When a customer buys prepaid credits, you have their cash but not your revenue. Under ASC 606 the payment is a contract liability (deferred revenue) until credits are used, and the credits customers never use, called breakage, can only become revenue under specific rules. That makes credit-based revenue lumpier, harder to forecast and easier to misread than a subscription, and it quietly inflates gross margin.
Credit models are spreading fast. In the PricingSaaS 500 Index, 79 companies offered a credit model by early 2026, up from 35 at the end of 2024, a 126% increase, with Figma, HubSpot and Salesforce among the new adopters (Rob Litterst in Kyle Poyar's Growth Unhinged, January 2026). Most writing on credits is about the pricing decision: what a credit buys, how to package it, how to stop customers feeling nickel-and-dimed. This page covers the other half, what the decision does to your revenue line, your balance sheet and your forecast.
For the pricing side, start with usage-based pricing and the SaaS pricing models guide. This is not accounting advice; confirm your policy with your auditor.
Credits are deferred revenue, not revenue
ASC 606 is explicit. When you receive a prepayment, you "recognize a contract liability in the amount of the prepayment" for your obligation to deliver, or stand ready to deliver, in the future (ASC 606-10-55-46). Revenue comes later, as credits are consumed. That is the same logic as any deferred revenue, with one important difference.
A subscription's deferred revenue unwinds on a calendar: an annual $12,000 contract releases $1,000 a month whatever the customer does. A credit balance unwinds on behaviour. The same $12,000 of credits might be gone in four months or barely touched in twelve. Your cash, your billings and your revenue now move on three different schedules, and only one of them is under your control.
| Annual subscription | Prepaid credits | |
|---|---|---|
| Cash arrives | Up front (if billed annually) | Up front |
| Revenue is recognized | Ratably over the term | As credits are consumed, plus breakage |
| Deferred balance falls | On a fixed schedule | At the customer's pace |
| What drives a forecast miss | Churn and new sales | Churn, new sales, and burn rate and breakage |
What breakage is, and the two ways to recognize it
Customers rarely use everything they prepay for. ASC 606 calls those unexercised rights "breakage" (ASC 606-10-55-47). Credits that expire are the obvious case: Anthropic's Claude API credit terms, for example, state that credits expire one year from purchase and that purchases are non-refundable. The standard gives two recognition patterns, and which one applies depends on whether you can reliably estimate how much will go unused (ASC 606-10-55-48):
- Proportional method. If you expect to be entitled to a breakage amount, you recognize it "in proportion to the pattern of rights exercised by the customer". In practice, every credit consumed also releases a slice of expected breakage.
- Remote method. If you cannot make that estimate, breakage stays in the liability until the likelihood of the customer using it "becomes remote", typically at expiry.
The estimate is subject to the same constraint as other variable consideration (ASC 606-10-32-11 to 32-13), so you should only book breakage you are confident will not reverse. A new product with no redemption history usually belongs on the remote method until the data exists. One more catch: any amount you are required to remit to a government under unclaimed property laws stays a liability and never becomes revenue (ASC 606-10-55-49). Whether those laws reach your credits depends on the jurisdiction and the terms, so ask counsel.
Worked example: one year of a credit pack
Illustrative inputs. In January, 20 customers each buy a $10,000 pack of 100,000 credits ($0.10 per credit) that expires after 12 months. That is $200,000 of cash and a $200,000 contract liability. Your history across similar cohorts says 15% of credits go unused, so you expect $170,000 to be redeemed and $30,000 of breakage.
Under the proportional method, each dollar of redeemed credits releases $1 ÷ (1 − 15%) of revenue, about $1.18:
| Quarter | Credits redeemed ($) | Revenue, proportional | Revenue, remote | Liability left (proportional) |
|---|---|---|---|---|
| Q1 | $51,000 | $60,000 | $51,000 | $140,000 |
| Q2 | $51,000 | $60,000 | $51,000 | $80,000 |
| Q3 | $42,500 | $50,000 | $42,500 | $30,000 |
| Q4 | $25,500 | $30,000 | $55,500 | $0 |
| Total | $170,000 | $200,000 | $200,000 |
Both methods land on the same $200,000. The difference is timing: the remote method holds the $30,000 back and releases it in one block at expiry, so Q4 jumps to $55,500 while usage is actually falling. If you present that Q4 as growth, you have mistaken an accounting event for demand.
The true-up: how a breakage estimate can sink a quarter
Breakage is an estimate, and you update it every period. When it changes, the correction is a cumulative catch-up, so it all lands in the quarter you change your mind.
Take the same cohort. After Q2 you have redeemed $102,000 and recognized $120,000. Then usage picks up (an AI feature launches and burns credits faster) and you revise expected breakage from 15% down to 5%, so you now expect $190,000 to be redeemed. Cumulative revenue should be $102,000 ÷ 0.95 ≈ $107,368, which means you over-recognized by about $12,632. If Q3 redemptions are $50,000:
- Q3 revenue = $50,000 ÷ 0.95 − $12,632 ≈ $40,000
- Q4 (redemptions $38,000) = $38,000 ÷ 0.95 = $40,000
Customers used more of the product in Q3, and reported revenue fell from $60,000 to $40,000. The year still totals $200,000. Nothing went wrong with the business; the estimate moved. This is why a credit business needs its breakage assumption on the same review cadence as its pipeline, and why you should explain the catch-up before an investor finds it.
