Every pattern, in one place
The workflow pattern and prompt templates from every module, collected as a reference you can search and copy at your desk. Each pattern is the same shape: inputs, an AI step, a human checkpoint, and a finished artifact. Take the prompt, swap in your own folder, and keep the validation discipline the module taught.
A prompt is a starting point, not a shortcut past the review. Every pattern here still ends at a human checkpoint: trace the numbers, verify the sources, and own the conclusion.
Module 0
Finance AI Foundation
You start with a recurring finance task you do by hand. You want a repeatable workflow you can trust. The route is the five moves.
A recurring task and the materials you normally start from: last period’s inputs, a blank template, and one finished example of the output you are aiming for.
Name the deterministic parts (the math, the roll-forward, the mapping) and have AI build a formula or template for those. Have AI perform only the non-deterministic parts (the narrative, the summary, the classification).
Here is last period’s input and the finished output as an example, plus this period’s input. The variance math is deterministic: give me the formula, do not compute it in prose. Then draft only the written commentary, using the computed figures.
Run the red-lines check before the task, then the five-point review after: verify every source, own every conclusion, strip AI writing patterns, trace every number, and reconcile.
A finished, defensible deliverable, and a repeatable pattern (or skill) you can rerun next period with fresh inputs.
Module 1
Close & Reporting
You start with a trial balance and last month’s numbers. You want close-ready flux commentary that ties. The route splits the math from the narrative.
The current trial balance, the prior-period trial balance, a flux template, and one example of last quarter’s finished commentary as the destination.
The template computes each account group’s variance in dollars and percent (deterministic). The model then drafts commentary for only the variances above materiality, naming a plausible driver for each.
Using the computed variances in the flux template (do not recompute them), draft close-ready commentary for each account group whose variance exceeds the materiality threshold. Name a specific driver for each and stay silent on immaterial movements. Match the tone of the example commentary in the folder.
Tie each figure in the commentary back to the trial balance, confirm materiality was respected, and check that every named driver is real and not fabricated.
Flux commentary a controller could send onward: reconciled, materiality-screened, correctly toned, and free of invented drivers.
Module 2
Variance & Driver Trees
You start with a favorable revenue variance and a budget. You want a decision-quality explanation of why, not just by how much. The route decomposes the variance, then reasons about cause.
Actual versus budget by product line, with the price and volume variance already decomposed, plus the materials cost line.
The model reads the decomposed variances and names the primary driver per line, then produces at least two alternative hypotheses and what evidence would distinguish them.
Using the decomposed price and volume variances in the bridge, explain the revenue beat by line. Then give two alternative hypotheses for what is really driving it (for example, a durable price gain versus a one-time stocking order) and state what evidence would tell them apart. Do not recompute the variances.
Confirm the price-volume math ties, and pressure-test whether the alternative hypotheses are genuinely distinct and testable rather than restatements.
An explanation that separates price from volume, sizes each, and carries a decision-quality layer of alternatives.
Module 3
Research & Benchmarking
You start with excerpted peer filings and metric definitions. You want a cited benchmark memo where every claim is fact or labeled inference. The route is extract, verify, cite.
Excerpted figures for the peer set plus the metric definitions to apply consistently.
The model computes each metric from the provided figures and tags every statement as a fact (cited) or an inference (reasoned).
Using only the provided filings and the metric definitions, build a benchmark of gross margin, revenue growth, and R&D intensity across the set. Tag every claim as FACT (cite the source line) or INFERENCE (label it). Where a definition difference makes a metric not comparable, say so rather than estimating.
Open each cited figure and confirm it resolves to the provided filings, and that inferences are labeled.
A one-page benchmark where facts are cited and inferences are labeled.
Module 4
Working Capital & Cash
You start with agings and a cash skeleton. You want a validated 13-week forecast and scenarios. The route is the deterministic split: the model builds the model, you validate it, then it narrates.
AR and AP agings plus a 13-week cash skeleton (beginning cash, collections, disbursements, ending cash).
