The Economics of AI: Four-Week Online Course

A four-week online course on what AI applications cost and what they are worth, with twelve 30-minute recorded lectures and a 90-minute live session each week. Week 1 covers unit costs, week 2 infrastructure and total cost of ownership, week 3 value and business cases, and week 4 controls and the wider economy. No programming is needed. See Course Formats and Grading for the format, the grading scale and the course policies, and The Economics of AI for an overview of the subject.

At a Glance

Students

Business, economics, finance and management students and professionals.

Deployment

A shared Smarter instance run by the instructor, with at least three models from two or more providers configured.

Workload

3 hours of instruction and 4 to 6 hours of assignments each week.

Cost

Expect less than $10 USD of model charges per student, capped by a per-student Budget.

Learning Objectives

By the end of the course, students can:

  1. Explain how model providers price requests, and predict the cost of a conversation.

  2. Measure and compare the cost, latency and quality of AI tasks across models and designs.

  3. Build a total cost of ownership model for an AI application, including infrastructure and people, and compare build, buy and self-host options.

  4. Build a business case with scenarios and sensitivity analysis.

  5. Design and implement budget policies and chargeback.

  6. Discuss the macroeconomics of AI: prices, market structure, labor and productivity.

Technical Prerequisites

  • Required: a computer with a current web browser, a reliable connection for video calls, and a spreadsheet program.

  • Required: spreadsheet basics: formulas, references and charts.

  • Required: reading a YAML file well enough to change a value (Learn YAML in Y minutes).

  • Helpful: introductory microeconomics, and net present value.

  • Not needed: programming.

Schedule

Week

Item

Content

1

Lecture 1

The unit of AI cost. Tokens, input and output prices, cached input, and conversation length.

1

Lecture 2

Quality and latency. Model tiers, reasoning models, and comparing models fairly.

1

Lecture 3

Design choices that cost money. Tools, retrieval, long system prompts and multi-step workflows as cost multipliers.

1

Live session 1

Smarter’s cost records and the dashboard, live. Groups price a conversation by hand. Start Assignment 1.

2

Lecture 4

Infrastructure. Fixed and variable costs, the baseline cost of a cluster (about $350 USD a month for Smarter), and the marginal cost per user.

2

Lecture 5

Self-hosting. GPU costs, utilization, and when an open model on LLMHost beats paying per token.

2

Lecture 6

Total cost of ownership. People, support, switching costs, and build versus buy.

2

Live session 2

Review of Assignment 1. A total cost of ownership exercise in groups. Start Assignment 2.

3

Lecture 7

Measuring value. Time saved, deflection, revenue, quality, and the difficulty of measuring productivity.

3

Lecture 8

The business case. Adoption curves, scenarios, sensitivity, net present value, and presenting to a CFO.

3

Lecture 9

Pricing an AI product. Per-seat, per-use and outcome-based pricing, and margins when your costs are per token.

3

Live session 3

Review of Assignment 2. Students critique each other’s business case assumptions. Capstone proposals discussed. Start Assignment 3.

4

Lecture 10

Controlling spending. Budgets, warnings, limits, cost accounting and chargeback.

4

Lecture 11

Market structure. Falling prices, open versus closed models, the data center build-out and vendor concentration.

4

Lecture 12

Labor and productivity. Automation and augmentation, who gains and who loses, and the evidence so far.

4

Live session 4

Capstone presentations (about 6 minutes each, with 2 minutes of questions), and a course retrospective.

Assignments 1 to 3 are due before the next live session. Assignment 4 is due in the middle of week 4. The capstone proposal is due before Live session 3, and the capstone at the start of Live session 4.

Graded Assignments

Assignment 1: The Cost of a Design

Students take one task and run it four ways: a small model, a large model, the small model with a tool plugin, and the small model with a long system prompt. They measure cost, latency and quality, and explain what drives the differences.

Turn in: the spreadsheet, a chart, and a one-page explanation.

Criterion

Pts

Exemplary

Developing

Data

6

Complete, from Smarter’s records, with repeated runs.

Missing or estimated.

Cost drivers

6

Explains each difference by tokens and calls.

Reports differences without causes.

Quality

4

Scored with a stated rubric.

Impressions.

Presentation

4

A clear chart and concise writing.

Raw data only.

Assignment 2: Total Cost of Ownership

Students build a three-year total cost of ownership model for an assistant used by 1,000 employees, comparing a commercial model provider with a self-hosted open model, and find the usage level at which the cheaper option changes.

Turn in: the spreadsheet model with a break-even chart, and a one-page summary.

Criterion

Pts

Exemplary

Developing

Completeness

6

Model charges, GPUs, cluster, people and support.

