More AI companies are charging for usage in credits than ever before.
Chargebee recently hostedUlrik Lehrskov-Schmidt, CEO of pricing advisory firm Willingness to Pay, for a live session on building credit-based pricing that holds up past the first sale. This recap pulls the frameworks, the mechanics, and the audience questions from that session. The timing fit the topic, with Notion overhauling its AI pricing the same week. Credits have become the default meter across Clay, Cursor, and a growing list of AI-native companies.
Getting credit-based pricing right means solving two problems. On the customer side, credits have a reputation for being a black box: people don’t know what a credit actually buys, and a balance that suddenly drops leaves them guessing why. On the vendor side, enterprise contracts used to lock in what a customer would get, upfront, for the life of the deal. AI breaks that assumption: a new model ships, an agent gets better at its job, and the value a product delivers can jump mid-contract without either side touching the price list. Pricing something whose value keeps moving, fairly and transparently, is the problem the rest of this piece works through.
Neither a license nor usage-based billing fits AI well. The volume is new and hard to estimate, and the cost behind each unit of consumption is real, not theoretical. Ulrik pointed to data from PricingSaaS — the pricing-intelligence platform that merged with Willingness to Pay this year — showing around 320 credit systems live on pricing pages today, about 8% of the market tracked, up from close to none three years ago.
Compare the Four Pricing Modalities on Cash, Risk, and Renewal
Ulrik breaks the choice into four modalities, or ways to charge for software: flat fee, license, usage, and credit.
- A flat fee charges one number regardless of volume — simple to sell, blind to how differently customers actually use the product.
- A license sells a fixed volume upfront; the customer carries the risk, cash arrives early, and renewals are a fresh negotiation.
- Usage-based pricing is the opposite: no upfront commitment, the vendor carries the risk, and predictability drops because nobody knows the bill until the period closes.
- Credits sit in between, and deliberately so. “Licenses and usage are polar opposites on almost every dimension that matters,” Ulrik said. “Credits come in and do the best of both worlds.” Cash arrives upfront like a license, but the customer, not the vendor, carries the risk of how it gets spent. Every other term is configurable rather than fixed:
Decide Whether Credits Are the Right Model for You
A credit-based pricing model makes the most sense when customers can’t estimate their usage volume; when usage carries a real cost that needs to be collected before it’s incurred; or when several different product actions should draw from one shared budget. Runway is an excellent example: nobody knows upfront how many video or image generations they’ll need, and every one of them costs real compute, priced in credits that scale with resolution and clip length. They’re the wrong choice when none of that complexity exists: a single, easily estimated metric is still better sold as a license, and a near-zero cost to serve is still better sold as usage.
Choosing Between Credits, Usage, and Outcome Pricing?
Still deciding on a pricing model for your AI product? Chargebee’s 7-factor framework for choosing an AI pricing model is the place to start.
Design Exchange Rates That Flex With Your Costs
Once credits are the right call, the real design work starts with making sure the system can flex as costs change. Every credit system runs on two exchange rates, and conflating them is the most common design mistake. The first is dollar-to-credit: how many credits a dollar buys. This is the stable layer. It lives in the contract, it’s what gets negotiated, and Ulrik advises that it should be simple enough to do in your head — 1 credit per dollar, or 10, or 100, never a fraction.
The second is credit-to-product: what a credit buys inside the product. This is the volatile layer, and it’s where the real flexibility lives. “You get to be your own central bank,” Ulrik said. “You control the exchange rate inside your own economy, without touching the stable contract layer with your customer.” Companies can reprice an action, retire a line item, or add a new one, all without renegotiating the contract, because the customer agreed to the dollar-to-credit rate, not the credit-to-product one.
Apply discounts at the dollar-to-credit layer, never inside the credit-to-product rate card, or every customer ends up with a different, unmanageable price list. And be careful about denominating credits in dollars at all. Rollover complicates this fast: it means a customer has purchased something but hasn’t used it yet, which you have to treat as deferred revenue, since it’s a liability until they spend it. Add a discounted dollar-to-credit ratio on top, and the revenue recognition math gets messy enough that most teams are better off calling the currency credits and leaving dollars out of the account entirely.
For the full mechanics of setting a credit ratio, seeChargebee’s prepaid credit pricing guide.
Know When to Add a Second Credit Pool
Most credit systems fall into one of three shapes: single-purchase, multi-purchase, or multi-pool.
Single-purchase is the simplest: one credit buys one unit of one thing, the way Audible sells one credit for one audiobook.
Multi-purchase is the default for complex products — many line items, different credit prices, one fungible pool — and it’s what Ulrik recommends for roughly 90% of credit systems.
Ulrik recommends reserving multi-pool for products with genuinely different cost or rollover logic across categories. Clay is the clearest example: data-enrichment credits and action credits run as two separate pools rather than one, because the two categories behave differently enough that pooling them would blur the economics. Clay’s pricing team has detailed how that model evolved as the company scaled, both in arecent article and on theThank You for Vibe Pricing podcast. The rule of thumb: default to one pool, and only split when the cost structure forces it.
