How usage-based billing changes MRR, churn, NRR, LTV and other SaaS metrics, with formulas and the usage-specific metrics worth tracking.
With usage-based billing, customers pay for what they use: API calls, tokens, orders, messages or storage. It's a natural way to let revenue grow as customers consume more value. It also makes traditional SaaS metrics a little trickier. Bills arrive after consumption, spend changes from month to month, and customers can stop using a product without ever cancelling. Here's how to make sense of the metrics you already know, plus new usage-specific ones worth watching.
Pricing vs. billing: usage-based pricing defines what you charge for and how much. Usage-based billing turns that into a bill: measuring usage, calculating charges, and issuing invoices. Section 1 covers pricing; the rest focuses on billing and the revenue it generates.
How to read this sheet: for each classic metric you get the standard definition, what changes with usage-based billing, and how to calculate it. Pair it with our SaaS metrics cheat sheet for the baseline definitions.
1. Understanding usage-based billing
Pricing models
| Model | How it works | Example | Predictability of billed spend |
|---|---|---|---|
| Pay-as-you-go | Customer is billed per unit consumed, in arrears | $0.002 per API call | Low |
| Tiered (graduated) | Each usage band is priced at its own rate | First 10k units at $0.01, next 90k at $0.008 | Low |
| Volume | All units are priced at the rate of the band the customer reaches | 100k units → every unit at $0.008 | Low |
| Hybrid (subscription + usage) | Fixed subscription fee plus metered charges, often as an overage above an included allowance | $150/month for 1,000 orders, then $0.10 per extra order | Medium |
| Prepaid credits | Customer buys credits upfront, consumes them, and tops up | 10,000 AI credits for $1,000 | Medium |
| Committed spend (drawdown) | Customer commits to an amount for the contract term, consumes against it, and pays overages above it | $120k annual commitment, overage at list price | High |
| Outcome-based | Customer pays per result delivered rather than per unit of activity | $1 per lead generated | Low |
Predictability depends on who your customers are, how they use your product, and what they're willing to commit to paying. Prepaid credits, for example, can make cash flow more predictable without making consumption more predictable.
Choosing a value metric
The value metric is the unit you charge for. A good one scales with customer value, is predictable enough for a buyer to budget, is easy to meter and audit, tracks your cost to serve, and is hard to game.
Three ways to look at revenue
These numbers answer different questions about your revenue. They overlap, so adding them together would count the same revenue more than once.
| Concept | What it answers | How to calculate |
|---|---|---|
| Usage revenue | How much revenue did consumption generate in the period? | Billable units consumed × price per unit |
| Usage MRR | What monthly run rate do we assign to the customer's latest billed consumption? | See Usage MRR |
| Contracted commitments | How much has the customer contractually agreed to spend? | Fixed subscription fees + minimum usage commitments + prepaid usage commitments |
Example: a customer commits to spending $120,000 on usage over the year. They use $8,000 in January and $12,000 in February. February's usage revenue is $12,000. That counts toward the $120,000 commitment; it isn't an additional charge.
Overages are charges for usage beyond a commitment or included allowance, often at a higher rate. They count as usage revenue and can signal that a customer is ready for a larger commitment.
Prepaid credits are cash received, not revenue earned. They sit in deferred revenue and are recognized as customers consume them. Unused credits (breakage) need a clear policy.
2. Growth metrics
Usage MRR
Standard: MRR is normalized monthly subscription revenue.
What changes: usage is billed in arrears, so you only know what a customer consumed once the period's usage invoice arrives. There's no single standard for turning that into MRR, and some businesses exclude variable usage from MRR entirely. One common method, and the one this sheet uses, treats the latest billed usage as the customer's current run rate until the next usage invoice replaces it.
Think of usage MRR as an estimated recurring revenue. It tells you what recent usage would look like as monthly recurring revenue, without promising that customers will spend the same amount next month. Because the last completed period carries forward, it can also differ from the usage revenue earned in the current calendar month.
How to calculate: add up all usage charges billed for the billing period, then normalize by the plan interval.
Examples
- $200 of usage one month, $20 the next: usage MRR drops from $200 to $20.
- Threshold billing can produce a mid-cycle invoice plus a cycle invoice for the remainder. They belong to the same billing period, so they add up. A genuinely new billing period starts a new total.
- A $120 out-of-cycle usage charge adds $120 MRR on a monthly plan and $10 on a yearly one ($120 ÷ 12).
- A $500 overage billed for a single day is $500 added to the period, not $500 × 30 days = $15,000 MRR. Never extrapolate a partial-period charge.
