AI vs the 10-K: How to Conduct a Forensic Examination of Company Accounts (Part 1)
Automate Quarterly and Annual Report Reviews without Missing the Nuance
This Part 1 of a 2 part post due to length and detail. Part 2 can be found here.
Many professional and individual investors read the Inferential Investor posts but aren’t yet subscribers. Click here to read why an II subscription is valuable to you or subscribe free with your email below, so you don’t miss out on a technique that will identify the next idea or save you from a costly investing mistake.
If you ask professional investors where edge still exists, many will point to the plain-vanilla 10-K & 10-Q (annual and quarterly reports for non-US investors). Not because others don’t read it, but because most readers skim the highlights and miss the connective tissue between footnotes, revenue recognition, accruals, cash flow idiosyncrasies, and management incentives. It is incredibly time consuming to conduct this work so many shelve it - to their detriment when short seller reports appear, class actions or surprise write-downs and downgrades.
Generative AI models mean you can conduct this work in as little as five minutes with the correct prompts. The risk is that speed without structure turns into superficiality. This piece lays out a practical, professional and tested way to pair classic financial-statement analysis with a prompt set that keeps nuance intact. The Inferential Investor prompt library already has a comprehensive, tested, forensic report master-prompt, bringing all the steps below together in a single copy-paste exercise. This will be available soon as more prompts are published.
The organizing idea is simple: for each core analysis that practitioners actually run - accruals quality, cash conversion, revenue recognition tests, capitalization choices, working capital stress tests, leverage and covenant risk, off-balance-sheet items, stock-based compensation and dilution, segment economics, auditor signals, and management discussion language, I’ll outline (1) why it matters, (2) the goal, (3) a manual checklist rooted in textbooks and best practice, and (4) a tested prompt pattern you can use to automate analysis and check your own logic.
We’ll reference principles from Graham & Dodd’s Security Analysis and Graham’s The Intelligent Investor (quality of earnings, margin of safety), Penman’s Financial Statement Analysis and Security Valuation (clean-surplus and residual income logic), Koller et al.’s Valuation (cash flow over accounting earnings, ROIC vs. growth), White, Sondhi & Fried’s The Analysis and Use of Financial Statements (ties between statements), and Schilit & Perler’s Financial Shenanigans (pattern recognition for red flags).
The point isn’t to replace judgment. It’s to let the machine handle extraction, cross-referencing, and first-pass diagnostics so you can concentrate on materiality and causality.
The following prompts are condensed for brevity and incorporate acronyms. The final copy/past prompt in the library is execution ready.
1) Accruals Quality: Are Earnings Built on Cash or Accounting
Why it matters
The accruals anomaly is one of the best-documented facts in empirical finance: companies with high accruals (big gap between earnings and cash) tend to underperform. Graham’s quality-of-earnings lens and Penman’s clean-surplus thinking both push you toward cash-validated earnings. If net income rises while operating cash flow stalls, you need to know why.
Goal
Quantify and explain the earnings–cash gap. Classify drivers (working capital timing vs. discretionary accruals) and judge sustainability.
Manual checklist
Reconcile net income to CFO; compute total and working-capital accruals as % of assets.
Identify drivers: receivables growth vs. revenue, inventory vs. COGS, payables vs. COGS.
Check cash taxes vs. book taxes, and deferred tax movements.
Tie CFO to revenue timing (billings vs. collections) and to any policy changes in the notes.
Cross-check with auditor critical audit matters (CAMs) on revenue recognition or reserves.
Prompt set (condensed)
Role & objective: You are an equity analyst performing accruals-quality diagnostics on a single company’s [10-K]. Inputs: Item 8 financials (income statement, balance sheet, cash flow), Notes on revenue recognition, inventories, receivables, taxes; MD&A liquidity section. Tasks 1. Extract: net income, CFO, CFI, CFF, AR, Inventory, AP, current taxes paid, deferred tax changes (last 3 years). 2. Compute: total accruals = NI − CFO; working-capital accruals = ΔAR + ΔInventory − ΔAP (use average assets scaling). 3. Diagnose: attribute accruals to specific line items; flag unusual jumps (>1 SD vs. 3-yr history). 4. Explain: summarize management’s stated reasons (quote short phrases) and whether they plausibly map to the numbers. Output: A table with metrics and a short, non-speculative interpretation with 2–3 alternative hypotheses to test.
