aaron

Quick win: Paste this into any AI and ask for a single compelling webinar title plus a one-line promise — you’ll have a working hook in under 5 minutes.

Problem: most people start writing slides before they have a clear hook, audience or CTA. The result is low signups, low attendance and wasted time.

Why it matters: one clear title and promise focus every element — outline, slides, emails and social — so your promotion converts and your webinar delivers actionable value that leads to the outcome you want.

What I recommend (what you’ll need)

• Working topic and target audience (e.g., “small business owners 40+ selling local services”).
• Outcome you want (bookings, paid course signups, leads).
• Logistics: date/time, length (30–60 min), platform (Zoom/Teams).
• 15–45 minutes to run AI prompts and edit outputs.

Step-by-step (how to do it & what to expect)

1. Decide the single goal for attendees (what they should do next).
2. Use the AI prompt below to generate: title, one-line promise, timed outline, 10 slide headlines, speaker notes, 3-email promo sequence (with subject lines + preview text) and 3 social posts. Expect 1–3 strong drafts you can refine in 15–30 minutes.
3. Edit the AI speaker notes to add 2 real examples or a short case study — this makes the voice human.
4. Create slides from the 10 headlines; paste AI notes into slide speaker notes and rehearse once aloud (15–30 minutes).
5. Schedule the first invite email and one social post now to validate messaging. Run A/B on subject line if possible (two quick variants).

Copy-paste AI prompt (use as-is)

“I’m planning a [length: 30–60 min] webinar for [audience — be specific]. Topic: [insert topic]. My goal: [what I want attendees to do after]. Produce: 1) a short compelling title, 2) a one-sentence promise, 3) a timed outline with speaker notes for each section, 4) 10 concise slide headlines with 1–2 sentence notes, 5) a 3-email promotional sequence (invite, reminder, last-chance) with subject lines and preview text and two subject-line variants for A/B testing, and 6) three short social posts (LinkedIn/Facebook/Twitter styles). Keep tone clear, practical and persuasive.”

Metrics to track

• Registration rate (signups per email/social impressions).
• Email open rate and subject-line A/B winner.
• Show rate (attendees ÷ registrants).
• Conversion after the webinar (your goal: purchases/appointments/subscriptions).

Mistakes & fixes

• Vague audience = generic copy. Fix: add age, role, top pain, and desired outcome to the prompt.
• Using AI text verbatim = robotic. Fix: insert one personal example and one plain-language line in every section.
• No clear CTA = low conversions. Fix: finish the webinar with a single, simple next step and repeat it 3 times.

7-day action plan

1. Day 1: Run the prompt, pick a title and one-sentence promise. Schedule date/time.
2. Day 2: Generate slides and speaker notes; rehearse once.
3. Day 3: Finalize CTA and landing page/register form.
4. Day 4: Schedule invite email + social posts (include A/B subject test).
5. Day 5: Send reminder email; test tech setup.
6. Day 6: Dry run with 2–3 people; collect feedback.
7. Day 7: Run the webinar; follow up with replay and CTA email.

Your move.

Hook: Good call — starting with a crisp headline is the single best way to make status reports useful. It forces you to answer “what changed?” before anything else.

The problem: messy notes, missed owners, and optimistic dates turn status updates into guesswork and slow decisions.

Why this matters: clear, repeatable updates reduce follow‑up friction, shorten decision cycles, and make your team look reliable. That’s measurable: fewer clarification emails, faster approvals, fewer missed deadlines.

Quick lesson I use: use a one‑line headline + 3 short sections and a confidence tag. It takes 5–10 minutes and prevents 30–60 minutes of back‑and‑forth later.

1. What you’ll need
   • Recent notes (last 1–2 weeks).
   • A phone or laptop and an AI summarizer or chat tool.
   • A short template: Headline, Progress, Blockers/Risks, Next Steps (owner + ETA), Confidence (High/Medium/Low).
2. Step‑by‑step
   1. Gather: Put the latest notes in one file — limit to 1–2 weeks.
   2. Tag: Add Owner: and ETA: next to items you care about (takes 1–2 minutes).
   3. Seed the headline: write one sentence that answers “what changed?” — that’s your pivot.
   4. Ask the AI to expand the seed into the 4 sections (see prompt below).
   5. Verify: 2–3 minute check on owners, dates, and a single metric (budget/hours/ETA).
   6. Publish: paste into email or project tool; save cleaned notes for the next run.

Copy‑paste AI prompt (use as written) — paste into your AI tool with the notes below the prompt:

“Create a concise project status report using the provided notes. Produce: 1) One‑sentence headline that answers ‘what changed’, 2) 2–3 progress bullets, 3) 1–2 blockers/risks, 4) Up to 3 next steps with Owner: and ETA:, 5) One‑word confidence (High/Medium/Low). Keep each item short and factual. Don’t invent dates or owners — if missing, mark ‘Owner: TBD’.”