The model builds and holds the 13-week arithmetic; once you validate it, the model narrates scenarios on top (a collections slip, a disbursement delay).
Validate this 13-week cash model: confirm each week chains (ending equals beginning plus collections minus disbursements, and each week begins where the last ended). Identify the trough week and minimum cash. Then narrate two scenarios (a 10 percent collections slip; a one-week disbursement delay) and name the levers that protect the minimum-cash line. Do not restate the arithmetic in prose.
Confirm the weekly chain ties and the trough is correct before trusting any scenario built on top.
A 13-week forecast with the trough identified and two scenarios with levers and actions.
Module 5
Cost & Spend Analytics
You start with a vendor spend cube. You want a defensible savings opportunity with every assumption written down. The route is total, rank, size, and document.
Annual spend by vendor and category.
The model totals and ranks the spend, finds the fragmented categories, and sizes an opportunity with every savings number tied to a stated assumption.
Total and rank this spend by category. Identify the most fragmented categories (many small vendors), where consolidation savings usually live. Size an opportunity, and for every savings number write the assumption behind it in an assumptions ledger (for example, a consolidation rate). Keep the narrative governance-safe: no invented external benchmarks and no vendor defamation.
Confirm each savings figure ties to an assumption in the ledger, and the narrative names no benchmark it cannot support.
A sized, governance-safe opportunity where every number is traceable to an assumption.
Module 6
Anomaly Detection
You start with a messy GL extract, mostly clean. You want an exception report and an audit trail. The route is rule-based testing with objective scoring.
A month of general-ledger entries: document number, date, vendor, account, description, debit, credit.
The model applies exception rules (duplicates, weekend postings, round-dollar entries, sign or account errors) and lists the entries that warrant review.
Run exception-based tests on this GL extract: duplicate payments (same vendor and amount), weekend postings, round-dollar entries, and revenue accounts carrying a debit. For each flagged entry give the document number, the rule it failed, and a proposed disposition. Do not flag clean entries.
Confirm each flagged entry actually fails a rule, and scan for anything the model missed. The scoring is objective.
A report of flagged entries plus an audit-trail note: tests run, thresholds, findings, disposition.
Module 7
Technical Memos
You start with a fact pattern and guidance excerpts. You want a supported issues memo, and a redline of a flawed AI memo. The route is draft, critique, revise.
The fact pattern, simplified guidance excerpts, and an AI-drafted memo with planted errors to review.
The model drafts a supported issues memo from the facts and guidance, then critiques the provided flawed memo clause by clause.
From the fact pattern and the guidance excerpts, draft a short issues memo: identify the issue, map it to the guidance paragraph, reach a supported position, and state the alternative you considered. Then review the provided AI memo and produce a redline: for each error, say what is wrong, why, and what it should say. Cite only the provided guidance.
Confirm each cited paragraph is correct and the conclusion follows from the criteria, not from a single fact.
An issues memo and a peer-review redline of the flawed memo.
Module 8
Contract Review
You start with a set of signed contracts. You want a clause-extraction and accounting-implication read with confidence ratings. The route is extract, map to guidance, rate.
Three fictional contracts: a SaaS subscription, an equipment-plus-service bundle, and a license with variable consideration.
The model pulls the accounting-relevant clauses, quoting the contract language verbatim, and maps each to its ASC 606 implication as an issue to resolve.
For each contract, extract the clauses that drive revenue recognition. Quote the contract language verbatim for each. Map each to its ASC 606 implication (performance obligations, timing, variable consideration) as an ISSUE to resolve, not a conclusion. Give a confidence rating (high, medium, low) and list open questions for the accounting team.
Confirm each extracted clause quotes real contract language, and that implications are framed as issues, not settled conclusions.
A first-pass read the accounting team can turn into memos, with confidence ratings and open questions.
Module 9
Valuation & Scenarios
You start with historicals and assumptions. You want a validated DCF and a sensitivity narrative. The route is the split-it thesis at full scale: the model builds the skeleton, you validate it, then it narrates.