Quick anecdote
In the late 1990s, Sunbeam reported strong profit while receivables ballooned. Cash conversion told a different story. Investors who tested accruals early were spared large losses when earnings were restated.
2) Cash Conversion and Free Cash Flow Quality
Why it matters
Koller’s Valuation centers on cash return on invested capital. If net income grows but cash conversion declines, growth may be value-destructive. Graham would call this the difference between an “earnings picture” and owner earnings.
Goal
Measure cash conversion of earnings and the quality of free cash flow (FCF) after maintenance investment.
Manual checklist
CFO/NI ratio over time; FCF = CFO − maintenance capex (estimate maintenance via depreciation or management guidance).
Separate working capital swings from structural shifts (payment terms, pricing power).
Identify “cash add-backs” such as customer prepayments or deferrals that might unwind.
Compare cash taxes to statutory expectations.
Prompt (continued)
Extract CFO, capex (gross and net), depreciation & amortization, cash taxes, and working capital detail for 5 years. Compute CFO/NI, FCF/NI, and FCF/Revenue; decompose ΔCFO into ΔWC vs. core. Identify one-off cash sources (prepayments, litigation proceeds) and tag them as non-recurring. Output a short memo: “Is FCF quality strengthening, flat, or weakening? Why, with evidence lines.”
Example
Subscription media businesses often benefit from upfront cash (annual plans). AI can tag these prepayments in notes and MD&A so you don’t misread “strong cash conversion” that is actually deferred revenue growth.
3) Revenue Recognition: Timing, Estimation, and Risks
Why it matters
Schilit’s revenue games are classic: channel stuffing, bill-and-hold, gross vs. net, multi-element arrangements. Under ASC 606/IFRS 15, estimation judgments (variable consideration, churn) place weight on management assumptions.
Goal
Understand when revenue is recognized, what estimates drive it, and where risk of reversal sits.
Manual checklist
Read the revenue policy note end-to-end. Highlight variable consideration, returns, rebates, breakage.
Map billings to revenue and deferred revenue.
Examine receivables aging disclosures and allowance methods.
Scan auditor CAMs for revenue-related risks.
Prompt
Extract the revenue policy note and summarize each revenue stream’s recognition trigger (delivery, over-time input/output method, etc.). Identify estimates (refunds, chargebacks, usage-based fees) and controls mentioned. Cross-link to the balance sheet: AR aging, allowance; deferred revenue rollforward. Output a risk map: “High/Medium/Low” per stream with a 2-sentence justification and the exact paragraph reference.
Anecdote
When a large e-commerce platform shifted its mix toward third-party sellers, net revenue growth slowed (a market concern for a while) relative to GMV, but working capital needs fell and cash conversion rose. The automated segment analysis would be able to flag margin expansion via mix, even as the revenue-policy prompt confirmed “3-rd party agent” treatment. The human conclusion: higher ROIC at the cost of optics on top-line.
Groupon in its early years recorded gross revenues from merchant deals before switching to net presentation. Analysts who studied the footnotes caught the inflated growth story.
4) Capitalization Choices: Capex, R&D, and Content Assets
Why it matters
Penman emphasizes matching and the investment/expense boundary. Capitalizing costs boosts current earnings and depresses future ones via amortization; expensing does the reverse. Industry norms differ: software often expenses R&D; media capitalizes content; telcos capitalize network build.
Goal
Spot where management draws the line and whether choices are consistent with economic reality.
Manual checklist
Identify capitalized development costs vs. expensed R&D.
Review amortization periods and impairment tests for capitalized intangibles or content.
Compare capex guidance with growth narrative (capacity vs. maintenance).
Track capitalization rates over time.
Prompt set
Extract R&D note, intangible assets note, PP&E note. Compute capitalization rate = capitalized development / (capitalized development + expensed R&D). Compare amortization periods to peer medians (if provided) and flag outliers. Summarize impairment triggers and the last impairment event, with page cites. Output: a 1-paragraph brief on whether capitalization choices inflate current margins, and what reversal risks exist.