Model charges and GPUs only.

Evidence

5

Measured or sourced figures.

Unsourced.

Break-even

5

Correct crossover, explained.

Missing or wrong.

Summary

4

States the conditions under which each option wins.

Picks one without conditions.

Assignment 3: Business Case

Students build a three-year business case for an AI application with three adoption scenarios, a net present value and a sensitivity analysis on the two most uncertain assumptions.

Turn in: the spreadsheet and a two-page memo.

Criterion

Pts

Exemplary

Developing

Costs and benefits

6

Measured costs, measurable benefits.

Vague benefits.

Scenarios and NPV

5

Three scenarios, correct NPV.

One scenario, or NPV errors.

Sensitivity

5

Identifies and tests the key assumptions.

Not done.

Memo

4

Clear recommendation.

No recommendation.

Assignment 4: Budget Policy and Chargeback

Students design budget limits and a chargeback scheme for an organization with three departments and four AI applications, implement the budgets in Smarter, and produce a mock monthly chargeback report from cost accounting data.

Turn in: a one-page policy, the manifests, evidence of a limit reached, and the chargeback report.

Criterion

Pts

Exemplary

Developing

Policy

5

Limits and chargeback rules that are fair and explained.

Arbitrary.

Implementation

6

Manifests enforce the policy; limit shown.

Mismatch, or not shown.

Chargeback report

5

Accurate allocation from real cost data.

Made-up figures.

Incentives

4

Discusses how the scheme changes behavior.

Not discussed.

Capstone Projects

Students choose one, propose it before Live session 3, and present it in Live session 4.

Capstone A: An AI Product Business Plan

Learning objective. Design the economics of an AI product: unit costs, pricing, margins and growth, grounded in measured costs.

Students propose an AI product sold to customers, such as an assistant for small law firms or a tutor for language learners. They build a prototype in Smarter, measure its unit costs on thirty realistic tasks, choose a pricing model, and build a three-year plan.

What students turn in:

  • The prototype manifests and measured results.

  • A financial model: unit costs, pricing, gross margin per customer, customer growth, fixed costs, break-even and sensitivity.

  • A five-page business plan, and a six-minute pitch.

How it is graded (100 points):

Criterion

Pts

Exemplary

Developing

Unit economics

25

Measured unit costs, margin per customer, heavy-user risk addressed.

Price-list costs; heavy users ignored.

Pricing

20

Pricing model justified against costs and competitors.

Arbitrary price.

Financial model

20

Complete, with break-even and sensitivity.

Incomplete.

Risks

15

Model price changes, vendor dependence and competition.

Not discussed.

Plan and pitch

20

Clear, persuasive and on time.

Unclear or overruns.

Capstone B: An AI Spending Strategy for an Organization

Learning objective. Advise an organization on how much to spend on AI, where, and how to keep control, using total cost of ownership, business cases and budgets together.

A fictional organization of 3,000 people asks for an AI spending strategy for the next two years. Students identify three candidate applications, build a business case for each, compare provider and self-hosting options, design budgets and chargeback, and implement the budgets in Smarter.

What students turn in:

  • A six-page strategy: priorities, business cases, sourcing choices, budget policy and governance of spending.

  • The supporting spreadsheets.

  • The budget manifests, with evidence that they work.

  • A six-minute presentation to the “executive team” (the class).

How it is graded (100 points):

Criterion

Pts

Exemplary

Developing

Prioritization

20

Applications ranked by value and feasibility, with evidence.

No ranking or rationale.

Business cases

20

Sound, with scenarios.

Superficial.

Sourcing

20

Build, buy and self-host compared on total cost of ownership.

One option assumed.

Controls

20

Budgets and chargeback implemented and shown.

Proposed only.

Presentation

20

Clear decisions requested, on time.

Data without decisions.

Further Study

  • Price and performance data: Artificial Analysis, which compares models on quality, speed and price, and the providers’ own pricing pages.

  • Cloud cost management: the FinOps Foundation, its framework, and its working group material on the cost of AI.

  • Reports: the annual AI Index from Stanford HAI, with chapters on the economy and on the cost of training and running models.

  • Podcasts: Exponential View with Azeem Azhar, on technology and the economy, Odd Lots from Bloomberg, which regularly covers AI and the data center build-out, and the a16z Podcast.

  • Writing: Ethan Mollick’s One Useful Thing on AI and productivity at work, and Benedict Evans’s newsletter and presentations on technology markets.

  • YouTube: talks by Ajay Agrawal, Joshua Gans and Avi Goldfarb on the economics of prediction, and Daron Acemoglu’s lectures on automation and inequality.