The economics of AI pricing

The pricing problem

AI products have a weird cost structure. The marginal cost of a query isn't zero — every API call to Claude or GPT-4 costs real money. This breaks the traditional SaaS pricing model (per-seat, all-you-can-eat) in subtle ways.

When we launched Ion, we picked the obvious option: per-seat pricing, $19/seat/month, unlimited queries. Three months in, we noticed a problem: 5% of our users were consuming 60% of our AI compute. We were effectively subsidizing power users at the expense of everyone else.

The options we considered

Option A: Hard usage caps

Limit each seat to N queries per month. Simple, but feels hostile — users hate hitting limits, and it punishes the power users who get the most value from your product.

Option B: Usage-based pricing

Charge per query. Aligns cost with value, but creates anxiety — users think twice before every query, which kills the exploratory use cases that make AI products valuable.

Option C: Tiered allowances + overages

What we ended up with: each plan includes a generous allowance (10,000 queries/month on Pro), with soft overages that warn but don't block. Power users who consistently exceed can upgrade to Enterprise with a custom allowance.

Why option C won

Three reasons:

1. Predictable billing. 95% of users never hit the cap, so their bill is the same every month. No anxiety.
2. Rewarding power users. The 5% who hit the cap are getting massive value from Ion — they're the most likely to upgrade to Enterprise, not churn.
3. Cost discipline. We can forecast AI compute costs per workspace and price accordingly. No surprises.

What we'd do differently

We should have started with tiered allowances from day one. The "unlimited" pricing was a marketing decision, not a business one — it sounded good in the pricing table but created a structural problem we had to unwind later.

If you're launching an AI product, price for the unit economics from day one. Your future self will thank you.