A four-step framework for CFOs to control AI token costs.

Read Now

Usage-based pricing models and revenue forecasting featuring Ray Rike

Last updated: November 202660 min watch

Speakers

  • Ray Rike, Founder of Benchmarkit

  • Kirk Kappelhoff, Senior Director of Strategic Finance, Drivetrain

A usage-priced customer with no minimum commitment budgeted $50K for your product. Its bill reached $60K, then climbed toward $120K, and now its CFO has told your account manager, "We have to consider options."

Ray Rike, founder and CEO of Benchmarkit, knows where that conversation leads. At the usage-priced company where he started his career, the largest customers were the most likely to leave. They were spending more than they had planned and didn't feel comfortable raising it with the vendor.

In this webinar, Ray walks Drivetrain’s Kirk Kappelhoff through how finance teams should count and forecast revenue that moves with usage. Watch the full session below, or read on for his working definitions of ARR, expansion, NRR, and CAC for usage pricing, a threshold-based approach to moving large usage-based customers onto commitments, and a customer-by-customer method for forecasting usage.

60:25

Why does usage revenue grow faster and fall faster?

Usage-based revenue rises and falls with the billable units your customers run through your product, such as transactions, messages, or API calls. When a customer’s business grows, it sends more of those units through your product, and your revenue grows at the same rate. Ray illustrates this with a vendor that bills Uber per transaction. If Uber’s transactions grow 300%, the vendor’s revenue from Uber grows 300% too.

Usage-based pricing also makes it easier to win new customers. With no minimum commitment, a customer can start paying for what it uses without signing a large upfront contract, and Ray sees pay-as-you-go pricing as one of the fastest ways for an early-stage company to add logos. In the benchmarking research he has reviewed, many early-stage companies use usage as their primary pricing model, mostly with no minimum commitment, and they tend to grow about twice as fast as companies on subscription or seat-based pricing. In strong markets, large companies with usage-based pricing models, such as Snowflake, Datadog, and Twilio have posted net revenue retention of 140% to 150%.

The decline is faster for the same reason. A customer can throttle usage as soon as its budget tightens, while a customer with 100 subscription seats keeps paying until the contract ends. In the billing data Ray has reviewed, growth at usage-priced companies slows sooner and more steeply than at companies with subscription-based pricing when the economy turns. Subscription-based companies feel the slowdown later, when contracts come up for renewal. Companies with usage-based pricing also recover first, so Ray watches their growth as an early signal that the market is turning.

Industry concentration can make the drop steeper. Ray’s example is ZoomInfo. Ray’s example is ZoomInfo, which priced partly on contacts used and drew about 40% of its revenue from early-stage B2B technology companies. No other industry made up more than 6% to 8%, so when those technology companies stopped buying, nothing offset the drop and ZoomInfo’s growth collapsed. Ray’s advice is to track the economic conditions in each industry your revenue depends on.

What counts as ARR when part of the revenue is variable?

Ray recommends reporting annual recurring revenue (ARR) in two ways: ARR from committed subscriptions only, and ARR that also includes variable usage. Prepaid usage counts as committed. A customer that buys a million transactions upfront has signed a minimum commitment, so Ray would count that purchase in both ARR and bookings. A company showing $22M of committed ARR and $28M of variable ARR tells investors how much of its $50M could fall next quarter.

The SaaS Metrics Standards Board, which Ray co-founded, had to work out how usage-based pricing changes standard SaaS metrics, including whether usage revenue counts as ARR at all or only as GAAP revenue recognized as the usage happens. Its recommendation is to start with the chart of accounts, putting committed subscription revenue and variable revenue on separate lines.

Most companies with pricing models that include a usage-based component blend the two. Research Ray cited shows that almost three-quarters of them convert monthly usage revenue into ARR. He considers this a dangerous practice because investors can’t see how much revenue is at risk. It also distorts gross revenue retention.

...if you include variable usage-based revenue in your ARR and the usage goes down, gross revenue retention could go from 85% down to 72%...I didn’t really have churn; I just didn’t have as much usage.
Ray RikeFounder and CEO, Benchmarkit

What counts as expansion, and what belongs in NRR?

