Credit-Based Pricing Explained

By Brian Tr · Updated Oct 5, 2026

Credit-based pricing means a customer buys a pool of abstract units up front, then spends them as they use the product. A short action costs a few credits. A heavy action costs many. The customer never sees the underlying compute bill, only a balance that goes down.

It looks like a small packaging choice. It isn't. Credits change how you recognize revenue, how you control cost, how buyers feel about the invoice, and how easily you can change prices later without a renegotiation. This guide covers what a credit system is, why AI products reach for it, the design decisions that matter, and the signals that tell you whether yours is working.

It's written for founders and owner-CEOs of B2B companies with a team. Where we use numbers in examples, they're round and labelled illustrative. They aren't data.

What a credit system actually is

A credit is a unit of account that you define. The customer pays real money for a bundle of credits, and the product deducts credits when the customer takes actions. Three parts make up every credit system:

• A currency. The unit itself, with a price per credit or per bundle.
• A rate card. The table that says how many credits each action burns.
• A wallet. The balance, with rules for what happens when it runs low, runs out, or expires.

Think of an arcade. You hand over cash and get tokens. Some machines take one token, some take four. The arcade can change what a machine costs without repainting a price tag on every game. That flexibility is the heart of the model, and it's also the source of most complaints about it.

Credits aren't the same as the broader family of usage pricing, though they overlap. We'll compare them properly below.

Why AI products adopt credits

Classic software has a near-zero cost to serve one more action. AI products often don't. Our overview of AI startup business models explains why delivery carries a variable cost, and credits are one of the main ways founders respond to it. Four reasons come up again and again.

Actions have wildly different costs. Summarizing a paragraph and running a multi-step research task are both "one use" to a customer. To you they can differ by an order of magnitude. A flat per-action price either overcharges the small task or loses money on the big one. A rate card in credits lets each action carry its own weight.

One currency spans many features. If you ship five AI features, five separate meters get confusing fast. Credits give the customer a single balance and let you add a sixth feature without a new line item on the contract.

Prepaid cash helps the company. A customer who buys credits pays before consuming them. That improves your cash position and, as we'll see, creates an accounting liability that you release over time.

Pricing becomes adjustable. Your underlying costs will move. If you've sold "1,000 credits" rather than "1,000 document analyses," you can change how many credits an action burns when your cost to deliver it changes, without rewriting the price list. Buyers may notice the change, so this lever has to be used carefully.

The model is spreading. Nalpeiron's 2026 State of B2B Software Monetization study, which surveyed 255 B2B software companies in June 2026, reports that 52% use AI credit or token bundles to price AI features. That's one vendor's survey of one sample, so read it as a signal of how common the practice has become, not a precise market share.

The core design decisions

Most of the work in a credit system is in the details. Five decisions shape how it behaves.

1. The conversion rate

How many credits does a dollar buy, and how many credits does each action burn? Two traps live here.

The first is a rate card so detailed that nobody can predict their own spend. If the customer needs a spreadsheet to estimate a month, you've moved the complexity from your cost structure to their desk.

The second is a rate card with false precision. Pegging one action to 0.37 credits looks rigorous and reads as noise. Round numbers that map to something a user can picture, such as "a standard task is 1 credit, a deep task is 5," are easier to trust.

An illustrative example: suppose a bundle of 1,000 credits sells for $100, so a credit is ten cents. A quick action burns 1 credit (ten cents), and a long, multi-step action burns 20 credits ($2). If the long action costs you $1.20 to deliver, your gross margin on it is 40%. If the quick action costs you four cents, it's 60%. The rate card is where you set those margins on purpose rather than by accident. For the margin math itself, see gross margin.

2. Rollover and expiry

Do unused credits carry over, or do they vanish at the end of the period? This is the most emotionally charged choice in the model.

Expiry protects your forecasting. Credits that disappear can be recognized as revenue, and you can plan capacity around a known monthly allotment. Rollover is friendlier to buyers, who feel cheated when they lose money they've already paid.