Breakage flatters your gross margin
This is the part most write-ups miss. Breakage is revenue with no cost behind it. You never served those credits, so there is no inference, compute or support cost against them.
Say each credit costs you $0.04 to serve. In the example, 1.7 million credits are redeemed, so cost of revenue is $68,000:
- Gross margin on reported revenue: ($200,000 − $68,000) ÷ $200,000 = 66%
- Gross margin on redeemed revenue only: ($170,000 − $68,000) ÷ $170,000 = 60%
Six points of margin come from customers not using what they paid for. That matters twice. First, it is not durable: if usage rises, as in the true-up above, the margin goes with it. Second, when you compare yourself with SaaS gross margin benchmarks or with the AI product margins in ICONIQ's data, compare the redeemed-only figure. On an AI product, where every credit carries real model cost, the gap can be wide. The calculator at the bottom of this page models cost per customer by usage band, so you can see what heavy users do to it.
High breakage is a churn signal, not a win
A finance team can be tempted to see breakage as free money. Treat it as a leading indicator instead. A customer who used 60% of last year's pack has a simple renewal decision: buy a smaller pack. Breakage revenue today often becomes contraction next year, and it pulls down net revenue retention with a lag. Track utilization by account and flag low burners for customer success well before expiry.
Rollover is the opposite lever. Snowflake lets customers roll unused capacity into a new order, generally when they buy more. That protects the relationship but defers revenue further. Snowflake's FY2026 10-K reports $9.8 billion of remaining performance obligations at January 31, 2026, with only about 46% expected to be recognized in the next 12 months "based on historical customer consumption patterns", and states plainly that its deferred revenue "is not a meaningful indicator of future revenue". If a public company with years of data says that, a startup's deferred balance tells you even less without a burn model behind it.
How to forecast credit revenue
A seats-times-price model cannot handle credits. You need to forecast consumption, by cohort. The minimum driver set:
- Packs sold per month, by size. This gives billings and cash, not revenue.
- Burn curve. The share of a pack consumed in each month after purchase, from your own cohort history. Early-life burn is usually fast, then tails off.
- Expected breakage by cohort. Different pack sizes and segments break at different rates. Enterprise commits often break less than self-serve top-ups, or more if they were over-sold.
- Expiry and rollover terms. When the remote-method breakage lands, and how much is pushed into new contracts.
- Cost per credit. For AI products this is mostly inference, and it can change with your model mix.
Then, for each cohort and month: revenue = credits consumed × price per credit ÷ (1 − expected breakage), with any change in the breakage estimate booked as a catch-up in the month it changes. Deferred revenue at month end = cumulative billings − cumulative revenue. FP&A practitioners also track credit utilization (consumed ÷ allocated) and days to credit exhaustion as early signals of upgrades and renewals (Isha Sharma, FP&A Trends).
Run at least two scenarios: one where burn speeds up (breakage falls, margin compresses, revenue pulls forward) and one where it slows (breakage rises, revenue looks better than demand, renewals shrink). The gap between them is the real uncertainty in a credit business. A single-line plan hides it. See revenue forecasting methods for the bottom-up approach this plugs into.
Common mistakes
- Booking credit sales as revenue. Cash is not revenue. The pack is a liability until used.
- Using the proportional method without history. If you cannot support the estimate, the constraint pushes you to the remote method.
- Mixing free and paid credits. Promotional credits granted at no charge are not a prepayment and do not create breakage revenue. Keep them in separate ledgers or your burn data is polluted.
- Reporting gross margin with breakage in it, without saying so. It overstates unit economics and will not survive due diligence.
- Calling credit billings ARR. A one-off top-up is not recurring. Split committed revenue from consumption revenue when you report ARR.
Credit-based pricing FAQ
Are prepaid credits revenue?
Not when sold. Under ASC 606 a prepayment is a contract liability (deferred revenue) until the credits are used. Revenue is recognized as they are consumed, plus any breakage recognized under the rules for unexercised rights.
What is breakage revenue?
Revenue from prepaid rights customers never use, such as expired credits or unredeemed gift cards. ASC 606-10-55-48 lets you recognize expected breakage in proportion to usage if you can estimate it reliably, or when the chance of use becomes remote if you cannot.
When can you recognize unused credits as revenue?
With reliable redemption history, gradually, as other credits are consumed (the proportional method). Without it, when redemption becomes remote, usually at expiry. Amounts owed to a government under unclaimed property laws are never revenue.
How do you forecast credit-based revenue?
By cohort: packs sold, a burn curve for how fast each pack is consumed, expected breakage, expiry and rollover terms. Revenue each month is consumed value divided by one minus expected breakage, with estimate changes booked as a catch-up.
Does breakage affect gross margin?
Yes. Breakage has no cost of delivery, so it raises reported gross margin. In the example above it adds six points (66% versus 60%). Benchmark on redeemed revenue to see the margin the product actually earns.
Do monthly credit allowances work the same way?
Mostly simpler. When included credits reset each month and expire with it, the unused portion usually settles within the same reporting period, so the timing question largely disappears. Rollover allowances bring it back.
Related: deferred revenue, usage-based pricing, outcome-based pricing, AI COGS and gross margin, bookings vs billings vs revenue.