Seat-based vs usage-based vs credit pricing for AI products
Seats win on simplicity and lose margin the moment usage per seat varies. Pure usage protects margin perfectly and scares buyers with unpredictable bills. Credits are usage pricing with better manners. In 2026 most AI products converge on a hybrid of all three: a base fee, an included allowance, and a meter past the cap. The deciding input is your own cost spread: how much more your P90 customer costs than your P50.
The three models side by side
| Per-seat | Usage-based | Credits | |
|---|---|---|---|
| Revenue predictability | High | Low | Medium (prepaid pools help) |
| Margin protection | None. Cost varies, price does not | Full. Bill tracks cost | Full, if burn rates are priced right |
| Buyer comfort | High. Everyone budgets seats | Low. Bill shock risk | Medium. Familiar from Cursor, v0, Lovable |
| Ops complexity | Trivial | Metering plus billing infra | Metering plus a burn table you must maintain |
| Fails when | One seat burns 50x another | Buyers throttle themselves; revenue caps | Burn-rate changes read as stealth price hikes |
The same 100 customers, priced three ways
An illustration with round numbers. Suppose 100 accounts, AI cost per account distributed like a real book: 80 light accounts costing $8/month each, 17 medium at $45, and 3 heavy at $400. Total AI cost: $2,605/month.
| Structure | Monthly revenue | Gross margin | What happens |
|---|---|---|---|
| Flat $79 for everyone | $7,900 | 67% | Looks fine blended. The 3 heavy accounts each destroy $321/month, subsidized by 80 light users. Any growth in heavy users erodes margin. |
| Pure usage, cost × 4 | $10,420 | 75% | Margin is uniform by construction. But light users now pay $32 and wonder why the bill moves, and heavy users see $1,600 and negotiate or churn. |
| Hybrid: $49 base + usage over $15 cost billed at 4x | $9,713 | 73% | Light users get a stable $49. Heavy accounts pay $1,589 each, tracking their cost. Nobody subsidizes anybody. |
The numbers are illustrative, but the shape is not: flat pricing converts your best-engaged accounts into your worst losses, and the hybrid recovers nearly all of pure usage's margin while keeping a predictable bill for the 80 percent of customers who are light. Run the same three structures against your own last 6 to 12 months and the decision usually makes itself. That replay is exactly what a Pricing Reset does, and the input it needs is the per-customer cost map from our cost-per-customer guide.
When do seats still make sense?
Seats survive where usage per seat is genuinely uniform: workflow tools where every user does roughly the same volume of work, or products where the AI feature is a small garnish on non-AI value. ChatGPT Business still sells seats at $25 to 30 per user per month, but note what OpenAI bolted on: credit top-ups for heavy usage. That is the 2026 pattern everywhere: seats keep the billing anchor, and a usage valve gets added so the tail pays for itself. What is disappearing is not the seat, it is the unlimited seat.
When does usage-based win?
When your cost and your customer's value scale together and your buyers are technical enough to accept metered bills: API products, developer tools, infrastructure. GitHub Copilot moved all plans to usage-based billing on June 1, 2026, and the developer reaction ("you will get less but pay the same," as one Visual Studio Magazine roundup put it) is the honest cost of the transition. If you go pure usage, three rules: publish the rate card, ship spend caps and alerts on day one, and never let the first surprise bill be how a customer learns the pricing.
When do credits win?
When you need usage economics but your buyers are not engineers, credits translate token math into a unit a human can budget. Cursor, v0, and Lovable all price this way, and credit models grew 126 percent year over year in 2025 across the PricingSaaS top 500. Two design rules keep credits from backfiring. First, publish the burn table (what each action costs in credits) and version it publicly; re-rating burn silently is how you end up in a Cursor-style apology cycle. Second, size the included pool from your P75 customer's real usage, not from what makes the tier page look generous.
How to decide in one afternoon
- Pull 90 days of cost per customer (method in this guide).
- P90 ÷ P50 under 3: keep seats or flat. Add a fair-use clause and revisit quarterly.
- P90 ÷ P50 between 3 and 10: hybrid. Base fee near your current price, allowance near P75 usage, overage or credit top-ups at 3x to 5x marginal cost.
- P90 ÷ P50 over 10: your tail is the business problem. Meter first, then choose the friendliest wrapper (credits for non-technical buyers, raw usage for developers).
Frequently asked questions
Is seat-based pricing dead for AI companies?
No. It is being fitted with meters. Seats remain the easiest model to buy, and hybrid designs keep them as the anchor while a usage component covers the cost tail. Pure unlimited seats are what is dying.
Why do AI companies use credits instead of dollars?
Credits decouple the customer-facing unit from volatile provider token prices, smooth over multi-model routing, and make prepaid pools feel like a budget rather than a taxi meter. They also give the vendor a margin lever in the exchange rate, which is why the burn table deserves public versioning.
What should overage cost?
A common pattern is 3x to 5x your marginal cost, set slightly above the effective in-plan rate so the plan feels like the deal. Price overage too high and customers churn at the cap; too low and heavy users have no reason to upgrade tiers.
See these three structures on your own data
The AI Pricing Reset replays your last 6 to 12 months against your current pricing, a hybrid with caps plus credits, and a hybrid with caps plus overage, then hands you the winning blueprint with exact thresholds and the migration pack. Delivered in 10 business days. One-time fee, no subscription.
See the AI Pricing Reset$1,500 founding rate · first 10 AI SaaS · $2,995 list