Pricing consultant Steven Forth looked at 63 AI products and found that only 9 of them (14%) offered rollover. He describes limited carry-forward, such as one extra period or a declining balance, as the emerging compromise. That's a small, non-random dataset drawn from a pricing database, so treat it as a snapshot of design patterns rather than an industry census. But it's consistent with what the incentives predict: vendors prefer expiry, buyers prefer rollover, and the middle path is partial.

3. Top-ups and overage

What happens when the wallet hits zero mid-month? There are four common answers:

The right choice depends on what your product does. If it sits in a customer's daily workflow, a hard stop at 2 p.m. on a Tuesday is a support ticket and a churn risk. If it's an occasional tool, a hard stop is fine. Whichever you choose, spend caps that the customer controls are what make automatic refills tolerable.

4. Pooled versus per-user credits

Does the company share one balance, or does each person get an allotment? Pooled credits suit teams where usage is lumpy: one analyst burns ten times what a colleague does, and nobody wants to manage individual quotas. Per-user credits suit products where each seat is a distinct role and the buyer wants predictability per head.

Forth's dataset found that infrastructure products lean toward pooled allocation and applications lean toward per-seat models, with some vendors splitting the difference by letting part of a seat's credits flow into a shared pool. A hybrid like that gives the admin flexibility without letting a single heavy user drain the whole company's balance. If you're mixing seats and credits, our piece on hybrid pricing for AI products goes deeper, and the per-seat pricing in the AI era article covers the seat side.

5. Visibility

The best rate card in the world fails if the customer can't see their balance and burn rate. A visible balance, a per-action preview of the cost, and an alert at a threshold they set are not nice-to-haves. They're how you prevent the bill-shock moment that defines a bad credit experience.

Revenue recognition and breakage

Credits create an accounting question, and founders should understand the basics before a finance person asks. This section is a plain-language summary, not accounting advice. Your auditor decides how it applies to your contracts.

When a customer pays for credits up front, you generally haven't earned the money yet. You owe them future service. Under US GAAP's ASC 606, receiving a prepayment creates a contract liability for the obligation to transfer, or stand ready to transfer, goods or services in the future. Deloitte's roadmap sets this out in its summary of ASC 606-10-55-46 to 55-49. As the customer spends credits, you release the liability into revenue.

Now the interesting part. Some credits never get spent. Accounting calls that breakage. Per Deloitte's summary of the guidance, the treatment depends on what you expect:

• If you expect to be entitled to a breakage amount, you recognize it as revenue in proportion to the pattern of rights the customer exercises.
• If you don't expect to be entitled to breakage, you recognize the remaining balance when the likelihood that the customer will use it becomes remote.

Both approaches are subject to the standard's constraint on estimating variable consideration, which exists to stop companies from booking revenue they may have to reverse. In plain terms: you can't assume customers will leave credits unused just because it would look good on the income statement. You need evidence.

An illustrative example makes the proportional approach concrete. Say a customer prepays $12,000 for a year of credits, and you've got good historical evidence that about 10% of such credits go unused. You'd expect $1,200 of breakage. If the customer burns credits steadily, you'd recognize the $12,000 over the year, with the expected breakage spread in proportion to their usage rather than dumped in the last month. If you had no basis to estimate breakage, you'd wait until use of the remainder became remote. Again, these round numbers are for illustration only.

Two founder-level takeaways follow. First, how you handle expiry and rollover isn't only a customer-experience choice, it also changes your revenue recognition pattern. Second, prepaid credits sit on the balance sheet as a liability, so a big bookings number and a big revenue number are different things. When you talk about ARR and MRR, be careful about whether credit purchases are recurring commitments or one-off top-ups.

The buyer's problem: predictability and credit anxiety

Credits help the seller more than the buyer. That's the uncomfortable truth behind the model.

The buyer sees an abstract currency with a conversion rate they didn't set and can't verify. They can't easily compare vendor A's credits with vendor B's. And they can't forecast, because the number of credits an action burns may depend on inputs they don't control. This produces what's sometimes called credit anxiety: hesitating to use a product because every click feels like it spends money.

That anxiety isn't just a feeling. A 2026 report from Spendhound, based on 172 finance and procurement leaders plus spend data from more than 1,300 companies, found that 46% exceeded their AI budgets in 2025, compared with 37% for traditional software. The report points to consumption-based pricing and decentralized buying as contributing factors. It doesn't isolate credits specifically, so don't read it as proof that credits cause overruns. But it shows the environment your buyers live in: finance teams are already burned by variable AI spend, and a credit model walks into that room.

