How to Price an AI SaaS: Credits, Prompts or Seats

The short version

AI products have a cost that classic SaaS mostly did not: every time a customer uses the product, you pay a model provider. A flat monthly price with unlimited use works until one heavy customer costs more than they pay. That is why most AI products end up charging, at least partly, for usage: credits, prompts, generations or tokens, often bundled into monthly tiers.

There is no single right model. The useful question is which unit your customer understands, which unit tracks your cost, and how close those two can be.

The four common models

Many products combine them: tiers with a monthly quota (predictable for the customer), plus overage or top-up credits (protection for you).

Our own case: prompts per month, with weighted actions

All Innovativ runs about 50 tools, many of them AI-powered. Here is how we price, as listed on our plans page in October 2026, and why.

Why a "prompt" and not a credit

A prompt is something a customer can picture: one request to a tool, one answer back. Credits would let us price every action precisely, but they add a conversion step in the customer's head ("how many credits is a blog post?"). For most of our tools, one request costs us roughly the same, so a simple count was the clearer choice.

Where one prompt is not enough

Some actions are much heavier than others. A full Pitch Deck Analyzer report reads a whole deck and writes a long, structured report, so it counts as 2 prompts. In Pitch Simulator, every AI turn counts as one prompt, so a 5-question session uses 6. We measured what each heavy action actually costs before deciding, and published the weight next to the tool, so nobody is surprised.

A separate meter for the most expensive output

Generating full SEO pages with upSEO costs far more than a chat answer, so it has its own monthly quota of pages instead of using prompts. Beyond the quota, extra pages are billed per page ($0.15, or $0.10 on Agency), there is a confirmation before any overage is generated, and a monthly hard cap of three times the quota protects both sides from a runaway bill. Customers without a subscription can buy one-off page credits that stay valid for a year.

Rules that keep it understandable

• Quotas reset monthly on the billing date and do not roll over.
• Free tools never count. Client-side tools such as the password generator or QR code generator do not use prompts.
• Cheaper models on the free plan. Paid plans unlock premium models; free users get free models, which keeps the free plan sustainable.

How to choose for your AI product

1. Measure your cost per action. Log the tokens and cost of real requests for a few weeks. Averages hide outliers; look at the most expensive 10% too.
2. Find the unit your customer already thinks in. Pages, videos, analyses, minutes of audio. If customers think in seats, keep seats and add a fair-use limit.
3. Weight the outliers. If one action costs several times more than the rest, count it as more than one unit or give it its own meter.
4. Set quotas from margins, not hope. A quota should leave a healthy margin even if a customer uses all of it, every month.
5. Add a ceiling. Overage, a hard cap, or both. Confirm before charging extra.
6. Revisit when model prices change. Model costs move; your quotas can become more generous, or need tightening, within months.

Common mistakes with AI pricing

• Unlimited plans. One power user or automated script can erase the margin of many customers.
• Abstract credits with no explanation. Always show what a credit buys, in the customer's own terms.
• Hidden weights. If an action costs 2 units, say so next to the button.
• Pricing from cost alone. Cost sets the floor; the value to the customer sets the price. A report that saves a founder a day of work is not worth only its token cost.
• Never revisiting. AI costs and customer behaviour both change quickly.

Model it before you launch

The Pricing Strategy tool can build this for your product. Describe it, pick developer/API or B2B as the market, add your current numbers and costs, and choose a goal. The report recommends a model, such as tiered, usage-based or hybrid, with tiers, price ranges, psychology tips, a competitor benchmark, a revenue projection with its assumptions stated, and a three-phase rollout. For the launch offer itself, see founding member pricing, and for changing prices later, how to raise prices.

Prices and quotas are as listed on allinnovativ.com/plans in October 2026 and may change. This guide is not financial advice.