Ray recommends defining how long a new customer takes to ramp, counting all its revenue during the ramp as new, and counting revenue from growth after the ramp as expansion. In subscription software, the contract draws that line clearly: the first annual contract is new revenue, and any increase from month 13 is expansion. Usage-based pricing has no contract term to mark the line, and almost every company’s new customers ramp up, sometimes to 5, 10, or 20 times the revenue they started at.

The ramp might last one, three, six, or twelve months, depending on the product and the customer base. With a six-month ramp, expansion starts in month seven. Ray recommends using the same number of months for every customer. Where you draw the line also changes NRR. If you start counting expansion before customers finish ramping, their normal ramp-up is counted as expansion, and NRR looks higher than their real growth after onboarding.

NRR must also include downsells and churned customers. As an example, Ray recalled an earnings call where Snowflake reported 160% NRR, meaning its existing customers were spending 60% more than a year earlier. About 20 minutes later, it presented its forecast for total revenue growth as 55% for the next quarter. An analyst asked how existing customers could be growing 60% while the company grew only 55%, and whether Snowflake was “paying new customers to become new customers.” As it turned out, Snowflake’s NRR had left out the customers who churned.

To analyze ramp revenue, Ray suggests using a separately named measure, net expansion rate, which compares a cohort’s ARR or GAAP revenue over the last 12 months with the same customers’ revenue in months 13 to 24. Because net expansion rate leaves churned customers out, so Ray treats it as a complement to GRR, which counts churn and downsells but not expansion. Together, they show both how much revenue you kept and how fast the customers who stayed are growing. Ray has seen several companies use the same calculation, under the name “two-year look-back,” as a proxy for net expansion rate.

How do you measure CAC payback when new revenue isn’t committed?

Ray says to choose the revenue basis for CAC payback and the CAC before you calculate them. CAC payback and CAC usually compare new ARR with sales and marketing expense. Payback is typically calculated after adjusting the new ARR for gross margin. But a small commitment followed by a six-month ramp leaves no settled new-ARR figure. Ray's default is to use the first 12 months of contract value, though three or six months can work if applied consistently.

The expense side can reflect how your deals are sold. Depending on your sales cycle length and average contract value, a company might take marketing expense from the previous quarter and sales expense from the current one, so each cost lines up with the customers it produced. Ray’s rule is to choose a time lag that fits your business and apply it the same way every period.

For land-and-expand deals, Ray offers a different option. Say an automotive customer commits $10K, and customers of its size in the automotive industry typically spend $100K in their first 12 months. Calculate CAC payback on that $100K of annual projected revenue, then recalculate it on actual GAAP revenue 12 months later. A projected 23-month payback might come in at 21 or 28 months, which holds the leadership team accountable for its forecasts. Ray recommends running both calculations by customer segment, because enterprise and SMB customers differ in acquisition cost and projected revenue.

So I strongly recommend you do both a booking/forecast model and an actual GAAP revenue model.
Ray RikeFounder and CEO, Benchmarkit

When and how should you move usage-based customers onto commitments?

Ray suggests moving usage-based customers onto commitments once your company has grown to the point where investors want predictable revenue, and once a customer’s spend has passed a threshold your company sets. Even then, the commitment works as a floor. The customer agrees to a minimum, usually in exchange for a lower unit price, and keeps paying for usage above it.

Ray doesn’t treat commitments as the goal for every customer. In his research, few companies charge on usage alone with no subscription or minimum fee. A minimum commitment with usage charges on top is far more common. He also noted that the goal for how many customers to move onto commitments differs from company to company.

The push for commitments usually comes with scale. As a company grows, investors start asking what revenue will look like next quarter and over the next 12 months. That’s when Ray expects the company to start asking its larger customers for commitments. Twilio followed that path, sending enterprise salespeople to offer a lower unit price in exchange for minimum commitments of $100K or $500K.

…History repeats itself. Early days, less friction, less commitment. As you scale, more commitment, lower unit economic cost.
Ray RikeFounder and CEO, Benchmarkit

Customers often prefer to skip the commitment at first. Early in Ray’s career, his company’s customers included Dell, Sun, Ford, and Toyota, and they liked not having to commit upfront. Two or three years later, their bills had kept rising, and their CFOs couldn’t budget for them.

