FIN 6150 · graduate financial management

How I teach FIN 6150

No textbook. AI on every assignment. Using it is never cheating.

Here is how the course is designed, how AI tutors coach students as they build Excel models and write executive memos, and how the grade stays honest.

One company
Backward design
The council
Three tutors
The scaffold
Live defense
The grade
The 2×2 grid
Monday moves

One company · three audiences

Same numbers. Three different questions.

Every team works one real company all term, Tesla or SpaceX, from public filings and disclosures. Students work in teams of three, and each student owns one business segment. Their job is the work they would do on that company’s finance team.

Arc 1
The CFO
Does the cash work? A working capital plan.
Arc 2
The CEO
Is the plan worth funding? Capex and financing.
Arc 3
The Board
Should we do the deal? Valuation and the exchange ratio.

Backward design · one source of truth

Start from what students must do alone.

I design backward. I start with six course learning outcomes, and the last one is to communicate a recommendation clearly and honestly, under pressure, with no AI. So every arc ends with AI switched off.

Then I work back. Seventy-seven concepts, each tied to an outcome. That list replaces the textbook, and students work from the 10-K and other public information. Then the sessions, then the assignments, which are built last.

6 outcomes. The last: communicate under pressure, no AI.
AI-off checkpoint. Every arc ends with AI switched off.
77 concepts. Each tied to an outcome. Replaces the textbook.
14 sessions. Built to reach the concepts.
Every assignment. Built last.

One source of truth, tested like code

The whole design lives in one workbook. For every assignment it records the outcome, the points, the grader, and whether AI is on or off. Before the term, I test the design against rules, the way an engineer tests code:

Every outcome is claimed by an arc.
Every arc ends AI-off.
Every assignment declares its AI mode.

Fail a rule, no launch.

Students using AI · the LLM council

Agreement between models is not evidence.

To build a capital budget, each student asks three different AI models for the company’s capital projects, using a prompt they co-create with the AI tutors. The point isn’t to average the answers. It is to see where they disagree, because three models agreeing is one source, counted three times.

Every number that survives needs a filing behind it, or the student owns it as an estimate.

Every surviving number carries

A source: a filing, a disclosure, the interview.

A $ class: disclosed · run-rate · estimate (you own it).

Model building by prompt

Students expand a 5-tab working capital model to 13 tabs by giving one specification prompt to the model of their choice. Every teammate uses the same prompt, so their tabs fit together, and the tutor flags structural errors before I see the file. They are learning to direct an AI, and to catch it when it gets things wrong.

Three AI tutors

Three tutors, each with one job.

Parker asks the questions

A Socratic questioner that helps students learn the 77 concepts the course is built around.

Avery builds and checks the model

Helps students build, edit, debug and expand their Excel models.

Rowan attacks the argument

Reads the executive memos and pushes back before they go to the executives, then helps students work the feedback into version 2.0.

The scaffold fades

The help fades. The work stays anchored.

Full scaffold
early
Shared
mid-term
Light
late
AI off
defenses

Any chatbot can answer

“What is days sales outstanding?”

Only a student who did the work can answer

“Your Energy segment collects in 70 days and Automotive in 12. Why?”

Peer evaluation · live defense

Score blind. Defend live. Then revise.

Each arc ends with a two-page memo to the CFO, the CEO or the board. Every student plays the executive the memo is written for.

Blind score. Students score classmates’ memos blind, on my rubric.
One question per memo, written before class.
A five-minute Q&A in front of the class, AI off.
Re-score, then revise. The revision carries the points.

What happened this term

Across 17 teams, 13 scored higher on ownership and truth-seeking after the defense, 3 went down and 1 stayed the same.

Defenses run in teams of three, or are spread across sessions, so the live check scales to a full class.

The grade · the audit trail

No point comes from an unsupervised model.

Peers score on my five criteria. Two AI models score the final memos independently, and I adjudicate where they disagree. Students earn the most points from the work that uses little or no AI.

Grade the polished file, and you’re grading the AI.

Every AI interaction is recorded in seven fields

Session ID
Timestamp
Prompt
Response
Persona role
Tool calls
Grading rationale

The tool you can use on Monday

Plot every graded item. The question is which cell it lives in.

Live in C to learn. Cash the grade in D to certify. Shrink A. Keep B small.

Assisted work can teach. Only independent performance can certify. The percentages are approximate targets for a course, not measurements.

The 2×2 grid as a table. Columns: was help allowed? Rows: what are you scoring?
What you score	With assistance AI, coach, notes, take-home	Without assistance live, no model, you watch
The product only	A · Pretty artifact (~5–10%). Polished memo, model, deck. After AI you cannot tell who did the thinking. The validity trap.	B · Closed check (~15–25%). Quiz, blue book, timed calculation. Integrity is better, but often too thin to prove they can do the job. Table stakes.
Observable competence	C · Coached practice (~20%). Attempt before teaching. A coach that will not write the deliverable. V1 → critique → V2. Learn here.	D · The credential (~50%). Live pitch, cold-call on their own artifact, rotating oral, in-class build. No model open. Certify here.

The honest part

Same work. Same points. With or without the AI tutors.

MBA students are demanding. They are taking time out of busy schedules and paying for their degree, and teaching this way has had a learning curve for them and for me. Students told me what wasn’t working, and some of it was fair.

So I offered one section a choice: the same 77 concepts, the same outcomes, the same points and the same activities, with or without the AI tutors. All 11 students who answered chose to keep them. They said the tutors pushed back on them and made them find the evidence.

Your Monday version · no special software needed

Three moves you can make on Monday.

Outcomes first, including one students must do without AI.
Run a council: three models, and a source for every number.
Grade the revision, after a five-minute live Q&A with classmates asking.

Get the worksheet

Are you a teacher? Do it yourself.

For teachers from K-12 through graduate school. Paste one assignment. AI-Proof gives back a version where AI coaches the practice and fades out, revisions earn the points, and the final checkpoint happens in class with AI off. You also get a rubric, plus a parent note for K-12 or a syllabus-ready AI statement for college courses.

Try AI-Proof free

Rebuilding a whole course? See AI-Proof Syllabus and Course with Agents.

3 free redesigns a month. Email sign-in, no card. Don’t paste student names, work or other personal information.

See a real redesign: an 8th-grade social studies essay →

Why it lasts

The model underneath will be replaced, repeatedly, by something better. The harness is what survives the replacement, which is why the harness is the asset and the model is the commodity.