Build the business case for an AI feature
Before you build, estimate what an AI feature will cost per use, what it will earn, the adoption at which it breaks even, and decide go, no-go or pilot.
Course designed by Alex Dionisio, Senior Product Manager. Updated November 14, 2026.
- 57 steps
- ~2 hr 46 min
Level: Advanced · Regular practice: you already use the tools or work on this topic.
Know, before you build, what an AI feature will cost per use, what it will bring in, at what adoption level it pays for itself, and whether to launch it. You learn to frame the decision and the starting situation, to estimate cost per use from tokens, resent conversation history and heavy users, to put a figure on value without confusing time saved with revenue, then to compute the break-even point and test your assumptions with scenarios. You write the go/no-go criteria before building and you present the decision to your leadership. You leave with the costed business case for an AI feature of your product, as one page and a spreadsheet, ready to present.
What you will be able to do
- State the decision the business case must settle, compare the options (including doing nothing and a non-AI solution) and put a figure on the starting situation, including the cost of doing nothing.
- Estimate, before building it, the cost per use and per user of an AI feature (input and output tokens, model, history, caching, heavy users), then its fixed and hidden costs.
- Estimate the expected value of an AI feature with explicit assumptions, separate time saved from captured value, and set the minimum quality and the way to measure that value.
- Compute the margin per user, the break-even adoption and the payback period of an AI feature.
- Build three scenarios, run the sensitivity analysis and identify the assumptions whose value flips the decision.
- Write quantified go/no-go criteria with a deadline before building, and design the pilot that tests them.
- Present the business case to leadership on one page, recommendation first, with the fragile assumptions and the answers to objections.
Prerequisites
- Work in product: you have already defended a priority or an investment in front of leadership.
- Know what a token and a prompt are, at the “I know what it is” level. Recommended courses before this one, not blocking: “Understand what LLMs do well, and where they fail” and “Prompting for Product Managers: delegate, describe, verify”.
- Have an AI feature in mind for your product, or use the one from the running example.
- This course requires no code and no advanced financial model: a spreadsheet and multiplication are enough.
Syllabus
What will I be able to do by the end of this course, and in what order?
- Course overview3 steps
Objective · By the end of this overview, you will know what you are going to produce (the costed business case for an AI feature of your product), which decision Atelio will settle throughout the course and which five modules take you there.
What decision must the business case settle, and what does the current situation cost if nothing is done?
Objective · By the end of this lesson, you will be able to write the decision an AI business case must settle, list the options that are really available, including doing nothing and a non-AI solution, and size the analysis effort to what is at stake.
Objective · By the end of this lesson, you will be able to cost the baseline of an AI feature, what the status quo costs in time and lost customers, without overstating it or treating it as zero.
How much will each use of the feature cost, and each user per month, before it is even built?
Objective · By the end of this lesson, you will be able to estimate the cost of one complete use of an AI feature before building it, from input and output tokens measured on a realistic prompt and the provider's dated prices.
Objective · By the end of this lesson, you will be able to go from the cost of one use to the monthly cost per user, factoring in the history resent at each turn, prompt caching, batch processing and the share of heavy users.
Objective · By the end of this lesson, you will be able to list and estimate the fixed and hidden costs of an AI feature (building, evaluation, observability, maintenance, model changes) and to date the prices your estimate rests on.
What will the feature bring to the company, and at what minimum quality does that value exist?
Objective · By the end of this lesson, you will be able to write the value hypotheses of an AI feature, separate the time the user saves from the value the company captures, and choose a monetization model and the metric that will measure each hypothesis.
Objective · By the end of this lesson, you will be able to set the minimum quality without which an AI feature does not create the expected value, estimate the cost of its errors and choose the indicator, such as the acceptance rate, that will check it.
From what adoption level does the feature pay for itself, and how do you decide without fooling yourself?
Objective · By the end of this lesson, you will be able to compute the margin per user of an AI feature, the number of users that covers its fixed costs and the payback period of the upfront investment.
Objective · By the end of this lesson, you will be able to build three consistent scenarios for an AI feature, run a sensitivity analysis and compute the switching value of the assumptions that decide the result.
Objective · By the end of this lesson, you will be able to write quantified go/no-go criteria with a deadline for an AI feature before building it, and design the pilot that tests the most fragile assumptions.
Objective · By the end of this lesson, you will be able to present the business case for an AI feature on one page, recommendation first, with the key figures, the fragile assumptions and prepared answers to leadership's objections.
What does the complete business case for your AI feature look like, ready to present?
Objective · By the end of this workshop, you will be able to produce the complete business case for an AI feature of your product, as one page and a spreadsheet, and to review a peer's with a self-assessment grid.
Glossary terms
Public definitions of the key concepts in this course:
Verified sources
The fast-moving facts of this course (tools, models, regulation) are checked against these sources, then reviewed according to how quickly they go stale.
- Anthropic, Batch processing (Claude API docs, 2026) · verified November 14, 2026
- Anthropic, Context windows (Claude API docs, 2026) · verified November 14, 2026
- Anthropic, Pricing (Claude API docs, 2026) · verified November 14, 2026
- Anthropic, Prompt caching (Claude API docs, 2026) · verified November 14, 2026
- Anthropic, Token counting (Claude API docs, 2026) · verified November 14, 2026
- FinOps Foundation, FinOps for AI Overview (FinOps Foundation, 2026) · verified November 14, 2026
- Google, Gemini API pricing (Google AI Studio, 2026) · verified November 14, 2026
- Google, Tokens (Gemini API docs, 2026) · verified November 14, 2026
- HM Treasury, The Green Book 2026 (GOV.UK, 2026) · verified November 14, 2026
- ICONIQ, 2026 State of AI, The Builder's Economy (ICONIQ, 2026) · verified November 14, 2026
- OpenAI, Conversation state (OpenAI API docs, 2026) · verified November 14, 2026
- OpenAI, Pricing (OpenAI API docs, 2026) · verified November 14, 2026