Separating spikes from baseline: usage MRR picks up every spike. A migration or a big campaign might make an account look like it's expanding, even if its usual consumption hasn't changed. To see the underlying baseline, compare usage MRR with trailing 90-day usage revenue.
ARR
Standard: ARR is MRR × 12. The formula stays the same, but ARR fluctuates each month as usage changes. Look at how it develops over several months rather than relying on one month's figure.
Usage MRR share
New metric. How much of your recurring revenue comes from consumption, and which direction it's trending. A rising share means more expansion upside, and more volatility.
For the share of revenue actually earned in a period, use usage revenue ÷ total revenue from your cash flow or revenue reporting instead.
ARPA
Standard: ARPA is MRR ÷ paying customers.
What changes: a handful of heavy users can pull the average up. Check the median and the spread too, so you can see what a typical customer actually spends.
CMRR
What changes: committed MRR (CMRR) is usually the most predictable part of a usage-based business, though how predictable depends on contract terms: the recurring revenue customers are contractually bound to, excluding variable usage above it. Track it alongside usage MRR.
Bookings
What changes: bookings capture the total value of commitments when a deal is signed. A $360k three-year commitment is $360k in bookings on day one, even though it's consumed and recognized over three years. Don't confuse it with CMRR or usage revenue.
MRR movements
What changes: the standard movement types still apply, they just happen more often.
- Expansion: usage MRR is higher than last period.
- Contraction: usage MRR is lower, or usage stops while the customer keeps paying for another subscription.
- Churn: usage stops and the customer has no other paid subscription (see Retention metrics).
- Reactivation: a churned customer starts generating usage again.
An increase or decrease in usage counts as expansion or contraction in the metric. That doesn't mean the customer's behavior has changed for good. Look for changes that persist across several periods before drawing conclusions. Segment movements by subscription type (usage-based vs. fixed) to see whether growth comes from consumption or from plan changes.
3. Retention metrics
Revenue loss, inactivity and customer churn
What changes: a customer can stop generating usage without cancelling anything. There are three different things that might be going on:
- Revenue loss: the customer generated no billable usage, so their usage MRR drops to zero.
- Inactivity: the customer is temporarily not using the product. They might run quarterly batch jobs, have seasonal demand, or be between projects.
- Customer churn: the customer has left the product altogether.
No usage could mean "we're done" or simply "not right now." You often only learn what happened when the customer starts using the product again, or stays away. Some teams track temporarily inactive (dormant) customers as a separate state; the right approach depends on how often customers are expected to use the product.
How this sheet treats it: when a usage subscription's billed usage drops to zero (a zero-amount usage invoice, or no usage invoice within a full billing period), its usage MRR is counted as churned, even without a cancellation. That's churn under the calculation method, not proof the customer has left.
- If it was the customer's only paid subscription, it counts as customer churn.
- If the customer still pays for another subscription, it's a contraction.
- If usage resumes later, it's a reactivation.
Spiky accounts can cycle through churn and reactivation, so read churn alongside reactivation.
NRR and GRR
Standard: revenue retained from a fixed starting cohort of customers, usually over 12 months. NRR includes expansion; GRR excludes it and can never exceed 100%.
What changes: expansion happens organically as customers consume more, so usage-based companies can post high NRR, but contraction is just as automatic, which makes it volatile. GRR becomes your clearest measure of how sticky consumption is.
Always calculate both on the same starting cohort, not from company-wide usage movements, which mix in new customers. Calculate them for usage MRR on its own as well as for total MRR.
Usage cohorts
Cohort by signup month as usual, and add three usage views:
- Usage revenue per customer over time, to see whether accounts deepen their consumption.
- Consumption volume retention, using the NRR formula but replacing revenue with units consumed. Track each usage meter separately to distinguish changes in consumption from changes in price.
- First-month usage level (light, medium, heavy), to see which starting behavior predicts long-term expansion.
Benchmark context: many AI products are billed on usage, and their retention looks very different from classic B2B SaaS. ChartMogul's AI churn wave report found median NRR for AI-native and consumer products far below the B2B SaaS median.
4. Usage-specific metrics
Usage adoption rate
What share of customers who can generate billable usage actually do. Define the denominator to match your model: subscribers on plans with a usage component, plus free accounts if they can incur billable usage.
Commitment utilization and consumption pace
For committed-spend models, utilization tells you how much of the commitment has been consumed; pace tells you whether the customer is on track to consume it by the end of the term. Measure both in the same units (currency, credits or volume).
