founder and CEO at Benchmarkit.

And I'm CAC, better known as Dave Kellogg, EIR at Balderton Capital, independent consultant, and the author of Kellogg.

And together we are the Metrics Brothers.

And we go together like meters and electricity.

Well, with AI's appetite for energy, I'm thinking maybe we're doing another episode about gigawatts or Manhattans.

Uh, no, not today, Ray. I thought we were gonna talk about kind of running the meter, um, specifically when you run the meter in usage and output of an outcome-based pricing and how that impacts the calculation of NRR.

Ah, I get it. In today's usage-based pricing environments, you need to meter the product usage, which feeds the key to calculating revenue, the bills or invoices, and then over time, the actual ARR and even NRR, Dave.

Yep, Ray. I still remember the session at SAS Metricspalooza '24 when I talked about ARR as the Achilles heel. of SaaS, basically, and SaaS metrics. And I think it's even more relevant today as we continue to have more and more companies moving to what I would call variable pricing, usually driven underneath by usage or outcome-based pricing.

Okay. Well, before we dive any deeper into Greek mythology or, God help us, existential philosophy, let's hear a word from our sponsor. The Metrics Brothers is proudly sponsored by Benchmarket, the The B2B AI and software industry's largest, broadest, and most contextualized benchmarking index. Dave Kellogg and Ray Groth Wright have put the Metrics Brothers podcast on hiatus for the remainder of 2026. All episodes we have ever recorded are still available on your favorite podcasting app. You can reach Dave at Kellogg and Ray on LinkedIn at Ray Reich. Thank you so much for listening to the Metrics Brothers podcast. Okay, Dave. So since we're talking about how variable pricing models, usage-based, outcome-based, impact NRR, I thought it might be a good time to revisit all the way back to our episode number 13, almost 2 years ago, to the definition of NRR. What do you think?

Let's do it.

Okay. So net revenue retention, also known as NRR. It measures how much recurring revenue you keep and expand from your existing customer base over a given period. Net revenue retention includes all the upsells, cross-sells, downgrades, and churn from those existing customers. And the value of NRR is it really tells you, are your customers buying more from you over time? It's a great signal for product-market fit durability, and it's a great signal for investors because net revenue retention, as measured by R-squared, is still the number 3 correlated metric to enterprise value to revenue multiples for public SaaS companies at 0.24. Only growth at 0.31 and Rule 40 at 0.32 is higher, Dave.

Got it. So a couple of other quick basics on NRR. Uh, one, probably the most frequent problem, is it really should be done on a cohort basis, meaning that it's done without survivor bias, meaning that we go back and get the cohort of people who were customers 1 year ago, regardless of how long they were a customer at that time. We get the cohort of people who were customers 1 year ago, and we take those same customers and look at them today, and then divide basically their value today by the value yesterday. That's how you get NRR using a cohort-based method. There are other ways to do it. But I think they all are very slippery slopes, some of which I think Ray will slide down with us in this episode. Um, the other thing that can be interesting about this is a little bit on timing. This is a bit of a nuance, but you know, as people talk more about land and expand, some companies will say, hey, if I have somebody who buys 100 units and then within 30 days buy another 100 units, so I got 200 units in a month. Then when I do my CAC ratio, I should really count $200 as kind of the initial land, right? So I can take that sales and marketing expense because they're at $200 for it. Some people get very aggressive about this and go, I signed a 3-year contract and the terminal value in year 3 is $300. So I paid for that, right? That's a deeper rabbit hole. Don't want to go down. But there is some notion of just the timing of this. And I think for CAC calculations, people like to kind of stretch the land out. And say, well, it's anything we get in the first 30 days, 60 days, 90 days. And when we're doing NRR, we need to have a precise point in time. Was it, was it at the end of the quarter or the start of quarter? In general, it should be at the end of the prior quarter, right? You take December 31st this year divided by December 31st last year. So if you do NRR with the cohort method, you won't step into trouble. If you don't, I think you might. So, but the main thing being here is that obviously NRR excludes new customers, as you talked about, but the answer why? Because they're not in that cohort. So if you use the cohort method, It's kind of self-correcting, right? You get the right answer because you wouldn't even be tempted to include them because they weren't in the cohort of existing customers one year ago.

Okay. Well, to calculate NRR in any environment, whether it's a usage-based, outcome-based, or traditional subscription, it does require you to have a standard definition of ARR. And I'm not sure we want to go down this rabbit hole, Dave, but I think it's really important that as variable pricing evolves, how are you calculating ARR? So do you want me to give you your 3 most common models I see out there today?