Ray suggests that your company set spend thresholds, such as $10K, $25K, and $50K, and have account managers offer each customer a lower unit price in exchange for a minimum commitment as it crosses one. The offer heads off a budget surprise before the customer’s CFO starts weighing options. To make the conversions happen, Ray recommends giving the account management team a compensated goal: convert a share of usage-based customers to minimum commitments worth at least 80% of what each billed over the previous year.

How do you forecast usage you don't control?

To forecast usage-based revenue, Ray recommends starting with each customer. Forecast usage for every customer and every customer cohort, from the bottom up and the top down, for the next three months and for each of the next four quarters. Ray calls the result annual projected revenue.

Earlier in his career, Ray held sales, account management, and finance jointly accountable for the forecast. And today, he recommends naming an owner for each forecast, such as the account manager or an FP&A analyst covering a region.

Investors will ask how accurate those forecasts are. Ray noted that the average company’s new-business forecast is already off by 11% to 15%.

Ray said this kind of forecast accounts for four drivers that vary by segment: industry trends, each customer’s business, seasonality, and how far each customer is into its ramp. If mortgage refinancing is expected to grow 30% next year, that growth belongs in the mortgage segment’s plan and quotas. If you charge retailers a percentage of their sales, your revenue from them will peak in Q4.

If I'm selling to OpenAI right now with a variable-based product, I'm pretty sure it's gonna grow a lot faster than if I'm selling to Salesforce.
Ray RikeFounder and CEO, Benchmarkit

Ray distinguishes cohorts, which are grouped by start date, from segments, which are defined based on customer type (e.g., enterprise in the US or mid-market in Europe). He recommends forecasting each cohort at least one or, ideally, two levels lower by both segment and product type, using trends from the last one to eight quarters. He acknowledged that this approach can get difficult to manage in Excel, so he advises building the capability when you first introduce usage-based pricing.

Billing cadence carries the same variability into cash flow forecasting with usage-based pricing. Most companies in Ray's research bill overages monthly, with comparatively few billing quarterly or annually.

How do you build these definitions into your model?

Ray’s definitions only help if your model applies them the same way every period: the ramp length, the revenue basis for CAC payback, the time lag between expense and new customers, and the split between committed and variable ARR. That takes a model that pulls data from your ERP, billing, CRM, and product analytics systems, and lets you change a definition in one place without rebuilding the calculation.

How do you change a metric definition without rebuilding it?

In the demo, Kirk built CAC in Drivetrain from two systems: sales and marketing expense from NetSuite and new customers from Salesforce. Both flowed into Drivetrain and lined up by period without manual matching, and Ray noted that getting that source data is the biggest issue with these metrics.

Kirk then showed how to adjust the definition: shifting expense to the previous month or quarter, limiting it to the sales department, and filtering new customers by vertical or region.

How do you build the cohort forecast?

Kirk called a waterfall cohort analysis the best way he's seen to forecast usage. It takes two inputs: the number of new customers expected in each monthly cohort, usually estimated from history, and a retention curve that defines how each cohort behaves after its first month.

Every new cohort starts the curve over from its first month, so at any point the forecast is the sum of several cohorts, each at a different stage of the same curve. The curve can also rise above 100%, which lets the same model forecast usage revenue from customers who stay and grow.

Kirk noted that the same waterfall can be built in Excel, but it usually takes nested IF and INDEX-MATCH formulas that are hard to update. In Drivetrain’s revenue forecasting tool, the waterfall is a single expression that multiplies each new cohort by the retention curve. To plan conservatively, a scenario can swap in a lower retention curve. If you’d like to see that forecast run on your own customer data, book a demo.

About Ray Rike

Ray Rike is founder and CEO of Benchmarkit, a SaaS benchmarking firm whose database covers 18,000 B2B SaaS companies. He spent the first ten years of his career, in the 1990s, at a provider of mainframe-hosted applications priced on usage. Over the following 25 years, he worked at venture-backed subscription software companies with five exits: two to strategic buyers, two to private equity, and one IPO. He co-founded the SaaS Metrics Standards Board. Ray also joined Drivetrain for a webinar on investor vs. operator metrics.

