Modern Pricing & UBB
February 26, 2026 ai-pricing

Credits Are the New Seats

For twenty years, the seat was the atom of SaaS pricing. One human, one license, one line item on the PO. Then AI showed up and broke the whole model. You can't charge per seat for an agent that runs 24/7 and never takes a lunch break. So the industry needed a new atom. It found one: the credit.

A credit sounds like a token, looks like a prepaid unit, and behaves like a seat — from finance's perspective. You buy a block of them. You have a number. That number is predictable. The CFO can put it in a spreadsheet. 44% of SaaS companies are now charging for AI features (Maxio 2025), and most of them, when they got serious about monetization, landed on some form of credit-based system. This is not a coincidence.

Why Credits Win

Credits solve the exact problem that sank pure usage-based pricing at the enterprise level: unpredictability. A procurement team cannot sign a contract that says "we'll pay whatever the API calls add up to." That conversation ends with someone asking if you accept net-180 terms and then never emailing back. Credits convert variable consumption into a committed, budgeted number. You pre-buy the credits. You draw them down. When they run low, you buy more.

This is brilliant for three reasons that nobody talks about enough:

1. Pre-commitment improves your cash flow dramatically. Instead of billing monthly in arrears for what was consumed, you collect upfront. The customer pays before they've extracted full value, which is the dream. AWS built its cloud on this logic. Every credit bundle sold is deferred revenue sitting on your balance sheet.

2. Credits decouple your value unit from your cost unit. If you bill by token, you're exposing your inference margin to every customer with a calculator. One credits = some black-box amount of compute. The customer sees "credits used," not "tokens consumed × $0.003 per thousand." You maintain pricing flexibility. You can quietly adjust the credits-to-compute ratio as your infrastructure costs fall — and with every GPU generation, they do fall — without renegotiating contracts.

3. The endowment effect works in your favor. Once a customer has purchased 50,000 credits, those credits feel like money sitting in an account. Unused credits are psychologically painful. Customers have a strong incentive to actually use the product — and when they run low, the expansion conversation is just "do you want more of what you already have?" No new vendor evaluation. No procurement cycle. Just a top-up.

The Design Choices That Kill It

Credits are powerful but not foolproof. Two things break credit systems faster than anything else:

Expiry policies that feel punitive. If credits expire in 30 days, you're telling the customer that their unused value evaporates — and they will hate you for it, especially enterprise teams with spiky usage patterns. Rollover credits cost you almost nothing in economic terms but massively improve satisfaction and renewal rates. The math is simple: the credits they're "keeping" would have been consumed anyway as they grow. Give them rollover. The exception is a steeply discounted promotional bundle — expiry is how you justify the discount.

Showing customers the tokens-to-credits ratio. The moment you publish a table that says "1 credit = 2,000 input tokens + 500 output tokens," you've turned your customers into analysts who will benchmark your margins against every competitor. Keep the abstraction. "1 credit = 1 AI generation" is enough. If your enterprise team pushes for transparency, offer an audit of total credits consumed vs. outcomes delivered — not raw compute metrics.

Credit Tiers: The New Seat Count

The maturation play is to tier your credit plans the same way SaaS tiered seat counts. Starter: 5,000 credits/month. Growth: 25,000 credits/month. Enterprise: custom. Each tier should be priced so a heavy user in the lower tier is strongly incentivized to upgrade — not because of hard limits, but because the per-credit rate at the next tier is materially cheaper. This is yield management, and it works.

The company that figures out credits has solved the AI monetization problem. Not tokens — credits. Tokens are for engineers. Credits are for deals. And deals are what keeps the lights on.

Sources

← Credit Systems · All posts