Well, first, before you do that, Ray, let me tee up one thing in advance. You either have ARR or you don't. And ARR, basically, from first principles, is annual recurring revenue. You have an annual contract with somebody at 100 units and they renew at 110, right? NRR, 110 divided by 100, so 110%. Everything else, and then all of SaaS metrics are based upon ARR, right? There's always ARR in a variable and you're plugging something in. And the way I put this, Ray, is proxy. It's just, what are you going to— since you don't have ARR, what are you going to use as a proxy for it? And I just think it's a very important starting off point because people can then do, ooh, how do you define ARR in this environment? How do you define ARR in that environment? It's like, whoa, whoa, whoa, we don't have ARR. If the business model isn't annual or contracts renewing once a year, It's not ARR. So what are we going to use as a proxy? MRR times 12, right? One of the simpler proxies. Or are we going to get into some of the more complicated proxies, which you want to cover right now?

Yeah. In fact, you know, so many people are using MRR times 12, which is really just looking at your recurring revenue from the latest month, multiplying it by 12. What I'm seeing a little bit more common, and a lot of public companies do it, and it's called implied ARR, Dave, Is you take last quarter's recurring revenue, you multiply it times 4, and that gets you implied ARR.

Yeah, what is recurring revenue? Most of them just use last quarter's revenue or subscription revenue or product revenue, but they're not— I mean, I think by recurring you meant to say not services, but I want to call that recurring. Maybe some people do, but to me, it's their subscription revenue is what I'd call it. Although even that's debatable because there's the overage part of the subscription, right? I'm not meaning to apply baseline, but I'll just do what Snowflake does. Product revenue. They take their product revenue, multiply it times 4 for the most recent quarter, and that creates implied ARR. And certainly with public companies, that's the standard way to do things.

Yeah. Now, this is a little bit of a rabbit hole, but would there be any problem just looking at your last 12 months of I'm still going to use the word recurring revenue and just say, hey, those last 12 months, that really is my annual recurring revenue.

Well, I mean, once again, I mean, the issue is— so look, Ray, to me, if you don't have ARR, you don't have ARR. So I don't like calling stuff— I like what Snowflake does, right? You find exactly zero instances of the word ARR in any of their 10-Qs or 10-Ks. So to me, it's just revenue— product revenue, software revenue, service revenue. Not services, that's even more confusing. But I wanna call it product revenue like Snowflake does. What's the difference between quarterizing Q4 or annualizing Q4? It's not annualizing. Yeah, it is annualizing or kind of run rating Q4, which is, I always think of run rating quarters as multiplying by 4. What's the difference between the Q4 run rate, right? And charting 12 months? The answer is there can be, particularly in a seasonal business, right? If you're a retailer, you do 40% of your year in December or more. So you're gonna be taking, you know, something that's really high and multiplying it. So it's not going to track very well. It's going to consistently overstate annual revenues, right? And the whole idea was the other way around, is that it was in ARR, it basically flowed into annual revenue. So like ARR could be a leading indicator of revenue, right? So I think implied ARR, but doing it the other way, but just taking revenue, multiplying by 4, it can work. It's what people do. But if you want to damp out that problem, you could just take I mean, Ray, trailing 12-month revenue is a name. It's called revenue, right? 2025 revenue is as of December 31st, trailing 12-month revenue. So, um, you could either just look at revenue, or if you want to play kind of games with it, um, and basically the whole exercise here, people are trying to find something they can plug into the formulas where ARR is supposed to go. Um, and most people who are public do that. by taking the quarterly revenue and multiplying by 4.

Okay. Well, let's now pivot to net revenue retention in a usage-based world. So I'm going to set it up like this. Hey, you're a hybrid revenue company. You have a platform subscription and you have a usage on a per-unit basis. Let's say you got $100,000 for the annual subscription, right? That's easy. Divide that by 12. Right. And you get your MRR or that $100,000 annual commitment is your ARR. But now this usage ramps over time and the first couple months it's only $100. It ramps up to $15,000 a month in the last 6 months. But at the end of the day, David, at the end of that first year, you have $100,000 of additional revenue from the usage base. So $100,000 subscription, $100,000 variable revenue over a 12-month period. So the first question is, is that $100,000 of usage revenue going to be called ARR?

So it depends how you define ARR, right? Um, some people like Ray— Ray likes to split the baseline and the overage. And I think for internal reporting, that can be meaningful. It depends a lot on your price list, but this is the way you— this is The way you price your software, these would be interesting reports to look at, right? What was the base and what was the variable and was it up to— public companies don't split it out, right? So in a public company, you just see revenue anyway. So to me, it all goes to revenue. Public companies generally don't release ARR, but so analysts calculate on their behalf implied ARR, and they're going to take quarterly revenue, which in this case will be $25,000, and multiply it by 4, um, to get $100,000 in, uh, ARR. So yes, I think is the short answer. It would go in there because we're not distinguishing between the two.