Example
Streaming platforms capitalizing content saw periodic amortization tweaks to reflect viewership half-lives. When amortization schedules were lengthened, reported profit improved temporarily but future write-downs loomed as viewership ultimately shortened with rising streaming competition. AI can spot a change in useful lives and quantify the effect on current vs. future margin.
5) Working Capital Stress Tests: Receivables, Inventory, Payables
Why it matters
Graham’s margin of safety exists not just in valuation but in liquid resources. Inventory spikes against slowing sales, receivables outpacing revenue, or payables stretched to fund growth are early stress signals.
Goal
Assess whether growth is being “leased” from suppliers or customers, and whether the cash conversion cycle is deteriorating.
Manual checklist
AR turnover and Days Sales Outstanding (DSO).
Inventory turns and obsolescence reserves.
Days Payables Outstanding (DPO).
Net working capital as % of sales, trend and seasonality.
Prompt set
Extract revenue, AR, Inventory, AP, COGS; compute DSO/DIO/DPO for 5 years. Flag shifts >10% y/y and annotate with management discussion and analysis explanations. Output a succinct table plus “top 3 hypotheses” that an analyst should verify manually (e.g., channel expansion, SKU mix, supplier renegotiation).
Anecdote
Several consumer hardware roll-ups grew by extending payable days materially. It worked unitl economic pressure forced supplier terms to snap back, erasing apparent free cash flow.
In 2012, Hewlett-Packard wrote down billions of dollars after receivables and inventory surged relative to revenue. Ratios flagged the issue well before management did.
6) Leverage, Covenants, and Liquidity
Why it matters
Graham urged a balance-sheet first mindset. Debt structure, covenants, and maturities turn small operating shocks into equity wipeouts.
Goal
Understand debt terms, early-warning triggers, and refinancing cliffs.
Manual checklist
Build a maturity ladder.
Note covenant types (leverage, interest coverage, liquidity) and headroom.
Check variable vs. fixed rate mix and hedges.
Tie interest expense to average debt to catch rate drift.
Prompt set
Extract debt footnote and MD&A liquidity. Build a table of maturities by year, coupon, rate type, covenants. Compute interest coverage (EBIT/interest) and sensitivity to +100 bps. Output: a ladder visualization description and a paragraph on “most binding constraint.”
Example
Airlines often carry seasonal working capital swings and flirt with covenant headroom. A seasonal dip in cash flow can trigger covenant breaches even if annual averages look safe.; covenant headroom in the trough matters more than annual averages. AI can calculate trough metrics if you supply quarterly data.
7) Off-Balance-Sheet Exposures, Committments and Leases
Why it matters
SPEs, guarantees, purchase commitments, and leases can hide leverage. Post ASC 842, most leases are on-balance-sheet, but short-term and variable leases still require judgment.
Goal
Catalog commitments and their cash impact; evaluate whether economics differ from accounting presentation.
Manual checklist
Lease note: terms, discount rate, maturity profile.
Purchase and take-or-pay commitments.
Guarantees and indemnities.
Unconsolidated affiliates.
Prompt set
Extract commitments & contingencies and leases notes. Summarize total undiscounted commitments by category and year. Estimate present value of off-balance-sheet items where feasible. Output: a one-page “shadow leverage” summary with bullets on materiality.
Anecdote
Energy midstream firms historically signed long-dated volume commitments. In downturns, minimums can create cash drains as volumes fall. Identification of these committments arms the investor with more information.
In the early 2000s, Enron’s use of off-balance-sheet partnerships concealed billions in purchase obligations. Analysts who dug into commitments notes could have seen the scale of exposure.
At this point we have covered the first seven pillars of forensic financial statement analysis with AI: accruals, cash conversion, revenue recognition, capitalization, working capital, leverage, and off-balance-sheet risks. These address the core financial architecture of a business.
In Part Two, we will move beyond the balance sheet into signals and governance: stock-based compensation, segment economics, auditor’s opinions, MD&A language, related-party transactions, reserves, and guidance. We will conclude with a master prompt for generating a complete forensic report automatically.
Remember - Part 2 is coming so keep an eye out.
As always, Inference never stops. Neither should you.
Andy West
The Inferential Investor