Anxiety hurts you in a specific way. A customer who's afraid of burning credits uses the product less. Lower usage means weaker habit formation, and weaker habit formation means a harder renewal. You can end up with a model that looks great on prepaid revenue and quietly kills adoption.

Founders can reduce the anxiety with a few concrete moves:

• Preview the cost before an action runs, especially expensive ones.
• Keep the rate card short. Five action types are memorable. Fifty aren't.
• Offer a free allowance for exploration so people learn the product before they worry about the meter. Our guide to free trials for AI products covers how to size that.
• Give admins controls: spend caps, per-user limits, and usage reports by team.
• Commit to price-change rules. If you'll change what an action costs in credits, say how much notice buyers get.

Credits versus seats versus pure usage billing

Credits sit between two other approaches. Here's how they compare.

The practical difference from pure usage billing is the wallet. Usage billing says "you used 40,000 units, here's the invoice." Credits say "you pre-bought 50,000 units and have 10,000 left." The prepayment and the abstraction are what make credits distinctive, and also what bring the breakage and trust questions. For the pure-usage side, see our explainer on usage-based pricing, and for where outcome pricing fits, the section's hub article on usage versus outcome-based pricing for AI.

Many companies end up with a hybrid: a seat or platform fee that includes a monthly credit allowance, with top-ups beyond it. That gives the buyer a predictable base and gives you cost alignment at the margin. The wider menu of options is mapped in SaaS pricing models.

Signals your credit model is working

A credit system is working when three things are true together.

Customers use most of what they buy, without feeling trapped. Healthy burn rates sit in a middle band. If nearly all credits are consumed early, customers feel starved and top-up friction rises. If a large share expires unused, they feel cheated, and you're borrowing against future renewals.

Margin is stable across actions. Track gross margin per rate-card line, not just in aggregate. A model that looks fine overall can hide a loss-making action that your best customers love. If one action consistently runs below target, change its credit cost or the way it's delivered.

Support questions about credits are rare. Count tickets that start with "why did this cost so much?" If they're climbing, the rate card is either too complicated or too opaque.

Signals it's failing

The failure signs are the mirror image.

• Usage drops after purchase. Customers buy a bundle, then ration it. Rationing is the clearest sign of credit anxiety.
• Expiring balances spike near renewal. A cluster of unused credits at term end predicts a fight over the next contract.
• Heavy users cost you money. If your top accounts have the worst margins, the rate card is subsidizing the wrong behavior.
• Sales spends calls explaining credits. If every deal needs a credit-to-task conversion walkthrough, the model has become a sales tax.
• Customers ask to convert to a flat fee. That request is feedback about predictability.

When several of these appear together, the usual fix isn't to abandon credits. It's to simplify the rate card, add visibility, introduce a limited rollover, or move to a hybrid base-plus-allowance plan.

Key Facts: Credit-Based Pricing

• A credit system has three parts: a currency (price per credit), a rate card (credits burned per action), and a wallet (balance, expiry, top-up rules).
• 52% of 255 B2B software companies surveyed in June 2026 use AI credit or token bundles to price AI features (Nalpeiron).
• Only 9 of 63 AI products studied (14%) offered credit rollover, per a February 2026 analysis by Steven Forth.
• Under ASC 606, prepayments create a contract liability and unused rights are called breakage; breakage is recognized in proportion to usage if expected, or when use becomes remote if not.
• 46% of surveyed finance and procurement leaders exceeded their AI budgets in 2025, versus 37% for traditional software (Spendhound).
• Credits help with cost alignment and prepaid cash. They cost you buyer predictability unless you add previews, caps, and a short rate card.

Credits are a tool for matching price to cost. They aren't worth the trust you spend if a new customer can't tell, in two sentences, what they're paying for.

Related Reading

• Usage vs Outcome-Based Pricing for AI
• Hybrid Pricing for AI Products
• Per-Seat Pricing in the AI Era
• AI Startup Business Models
• Usage-Based Pricing
• Gross Margin