Okay. So now let me go into a model we were discussing before the episode, because, and this is all about NRR. You sign up a new customer. Hey, the first 4 quarters, they just do $100. Then quarter 5, they do $100 again. And then quarter 6, they go to 120. Quarter 7, they go to 120. Quarter 8, they go to 110. So the question is, in that model, Dave, how is NRR calculated at the end of the second year?

So, I mean, the answer is there's different ways to do these things, right? The 2 that spring to mind to me for NRR are the snowflake— what I call the snowflake method. Which is just, I think you call it a 2-year lookback. It's basically you take product revenue for year N, the most recent year, and divide it by product revenue for year N minus 1. Actually, other way. Yeah, yeah, today's one. So product revenue for year N divided by product revenue for year N minus 1 for all customers who were customers at the start of year N minus 1.

Okay.

Right? So you got to get the cohort. Which is, okay, let's go back effectively 2 years, right? So, if it's December 2025, I got to start there. And then I got to go back to people who are customers as of January '24, right? That's going to be my cohort. And then I'm going to say, okay, what was their revenue in '24? That's going to go on the bottom. What was their revenue in '25? That's going to go on top. Divide one by the other. That's the snowflake method called a 2-year lookback. That's one way to do it.

Okay. So in that model, using the numbers I threw out, they did $450,000 in year— the most current year, $400,000 in the previous year. $450,000 divided by $400,000. So you have about 112.5% NRR.

Yeah. I'll trust you on the math. That sounds right. The other way to do it, which will, I don't believe, produce the same result, or certainly in many cases won't. I'm not sure if in the example it will. is you do change in implied ARR, which is probably the wrong way to do it, so I won't belabor it. But theoretically, if you wanna get a proxy for ARR and shove it into your formulas, you could do that because ARR today, the issue is you'll never be able to do this for a public company 'cause you won't be able to get the cohort. So you actually can't do it outside in. But if you wanted to, theoretically, another way to look at it is just plug implied ARR into the formulas where ARR goes. You won't be able to do it unless you have the insider data. But that would be, you know, will not get you the same answer due to the problem we talked about. If December is a huge quarter for you, or if you have a big deal signed, when you do kind of trailing analysis or you look at a whole year, it's going to spread things out. Whereas on these quarterly analysis, for example, if you had a huge Q2, it'd be entirely missed, right? Because it's going to look at Q4 times 4 divided by a year ago Q4 times 4.

Yeah.

And that will give you quote unquote change in implied ARR, or basically you divide the top one by the bottom one in this case because we're trying to get NRR. So, but you could do that. My recommended, my preferred, look, I have a simple rule here, Ray, which is either you have ARR, if you don't, if you don't use a proxy and anytime you want a good proxy, use trailing spend, right? Just what did they spend? I don't care if it was subscription or overage. I just want to know how much this customer spent in this quarter. And I'm going to compare that either to the prior quarter or the year-over-year quarter. But basically, spend is truth. And all kinds of fancy, hey, let's proxy ARR with December times 12. You just get in a lot of trouble real fast, right? Because December, particularly enterprise software, December is a huge month. So, you're going to take the biggest month of the year by far, multiply it by 12 and call that ARR. And it's just not, right? It's going to overstate the forward 12 months revenue by quite some margin.

Yeah. Okay. So in this usage-based world, 2 basic models we've talked about, the 2-year lookback, which is the most current whole 12-month period versus the previous 12-month period. Or if you want to do it on a quarterly basis, you could look at Q4 2025 versus Q4 2024. But once again, that doesn't really show you a rolling 12 months.

Well, bear in mind, you could do the snowflake method for quarters, right? It's just the trailing 4 quarters at each time period. So if you want to do it quarterly, it works out easily. If you wanted to do it only off implied ARR, you could notionally plug that into the formula, but you suffer from this problem of annualization of quarters because the year is rarely 4 times any quarter.

Now, I didn't want to do this, but I'm going to give you a little bit of a rabbit hole. We can decide whether to go down or dedicate a new episode to it. But some companies, private companies especially, they want to show their expansion ARR on a quarterly basis. But in a usage-based model, that expansion could go up in one quarter. And if it's a down quarter seasonality, it could go down. So my recommendation is it's very hard to define and report expansion ARR every quarter. It might be safer to do it on an annual basis.