Use the Volume Terms Matrix to Set Risk Terms Deliberately
Every sale, whether license, usage, or credit, makes four implicit decisions about volume:
- How much the customer buys today
- What happens if they need more
- What happens if they use less
- What they buy at renewal
Ulrik calls this the Volume Terms Matrix: the ability to set initial volume, overconsumption, underconsumption, and renewal independently, instead of inheriting a license’s or a usage model’s defaults wholesale. Get it right, and it pays off in several places at once — the purchase feels frictionless, anxiety at checkout drops, the customer’s momentum inside the product holds, and you still capture revenue for anything they consume beyond what they originally bought.
Here’s the setup Ulrik defaults to:
- Initial volume: let a framework suggest the number, but have the customer choose it
- Overconsumption: let more volume unlock the next discount tier rather than penalizing it
- Underconsumption: roll over credits in full if the customer renews at roughly 80% of what they consumed
- Renewal: set volume to match actual usage rather than restarting from zero
Partial rollover is the trap teams fall into when they try to split the difference. “If you have 40 credits to roll over and I only roll over 39 of them, what are people going to focus on?” Ulrik asked. “The one credit they lost.” The math may work, but psychologically, people fixate on what they lose, not what they keep.
Get this wrong at the initial sale and it compounds. Ulrik described a Silicon Valley company that grew from $8 million to $90 million in annual recurring revenue on six-figure license deals, then spent the next three years stalled getting to $100 million, because every contract came up for renewal over-committed, on bespoke terms, with a disgruntled customer and a procurement team pushing for concessions.
Watch the Full Volume Terms Matrix Walkthrough
Watch the full session on demand for the complete walkthrough of the Volume Terms Matrix, slide by slide, in “How to Make Credit-Based Pricing Work.”
Build Guardrails So Credits Don’t Feel Like a Black Box
Credits move estimation risk onto the customer, and that risk has to be handled somewhere. Here’s how Ulrik hands the customer the tools to carry it:
- Measure: real-time visibility into what’s been consumed, by whom, on what
- Control: caps, pauses, and alerts before the pool runs dry
- Permissions: who in the customer’s organization can approve more spend
Skip these, and customers negotiate the risk back into the contract: hard caps, deep discounts, unlimited-volume clauses, because nobody wants to be surprised by an invoice. Build them well, and the customer can manage the risk themselves, which is usually what they want.
Flat per-user allowances are a common trap. It’s tempting to cap every seat at the same number of credits, but Ulrik has seen a consistent pattern across AI products: a normal outcome is 75% or more of total AI cost concentrated in 1% of users or fewer. Cap every individual the same way, and the system throttles the power users who should be driving account-wide adoption. Give the account the controls instead, and let usage find its own shape.
The same logic applies to the contract itself. “Don’t sign a three-year fixed contract with no ability to change pricing or packaging,” Ulrik said. “Build the contract today so you can stay flexible as the market moves,” and tell the customer that’s the deal, rather than pretending today’s price list is permanent.
Read all about how to make AI profitable in Ulrik’s new book.
The AI Pricing Roadmap
How to Make AI the Most Profitable Thing You’ve Ever Sold.
Design Credit Risk on Purpose
Credit-based pricing works because it lets you configure risk deliberately: initial volume, overconsumption, rollover, renewal, instead of inheriting those terms from a license or a usage model that was never built for AI’s unpredictable curve. The mechanics (two exchange rates), the architecture (usually one pool, not three), and the guardrails (measure, control, permissions) are what make that configuration hold up past the first sale.
Watch the full session on demand for the complete walkthrough, or runChargebee’s AI Pricing Diagnostic to see where your own model might be leaving margin on the table.
For more frameworks like this, exploreChargebee’s AI Pricing and Monetization resources, including our podcast on thehigh-stakes pricing pivots Notion, Miro, Vercel, and other SaaS leaders are making.
Frequently Asked Questions
What is credit-based pricing?
Credit-based pricing sells customers a prepaid pool of credits upfront, then draws the pool down as they consume different features, actions, or API calls. Unlike a license, unused credits can roll over. Unlike usage-based billing, the cash arrives before consumption, not after.
How is credit-based pricing different from usage-based pricing?
Usage-based pricingbills after the fact, from zero, with no prepayment. Credit-based pricing collects cash upfront and draws down a pool, which gives the vendor cash flow and the customer a spending cap they control, at the cost of needing a rollover and renewal policy.
What is one credit worth, and how do you set the ratio?
Work backward from the smallest thing you sell, keep the dollar-to-credit ratio simple (1:1, 1:10, or 1:100), and apply discounts at that layer rather than inside the product’s credit-to-product rate card.
Should credit value be measured from the vendor’s perspective or the customer’s?
Always the customer’s. A rate card priced around what something costs you to run, rather than what it accomplishes for the customer, creates friction and gets hard to explain. Price the outcome the customer recognizes, not the compute behind it.
Should credits expire, and when?
Most teams use a rolling policy rather than a hard expiration: annual credits roll over once then expire, commonly capped around 24 months, monthly credits roll twice, and the oldest credits are spent first.
What happens if a customer runs out of credits mid-period?
Self-serve products typically use a hard paywall. Contract customers get more flexibility: allow a 10 to 20% overage, bill it in arrears, or pull the renewal forward, but avoid letting anyone consume for free or get cut off mid-work.
How do tiered licenses and credit discounts work together?
A base license fee buys access to the best rate card, and credits are purchased at a discount within that tier. It’s effectively a flat fee stacked with a credit system: the license pays for the better pricing, the credits pay for consumption.
