MGT 816 · Fall 2026

AI Toolkit

Two things here: the CFO AI Skills — eight skill files you keep after the course — and the PS2 forecasting scaffold from Sessions 6–7. The rule is the same for both: AI does the analyst work; you certify every number.

The CFO AI Skills

Eight skills, one per block of the course. Each is a folder: a SKILL.md that teaches an AI assistant the methods from class — the formulas, the decision rules, the red flags — plus the working templates that go with them. They are written for the jobs you will hold one to five years out: first finance hire, founder, fund associate, operator. The course ends in October; the skills don’t. Each one closes with instructions for extending it with your own firm’s numbers and policies.

The course rule is built into every one. The skill shows its formulas, refuses to invent a number it wasn’t given, stamps unsourced benchmarks UNVERIFIED, and names the moment to call a CPA, a lawyer, or a banker. Every worked example uses invented numbers. None of them contains a problem-set or case answer — don’t bother looking.

How to use a skill

  1. Download the zip. Inside: SKILL.md and a resources/ folder.
  2. Install it:
    • Claude (claude.ai): create a Project; upload SKILL.md and the resource files as project knowledge, or paste SKILL.md into the project instructions.
    • Claude Code / Agent SDK: drop the folder into ~/.claude/skills/. Done.
    • Any other AI tool: paste SKILL.md as your first message. It is self-contained.
  3. The resources/ templates work on their own — checklists, calculators, and log templates you can fill in with or without the AI.
  4. The skill drafts; you certify. Every package ships a verification-log template. Use it whenever a number matters.

Find your skill

Type to filter the cards below by name, topic, or keyword. Or browse the index table.
Skill The job it does From sessions
Startup Finance Infrastructure & Runway Stand up the finance function; always know the out-of-cash date S00 · S01 zip
Cap Table & Financing-Round Modeler Model rounds, convert SAFEs/notes, run the exit waterfall S00 · S02 · S11 · PS1 zip
Entity Choice, QSBS & Worker Classification Pick the entity, capture QSBS, classify workers correctly S03 zip
Equity Compensation Tax & Accounting Advisor Options, restricted stock, AMT, 83(b) — for the employee and the CFO S04 · S05 · S08 zip
Driver-Based Forecast with AI Verification Build a forecast with AI as analyst, you as certifier S06 · S07 · PS2 zip
Revenue Recognition & Margins (ASC 606) Recognize revenue correctly; read the bookings-revenue gap S09 zip
Board Package & Audit Readiness Draft, run, and red-team board packages; prep audit #1 S10 zip
Exit & M&A Deal Structuring Run the sale process; the peg, the earn-out, the step-up S12 zip

Download all eight (one zip)

Startup Finance Infrastructure & Runway

Built from S00 Bootcamp · S01 CFO & Infrastructure
Stand up a young company's finance function and read its cash clock.
  • Gross/net burn, runway, and the out-of-cash date — plus which lever actually buys more time
  • Cash-vs-accrual switch trigger, chart of accounts, software stack
  • Controls triad and the six-step monthly close, sized for a 10-person company
  • The "credibility engineer" play when a counterparty demands financials you don't have
Reach for it when: you're the first finance hire, a board asks "when do we run out?", or an acquirer wants three years of audited statements you don't have.
DownloadSKILL.md + 7 templates

Cap Table & Financing-Round Modeler

Built from S00 · S02 Cap Table Lab · S11 Fundraising · PS1
Build and defend a cap table through priced rounds — and know who gets what at exit.
  • Priced-round algebra, the option-pool shuffle, and the effective pre-money
  • SAFE and convertible-note conversion (caps, discounts, MFN) into the round
  • Liquidation waterfalls: preferences, participation, the conversion crossover
  • Term-sheet triage — the five terms worth your negotiating capital
Reach for it when: a term sheet lands, SAFEs are stacking up, or the board is weighing two exit offers and common's share is unclear.
DownloadSKILL.md + 7 templates

Entity Choice, QSBS & Worker Classification

Built from S03 Legal & People Infrastructure
The formation decisions that get made once and bind the company for a decade.
  • C-corp vs. LLC after-tax showdown, the indifference rate, and the deferral flip
  • QSBS §1202 under both regimes (post-July-2025 and grandfathered) — and the clock rules
  • W-2 vs. 1099 (the IRS control test) and exempt vs. non-exempt (the FLSA gates)
  • Pricing the misclassification bill before an auditor does
Reach for it when: you're incorporating, converting an LLC at a term sheet, or about to write "contractor" on an offer that walks like an employee.
DownloadSKILL.md + 7 templates