The assumptions ledger (growth, margin, tax, capital, discount) and three years of historical financials.
The model builds the five-year free-cash-flow build and terminal value from the ledger; once you validate the math, it writes the sensitivity narrative.
Build a five-year free-cash-flow model and a Gordon-growth terminal value from the assumptions ledger, and compute enterprise value. Show every figure traceable to a ledger assumption. Then write a sensitivity narrative: how enterprise value moves with WACC and terminal growth, and a board pre-read framing of the base case and swing factors. Do not restate the arithmetic in prose.
Confirm every DCF figure traces to a ledger assumption and enterprise value recomputes before trusting the narrative.
An enterprise value with a sensitivity narrative and board framing.
Capstone
Capstone Simulation
You start with one month of Meridian data. You want a single, defensible board package. The route chains four workflows you have already run into one integrated deliverable.
One month of Meridian: the trial balances, the price-volume bridge, the GL extract, and a technical-accounting issue.
Produce the close-ready flux commentary, the driver-based variance explanation, the exception-review note, and one memo issue, then a one-page board summary that ties them together.
Using the folder, produce, in one package: (1) close-ready flux commentary that ties to the trial balance; (2) a price-versus-volume explanation of the revenue beat with one alternative hypothesis; (3) an exception-review note on the GL with each anomaly and its disposition; (4) one technical-accounting issue with a supported position and the alternative; and (5) a one-page board summary. Reuse the discipline from each earlier module; do not introduce new methods.
Confirm every number ties across components, every driver is supported, every anomaly is real, the memo cites correctly, and the summary is honest about uncertainty.
One integrated, reconciled deliverable graded against all eight learning objectives.
Appendix
Financial Due Diligence
You start with reported EBITDA and a set of adjustments. You want a normalized earnings number and a working-capital peg. The route is bridge, support, and peg.
Reported EBITDA, the candidate adjustments, and 12 months of net working capital.
The model totals the bridge from reported to adjusted EBITDA and drafts the support note; each add-back is labeled one-time or run-rate.
Build a quality-of-earnings bridge from reported to adjusted EBITDA using the adjustment schedule. Label each add-back one-time or run-rate and note its support. Compute the 12-month average net working capital as the peg. Do not invent adjustments; use only the schedule.
Confirm the bridge foots and each adjustment has a defensible basis; a buyer will challenge each one.
A normalized EBITDA bridge, a working-capital peg, and the diligence questions raised.
Appendix
Procurement & Sourcing
You start with vendor bids and a should-cost estimate. You want a best-value recommendation and negotiation levers. The route is total, compare, and reason.
Three vendor bids broken into cost elements plus an independent should-cost buildup.
The model totals each bid, computes the gap to should-cost, and drafts a best-value recommendation with negotiation levers.
Total each vendor bid and compute its gap to the should-cost estimate by cost element. Recommend the best value (not simply the lowest price), state the annualized savings versus the highest bid, and name the negotiation levers the should-cost reveals (for example an out-of-line overhead or margin). Do not invent an external benchmark.
Confirm the totals foot and the best-value logic holds beyond price alone.
A bid comparison, a best-value pick, and the negotiation levers.
Appendix
FP&A Operating Model
You start with a driver-based plan. You want a validated operating model and scenarios. The route is the deterministic split: the model builds the arithmetic, you own the assumptions.
A segment plan (units, price, cost ratio) with the totals shown.
The model builds and holds the arithmetic (revenue equals units times price, cost ties to the ratio, operating income foots) and narrates the drivers and scenarios.
Validate this operating plan: confirm revenue equals units times price by segment, COGS ties to the cost ratio, and operating income foots. Then name the assumptions that move operating income the most, and narrate two scenarios (a price-hold and a volume-miss) with the operating-income impact. Do not restate the arithmetic in prose.
Confirm the plan ties and that the assumptions are yours to defend, not the model’s to invent.
An operating plan with driver commentary and two scenarios.