So I don't know. I think that might be a little bit of a cop-out. I mean, I think you can do it. First, look, did we just talk about the real underlying problem here? I'm just going to call it noise. Because usage variation is either kind of signal or noise, right? If somebody's constantly growing and using more of your software and you could find this trend line of, wow, this customer is consistently growing at this rate over time, well, that's information. Accordingly, you could have just straight noise. And I'll just do a simple example of where a given customer just follows a pattern like they spend $100, then they spend $110, then they spend $100 again, then they spend $90, then they spend $100 again. So that takes 5 data points. But after 5 data points, they've gone from 100 to 100, right? So if you look at the NRR over that period, whether that period is months or quarters, right, the NRR, if we looked at it and we're smart, we go, oh, it's 100%. They started at 100, they ended at 100, and they have this pattern, this kind of sinusoidal pattern where they go up and then down, but they always end up at the same place, right? Which would be different from that sinusoid riding on a line, right? That would be the case where there was signal and noise. Or you can think about it, right? But the issue is you got to be very careful because you might aggregate. Like, for example, if you wanted to calculate expansion ARR, you're going to add every one of those pluses, right? And you're going to have a whole bunch of expansion ARR. Just say it says it's 5 at a cycle, there's 50 weeks. So you run 10 of those cycles and you've got 1, 2 up motions per cycle and 2 down. So you're gonna have 20 units of expansion ARR if I did the math right. Not 20. You're going to have how many cycles? You're going to have 10 cycles times 2. Yeah, 20. 20 units of expansion. You'll have 20 units of shrinkage. And if you add them all up for the year, you're going to have, you know, be net zero. But if you look at a quarter, particularly in this one, because that cycle, if those are months, is not completing within a quarter, it's going to really break things because some quarters they'll be up 10, some quarters they'll be down 10. I don't think they might actually be flat. It depends. Up, down, up. But in any case, It's noise because you're looking at a wave and you can't see it because you're getting pieces of it. So again, trailing spend will tend to damp that out, right? If you look at trailing 12 months, it certainly damps it all out. If you even look at trailing— well, trailing 3 won't do it because it's a 5-point cycle. But in any case, the point here, Ray, is you got to be careful that you're not aggregating all these pluses and aggregating all these minuses. And you think you have all expansion and you think you have all this churn. What you actually have is a whole lot of nothing. And if you think it's bad on NRR, it's even worse on GRR because potentially you only aggregate the subtractions, right? Like, oh, because you can't count the expansion. So you have to go, they went down, they went down, they went down, they went down, they went down. So this sequential analysis can be very hard. So that's why I think it basically works over long periods, which is what I think you're saying. I just think you could do it over quarters. It definitely works well over a year. It'll damp out seasonality.

Yeah.

But I think it works over quarters. You could just look at the sequential change quarterly and how much you got from a given customer. I think that's a meaningful metric. If I wanted to call it— the thing I always look for, Ray, is expansion as a percent of new. So, I'd actually just make a table that says, how much growth did I get from existing customers in this quarter? How much did I get from new customers? And by the way, it's going to be bookings, not revenue typically, right? We could do it on revenue. But if you care about sales, we care about bookings. Right? I care about the orders we got and how much of them came from new customers versus existing. And then looking at it on the revenue side is a little bit lagging because you're saying, okay, for those new customers, how much did they actually burn? Which is kind of a different question.

Yeah, totally agreed. Well, I think I'm going to wrap up here on today's episode with if you're usage-based or outcome-based, right? And you're trying to calculate NRR. Best practice, do it on a cohort method. Next best practice is look at the most recent 12 months versus the previous 12 months. So you're doing year over year. And if you do want to do it on quarterly, look at those customers from 5 quarters ago, compare that to those same customers in today's quarter, and that will give you the year over year, but on a quarterly basis, your net revenue retention. So that's kind of my recommendation, Dave.

Yeah, I, I'm not sure about the 5 quarters, right? Because to me, you just use the smooth— the snowflake method on a trailing 8-quarter basis, in effect, right? Take quarters 5 through 8, put them on the bottom, take quarters 1 through 4, put them on the top, and there you go. And you could, you could hang that off either the end of the year, off the end of a quarter, it doesn't change much. Um, and so that 2-year lookback can easily just slide across. But yeah, I think that's it. Um, other than that, I think we're good.

Okay, so that snowflake method, also called the 2-year lookback, definitely the safest way to go if you're in a variable pricing environment. Thanks, Dave. Thanks, everyone.

Thanks, Ray. The Metrics Brothers AI, SaaS, and More is a production of the Metrics Brothers Growth and CAC and a member of the Benchmarkit Podcast Network. By accessing this podcast, you acknowledge that the Metrics Brothers make no warranty, guarantee, or representation as to the accuracy or sufficiency of the information provided Or the humor content of the jokes told. Ray, information, opinions, and recommendations presented are, according to our spouses, probably wrong and provided for information purposes only. This podcast should not be considered professional or, for that matter, unprofessional advice. We disclaim any and all liability for any direct, indirect, undirect, misdirect, incidental, special, ordinary, consequential, inconsequential, or other damages arising out of any use of or God help you reliance upon. Upon the information presented here. Ray Grothreich is based in New York City and available on X at Ray Reich. Dave Kellogg is based in Silicon Valley and available on X at Kellogg. Schenectady, which is French for unspellable, is not our actual production location. Thanks for listening.