Tool · Pricing

A pricing model is a choice about who pays the bill.

Answer up to four questions and get a starting hypothesis for your product, why it came out that way, and where it can mislead you. The result is a first idea to test, not a verdict.

Author: Alp SuleimanPublished 5 October 2026

01Question 1 of 4

Does your product finish the job end to end, or does it offer suggestions and drafts?

If the product does not close the loop, it is hard to attach an outcome to it. This question rules outcome-based pricing out early or leaves it open.

Method

What this tool rests on, and what it does not

The tree is a decision skeleton drawn from public sources: Bessemer's AI pricing playbook, three pieces by Poyar, Poyar and Broe's review of 40 AI startups and, for the survey data in the FAQ, ICONIQ's State of AI 2026 report. The order of the questions, the thresholds and every sentence in the reasons that carries neither a source nor an (interpretation) tag, advice or observation alike, are our interpretation and rules of thumb; none of the sources proposes this tree in this form. It can change as new information arrives.

There is no perfect pricing model. As Poyar wrote in October 2025: there's no such thing as a perfect pricing model. The tool does not find you the right model; it makes visible which risk you are taking on and gives you the first hypothesis to test.

The angle here is our own, and it is interpretation: a pricing model is a choice about who keeps the inference bill and the exchange-rate risk. In flat pricing the bill stays with you; in usage-based pricing it passes to the customer; in hybrid it is shared; in outcome-based pricing it stays with you again. Look at the distribution of usage first, then decide on the model.

The signature question: what does your worst customer cost you at month end, and does your price page know it? If you have no answer, which model you pick comes second; first model that customer in the AI unit economics calculator.

Turkey note (interpretation): in usage-based pricing the exchange-rate risk passes to the customer; in fixed lira pricing and in outcome-based pricing it stays with the seller. If your inference cost is in dollars, that difference can decide the outcome.

The idea behind this tree, that a pricing model decides who carries the inference bill, is set out in full in the pricing section of the guide. Growth in the AI Era: pricing →

Frequently asked

Why is outcome-based pricing not right for everyone?

Because inference cost and the failed attempt stay with the seller. Bessemer lists three conditions: confidence in performance, the ability to absorb cost variance, and a clear, measurable outcome. Poyar notes that many enterprises still crave predictability. That is why the tool recommends outcome-based pricing only when all three conditions hold together.

Why is hybrid so common?

In Poyar's 2026 survey (230+ software companies, April–May 2026, self-reported) hybrid was the most common model (hybrid = a combination of two or more models): 37%, up from 25% twelve months earlier. In ICONIQ's July 2026 report (Q4 2025 and Q2 2026 surveys of ~300 software executives, self-reported), consumption-based pricing rose from 35% to 42% in six months and outcome-based from 18% to 23%; companies blend 1.7 models on average. These are survey results; they do not show what is right for you.

Is seat-based pricing over?

In Poyar's words, seats are not necessarily dying but are less effective than they used to be. Where value scales with the number of users, a seat can still make sense. As automation grows, customers need fewer seats, and revenue can fall as the product succeeds.

Can I offer more than one model at once?

In Poyar's 2026 survey, 29% of companies let customers choose between several models (230+ companies, self-reported). The option gives customers flexibility but complicates the price page and billing. Starting with one model and adding a second once usage data arrives is the healthier order (rule of thumb).

Why does token pricing not fit everyone?

In Bessemer's words: Tokens work for technical buyers but confuse everyone else. A token relates to your cost, not to the customer's work. For a non-technical buyer, a unit closer to value (a resolved ticket, a prepared file) or included usage inside a plan is easier to understand.

Does your product finish the job end to end, or does it offer suggestions and drafts?

Frequently asked

Sources