Equity Compensation Tax & Accounting Advisor

Built from S04 FeedBurner · S05 Equity Comp · S08 Solstice
What a grant really costs and pays — for the employee and for the CFO.
  • ISO vs. NQSO on every axis, and the fine print that ruins people (90-day windows, the 3-month statute)
  • The 5-line AMT worksheet for exercise-and-hold, with the phaseout shadow bracket
  • 83(b) elections, QSBS clocks, and the exercise-funding bridge
  • ASC 718 entries and why the 409A price sits below the preferred price
Reach for it when: you (or your team) get a grant, an exercise deadline, a tender offer — or you're the controller who has to withhold and book it.
DownloadSKILL.md + 9 templates

Driver-Based Forecast with AI Verification

Built from S06 KPIs & Budgeting · S07 AI Workshop · PS2
Build an operating forecast with AI doing the analyst work — and you certifying every number.
  • The four-role protocol (Analyst → Skeptic → Builder → Red-Team) with the six copy-ready prompts
  • The Five Manual Checks and the verification log — the discipline that makes AI use defensible
  • The unit-economics stack: CAC, NRR, LTV:CAC, payback vs. runway, burn multiple
  • Scenarios run through drivers, and a board one-pager where every number ties to a cell
Reach for it when: you own the annual plan, inherit an AI-built model you must present, or need to defend unit economics in a raise.
DownloadSKILL.md + 8 templates

Revenue Recognition & Margins (ASC 606)

Built from S09 Revenue Recognition
Turn contracts and a billing export into numbers an auditor and a diligence team will accept.
  • The revenue bridge: bookings → billings → GAAP revenue → deferred revenue
  • The ASC 606 five-step model as a scorecard, contract by contract
  • Gross vs. net (principal/agent) and why GMV is a KPI, never revenue
  • The margin ladder and commission capitalization done right
Reach for it when: sales says $12M, the model says $9M, the auditor says $7.4M — and the CEO wants to know which is "real."
DownloadSKILL.md + 7 templates

Board Package & Audit Readiness

Built from S10 Boards & Audit (leans on S06 · S07 · S09)
Draft, run, and red-team a board package — and be ready when the first audit comes.
  • The package spine: cash-burn-runway first, KPIs vs. plan, base/best/worst, asks
  • Picking the ~6 board KPIs — and the three kill tests for vanity and destructive metrics
  • The six-flaw red-team scan for any deck or investor update
  • Board composition by stage, the meeting rhythm, and the first-audit PBC list
Reach for it when: "board meeting in two weeks, can you pull the package together?" — or you're the one reading the deck before wiring money.
DownloadSKILL.md + 9 templates

Exit & M&A Deal Structuring

Built from S12 Exits (closes loops from S02 · S03 · S10 · S11)
The purchase price is not the price — manage what actually lands in the bank.
  • The sell-side funnel from teaser to close, with the gates that destroy value if skipped
  • The working-capital peg and dollar-for-dollar true-up, computed cold
  • Earn-outs, seller notes, and rollover equity — matched to the disagreement they solve
  • Stock vs. asset sale, the buyer's step-up shield, and the IPO decision framework
Reach for it when: an LOI lands, a buyer proposes an earn-out, or the board is split between selling and going public.
DownloadSKILL.md + 7 templates

No skills match that filter — try a broader term, or browse the index table above.

One warning. An AI running one of these skills is still an AI: it will be confidently wrong on schedule. The skills are built to catch that — every one ships a verification-log template — but the log only works if you use it. The draft is the machine’s; the number is yours.

The PS2 forecasting scaffold

The AI-assisted forecasting module (Sessions 6–7 and Problem Set 2) uses a structured scaffold: four AI roles, copy-ready prompts, a graded verification log, and five manual checks. AI is required on PS2; every number you submit must be human-certified. (The Driver-Based Forecast skill above generalizes the full protocol for use after the course.)

AI-for-FP&A ground rules

  1. The AI never touches your Excel file. Paste text in; implement schedules yourself.
  2. Four roles, four jobs: Analyst → Skeptic → Builder → Board Red-Team (plus a Drafter for the memo).
  3. Every external benchmark must be traced to the professor’s benchmark sheet or stamped UNVERIFIED in your log.
  4. Every number in your memo must exist in your model — unverifiable numbers cost points.
  5. The final exam is closed book, no AI — the backstop that keeps incentives honest.