Example: a $1M annual commitment with $200k consumed after eight months has 20% utilization with 66.7% of the term elapsed, so it's consuming at 30% of the expected pace. That's a warning sign for the renewal and worth a customer success conversation, though not a certain downsell. A pace above 100% points to an overage or upsell conversation.
Revenue growth vs. consumption growth
Revenue growth doesn't always mean customers are using more. Prices may have changed, or customers may be moving between cheaper and more expensive units. Track consumption and revenue per unit alongside usage revenue to see which is happening.
| Month 1 | Month 2 | |
|---|---|---|
| Units consumed | 1M | 1.5M |
| Revenue per 1M units | $100 | $80 |
| Usage revenue | $100 | $120 |
Consumption grew 50%, but revenue grew only 20% because the price per unit fell (for example if the customer reached a cheaper volume tier).
Leading indicators
Keep volume metrics tied to the unit you're billing. Tokens and orders tell different stories, so combining them won't help.
| Metric | What it tells you | How to calculate |
|---|---|---|
| Active consumption accounts | How many customers are actually using what they pay for | Count of accounts with billable usage greater than 0 in the period |
| Consumption volume | The underlying usage that drives revenue | Sum of billable units consumed, per meter |
| Consumption per account | Depth of adoption and expansion potential | Consumption volume ÷ active consumption accounts |
| Revenue per unit | Pricing efficiency, discounting and mix shift | Usage revenue ÷ consumption volume (per thousand or million units if units cost fractions of a cent) |
| Activation rate | Share of new accounts reaching meaningful usage | Accounts passing an activation threshold ÷ new accounts |
| Time to first billable usage | How quickly new accounts start consuming | Median days from signup to first billable usage |
| Revenue concentration | Exposure to a handful of heavy users | Usage revenue from top 10 accounts ÷ total usage revenue |
| Free credit exhaustion | How often free allowances run out: a conversion trigger | Accounts using up free credits ÷ accounts with free credits |
| Prepaid balance depletion | How often paying customers run down their balance: a top-up and expansion trigger | Accounts depleting prepaid balance ÷ accounts with prepaid balance |
5. Efficiency metrics
Gross margin
Standard: gross margin is the share of revenue left after the direct costs of delivering the product.
What changes: gross margin moves with usage, especially when each unit carries real infrastructure or inference cost. Track it per unit, and watch cost to serve per account: your heaviest users on discounted rates can be your least profitable. For AI products, some finance leaders also track revenue per compute dollar (revenue ÷ inference and infrastructure cost), the compute equivalent of revenue per employee.
LTV
Standard: LTV is ARPA × gross margin ÷ customer churn rate. Dropping gross margin gives revenue LTV, the version tested in ChartMogul's LTV research.
What changes: LTV can look more precise than it really is. ChartMogul's research found that its biggest source of error is using blended ARPA for new customers, who typically pay less than the average, while expansion later pushes actual revenue above the prediction. Usage-based billing can introduce similar challenges, particularly when customers ramp up consumption over time, and consumption can grow, contract, spike or stop independently of the subscription.
How to use it: use LTV as a guide, not an exact forecast. Check it against what cohorts actually generated, using cohort revenue curves and cumulative gross profit per cohort, and use net MRR movements to see what drives the difference.
CAC payback period
Standard: CAC payback is the months of gross margin needed to recover acquisition cost.
What changes: usage customers often start small and ramp up. A payback estimate based on first-month revenue will overstate payback if they ramp up, and understate it if their usage declines. Track payback per cohort as the month when cumulative gross profit per customer exceeds CAC.
Quick Ratio
Standard: Quick Ratio is (new + expansion + reactivation MRR) ÷ (churned + contraction MRR).
What changes: usage expansion and contraction flow into the same ratio every month, so it's noisier. Read the trend rather than a single month.
6. Common mistakes
- Treating a missing renewal invoice as past due: usage is billed in arrears, so the invoice arrives after the period. Only an unpaid invoice is past due.
- Extrapolating a partial-period charge: a one-day overage is not a month's recurring revenue.
- Double counting: commitments, usage revenue and overages overlap. Don't add them together.
- Reading one month as a trend: usage MRR moves every period by design.
- Equating zero usage with a lost customer: it's lost revenue, but the customer may come back. Track reactivation alongside churn.
- Mixing usage units with seat counts: two million tokens is not two million seats.
- Ignoring unit margin: revenue can grow while margin shrinks if heavy users sit on discounted rates.
- Choosing a value metric customers can't predict: unpredictable bills slow procurement and drive churn. Usage alerts, budgets and caps help prevent bill shock.