How I price indie AI products

The token-cost problem

Traditional software pricing assumes fixed marginal cost: serving customer N+1 costs roughly the same as serving customer N. AI software breaks that. A power user can drive 100× the inference cost of a casual user, and they'll keep doing it because the marginal cost to them is zero.

The most common mistake — and the one I made repeatedly — is using flat-rate subscriptions on usage-elastic features. You end up subsidising power users with new-customer revenue. It looks fine until growth slows.

Three buckets

Bucket 1: Closed-form features → one-time purchase or low monthly

Anything where the user gets a discrete output (a portrait, a summary, a checklist) and walks away. Cost per use is bounded. Price it like a tool. App Store one-time IAPs or simple monthly subscriptions work fine.

Bucket 2: Open-ended chat or agents → usage-aware tier

If the user can drive cost linearly, the pricing must too. Either token-credit models, hard daily caps with reasonable defaults, or fair-use throttling. Don't pretend the cost isn't there.

Bucket 3: Vertical or enterprise → custom

At the high end, value is wildly variable and so should pricing be. “$39/seat/month” is fine for generic SaaS; it's a leak for vertical AI deployment. Charge based on the problem solved.

Free tier as discovery, not freemium

Freemium with AI is dangerous because your “free” users are still costing you inference dollars. The trap is converting them to paid before you've covered acquisition cost.

I treat free differently: it's a one-screen demo, not an indefinite subscription. The user gets to feel the magic once, sees what unlocks paid, and decides. This is harder to game and more honest.

Pricing as positioning

Pricing tells users what kind of product this is. $0.99 says “casual app.” $19/month says “tool I use weekly.” $499/month says “this saves my company $5K/month.” Picking the wrong number for the right product is one of the fastest ways to fail.

For indie AI products specifically, I bias toward higher prices than feels comfortable. The customers who will get the most value are usually the ones who can afford it. Cheap pricing attracts churn-prone users; appropriate pricing attracts customers who treat the tool seriously.

A simple test

Before I set a price, I check three things:

1. Worst-case unit economics. If my heaviest 5% of users are at this price, am I still profitable? If not, adjust pricing or add limits.
2. Anchor. What does the closest non-AI alternative cost? Price relative to that, not to other AI products (which are often mispriced themselves).
3. One-sentence justification. Can I explain in one sentence why this price is fair for this value? If not, I haven't thought about it enough.

What I don't do

• “AI credits” as the primary unit. Users can't translate credits into outcomes. I'd rather show usage in their language (“3 portraits / month”).
• Pricing-tier sprawl. Three tiers max. More than that is a way of avoiding a hard pricing decision.
• Discounts as marketing. A discount is a confession that the regular price is wrong.

Pricing isn't a one-time decision. Re-examine when your cost basis changes (a new model release), when you launch a new feature, or every six months at minimum.