Four role cards

Analyst — driver extraction

Bias / blind spot: Over-extracts plausible-sounding drivers not in the case text.

Verification: Cross-check every extracted assumption against the case narrative.

Skeptic — assumption benchmarking

Bias / blind spot: Cites benchmarks confidently without real sources.

Verification: Provenance rule — source every external number or stamp UNVERIFIED.

Builder — schedule logic drafting

Bias / blind spot: Outputs formulas that look right but violate timing rules (ramp, lag, hiring).

Verification: Hand-recompute one month; tie out beginning + adds − churn = ending.

Board Red-Team — forecast critique

Bias / blind spot: Asks generic board questions instead of finding internal inconsistencies.

Verification: Run a one-variable data table on the AI’s “most sensitive driver” claim.

Copy-ready prompts

Use these in any chatbot (free tier is sufficient). Paste your case facts between ### delimiters.

P1 — Driver identification (Analyst)

You are an FP&A analyst at an early-stage SaaS company. Below is our planning case, delimited by ###. Extract every quantitative assumption into a table with columns: driver name | value | unit | financial statement line it feeds | controllable by management (Y/N). Then present the driver tree from sales capacity to ending ARR as an indented list. Do not invent any assumption that is not in the text; instead, end with a list titled "OPEN QUESTIONS" for anything the case leaves unspecified. ### [paste case_facts.txt] ###

P2 — Assumption benchmarking (Skeptic)

You are a skeptical FP&A reviewer. Here is our assumption table: [paste]. For each assumption, classify it as aggressive / typical / conservative for a Series-A B2B SaaS company in 2026, with one sentence of reasoning. Wherever you invoke an external benchmark, name the source and your confidence in it, and tag the line "VERIFY BEFORE USE." Finish by naming the three assumptions that, if wrong, do the most damage to year-end cash, and why.

P3 — Schedule drafting (Builder)

Build a monthly FY2027 customer-count schedule from these assumptions: [paste relevant rows]. Output (a) a 12-column table with beginning balance, new logos, churned logos, ending balance; and (b) for each row, the formula logic in plain words so I can implement it in Excel myself. Explicitly flag any month where a ramp, training lag, or hiring rule changes the calculation. Do not give me Excel formulas with cell references — I own the spreadsheet.

P4 — Scenario building

Create base, upside, and downside scenarios by varying ONLY these four drivers: [list]. For each scenario, give the driver values and a one-sentence rationale grounded in the case facts — no new facts. Present as a table I can paste into my assumptions sheet. Then state which single driver you believe moves December cash the most; I will test your claim with a data table and report back.

P5 — Forecast critique (Board Red-Team)

You are a board director with a venture background reviewing a Series-A company's annual operating plan. Here are our forecast outputs: [paste one-pager numbers]. Ask the five hardest questions you would ask in the board meeting. Then identify any internal inconsistencies — revenue growth vs. sales headcount, CAC vs. marketing spend, support ratio vs. customer count, commission expense vs. bookings. Be direct; do not soften.

P6 — Board narrative (Drafter)

Draft a 250-word board narrative from ONLY the following verified numbers: [paste]. Structure: the headline, the base case, the single biggest risk, and the pre-agreed trigger that would cause us to re-forecast. If a sentence would require a number I did not give you, leave a bracketed placeholder instead of inventing one.

Five manual checks

Each must appear in your PS2 verification log:

  1. Extraction fidelity — every AI-listed assumption traced to the case text.
  2. Roll-forward integrity — beginning + additions − churn = ending, all 12 months; cash walk ties to net income ± working capital.
  3. Benchmark provenance — every external number sourced or stamped UNVERIFIED.
  4. Sensitivity replication — your data table confirms or refutes the AI’s “most sensitive driver” claim.
  5. Column recompute — hand-recompute December, top to bottom.

Downloads

File Status
Verification log template (Excel) On the Problem Sets page with the full PS2 Model Pack
Benchmark sheet (PDF) On the Problem Sets page with the full PS2 Model Pack
Prompt card (PDF) The P1–P6 prompts, print form — content is also on this page

The full workshop kit and Model Pack are on the Problem Sets page; submissions should be uploaded to Canvas for grading.