Summary

  • Usage-priced revenue grows faster than subscription revenue while customers’ businesses are growing and falls faster when they cut back because it moves with how much each customer uses your product, and customers can throttle usage at any time.
  • To predict revenue with unpredictable usage-based pricing, Ray recommends reporting committed ARR separately from ARR that also includes variable usage, and offering large usage-based customers a lower unit price in exchange for a minimum commitment.
  • Ray recommends that companies with usage-based pricing set a ramp period for new customers, count revenue growth during the ramp as new ARR, count growth after the ramp as expansion, and include downsells and churn in NRR.
  • CAC payback and the CAC ratio both need a defined basis for new revenue. For land-and-expand deals, Ray suggests calculating CAC payback by segment twice: once on projected revenue at signing and again on actual GAAP revenue 12 months later.
  • Ray recommends forecasting usage customer by customer and cohort by cohort, bottom-up and top-down, with an account manager or FP&A analyst owning each forecast, because each customer’s usage depends on its business, industry, its seasonality, and how far into its ramp period it is.
Summarize with AI:

Questions from the webinar

Does usage-based pricing earn a higher valuation multiple than fixed-license pricing?

Ray’s answer is no. Among public SaaS companies, growth has the strongest effect on enterprise-value-to-revenue multiples. Next comes a minimal level of profitability, which Ray tied to the Rule of 40, and then NRR. Rule of X research that Ray attributed to Bessemer and Meritech Capital found growth has 2.3 times the impact of free cash flow margin. A pricing model affects valuation through its effect on those three measures.

How do you decide when usage revenue from customers who are still ramping can be counted as ARR?

Ray ties the decision to the ramp period. If the ramp typically lasts three months and about 90% of customers are 80% to 100% ramped by then, converting their usage revenue to ARR at that point is defensible. Ray’s own preference is to report usage as GAAP revenue and keep it out of committed ARR until the customer signs a minimum commitment.

How do you show new ARR in an ARR bridge while new logos are still ramping?

Ray suggests a separate report for customers in their ramp period, alongside the standard ARR bridge. It shows where each customer is today and where you project it to be over the next one to four quarters, with customers grouped by cohort.

What holds companies back from testing usage-based pricing?

The most common reason is customer pushback, according to Ray. Customers worry they won't be able to budget for usage and that their costs will rise unless the vendor protects them from overruns. Ray's advice is to talk to customers before choosing a usage-based pricing model, then test it with a new customer segment or a new product. Citing Chris Mele of Software Pricing Partners, he warns against starting with existing subscription customers.

How does Drivetrain support usage-based revenue forecasting?

Drivetrain pulls metric inputs from separate systems, such as expenses from NetSuite and new customers from Salesforce, and lets you change a metric's time period or dimension without rebuilding it. Its cohort forecast multiplies each new cohort by a retention curve that restarts with every cohort and can be swapped for a lower curve in a conservative scenario.

Speaker spotlight

Ray Rike

Founder of Benchmarkit

Ray Rike is the founder & CEO of Benchmarkit and one of the most respected voices in SaaS metrics. With 20+ years in B2B SaaS, Ray has worked across operating and investment roles, and now leads Benchmarkit’s data-driven insights for SaaS leaders. His research has been featured by OpenView, TechCrunch, and SaaStr.

Kirk Kappelhoff

Senior Director of Strategic Finance, Drivetrain

Kirk Kappelhoff is a financial modeling expert with a BBA in Finance & Accounting and nearly a decade of experience at Deloitte, EY, and KPMG, where he built models for pre-IPO companies, M&A transactions, and strategic planning initiatives. At KPMG, he led the Business Modeling Services team, specializing in equity stories and financial forecasts that help high-growth companies communicate their value to investors. At Drivetrain, Kirk writes about strategic planning, granular reporting, and modern FP&A best practices for rapidly scaling finance teams.

Ready to start your journey?

Fewer spreadsheets, faster planning, better decisions.

Book a Demo