Metrics to track

• Time to draft (target: <10 minutes)
• Edit time (target: <5 minutes)
• Number of clarification replies after sending (target: 0–1)
• Decision lead time (average days from report to decision)

Common mistakes & fixes

• Missing owners — fix: tag Owner: in your notes before summarizing.
• Optimistic dates — fix: add a conservative ETA or mark as tentative.
• Too much context — fix: keep progress strictly outcome‑focused (what was done, not the whole story).

1‑week action plan

1. Day 1: Pick one active project and run the 2‑sentence headline routine for a single note. Time and record edits.
2. Day 2–3: Apply the AI prompt to two more weekly notes; refine which items you keep.
3. Day 4–5: Add the Confidence tag; compare clarification emails vs. Day 1.
4. End of week: Review metrics (draft time, edit time, clarifications) and pick one tweak to the template.

Closing — your next step: run the prompt above on one recent note now, track time, and report one metric back (draft time or clarifications). Your move.

— Aaron Agius

Smart call on the change log and the “don’t rewrite” rule. Here’s how to lock in your voice with guardrails that make drift almost impossible—and show you numbers to prove it.

Core idea: freeze your voice first, then proofread. Build a light “voice fingerprint” (simple metrics), cap how much the AI can change, and ask for hard stats after each pass.

High-value move: the Voice Fingerprint

• Contractions target (e.g., 70–90% of eligible cases)
• Average sentence length (e.g., 12–16 words)
• Never-change list (idioms, signature phrases, product names)

Copy-paste prompt — build your fingerprint (60 seconds)

“Analyze the following paragraph and create a simple voice fingerprint. Return: 1) contraction rate (% of eligible contractions used), 2) average sentence length (in words), 3) tone in 3 plain adjectives, 4) a list of 5–10 phrases or idioms that feel signature to the voice, 5) a one-paragraph style guide I can reuse. Text: [paste a representative paragraph you like].”

Copy-paste prompt — strict grammar with edit cap + stats

“You are a careful copy-editor. Fix grammar, punctuation, and clarity only. Respect my voice fingerprint. Do not change metaphors, idioms, or brand terms. Keep sentence structure unless a fix is required for correctness. Edit cap: change no more than 10% of words. Keep contractions within ±5% of the target. Keep average sentence length within ±2 words of the target. Return 4 sections: 1) corrected text, 2) change log with brief reasons, 3) any sentences you’re unsure about (max 2) with one suggested alternative each, 4) stats: words changed, % changed, contraction rate, average sentence length, any rule you couldn’t follow. Voice fingerprint/style guide: [paste your fingerprint/style guide]. Now edit this text: [paste your draft].”

Variant — delta-only view (fast approval)

“Same instructions and caps. Return two sections only: 1) corrected text with brackets around changed words so I can see edits inline, 2) stats (words changed, % changed, contraction rate, average sentence length). No rewrites beyond grammar.”

Why this matters: You keep rhythm, word choice, and idioms while fixing errors. The edit cap and stats prevent silent voice drift. Your acceptance rate goes up, and review time shrinks.

What you’ll need

• One paragraph you love (to build the fingerprint)
• Your never-change list (idioms, product names, signature lines)
• A short draft (100–300 words) for the first run

Step-by-step

1. Fingerprint in one minute: run the fingerprint prompt on a paragraph you’re proud of. Save the style guide and targets.
2. Set your rules: keep contractions, keep first-person, don’t change metaphors/idioms, edit cap 10% words, sentence length ±2 words.
3. Run the strict grammar prompt. Expect corrected text, change log, and stats. If % changed is over your cap, ask the AI to roll back the least essential edits first.
4. Approve quickly: skim the change log; accept anything that fixes correctness and preserves rhythm. Reject anything that swaps words for style, unless you requested it.
5. Second pass (optional): only for the 1–2 flagged unclear sentences. Approve if it still sounds like you.
6. Save the winning fingerprint and rules as your template for future pieces.

What to expect: A clean, readable draft that sounds like you, with edits you can see and stats that confirm nothing drifted. Over time, your acceptance rate improves and your edit cap can drop to 5%.

Metrics to track (weekly)

• % words changed: target ≤10% (then tighten to ≤5%)
• Contraction rate delta vs. target: within ±5%
• Average sentence length delta: within ±2 words
• Acceptance rate of AI suggestions: 70–90% is healthy
• Time to final: aim for 10–20 minutes per short piece
• Reader response: replies or comments per 100 sends/posts

Common mistakes and fast fixes

• AI over-formalizes. Add “prefer short, everyday words; keep contractions,” and restate your never-change idioms.
• Too many edits. Lower the cap to 5–8% and request a rollback of low-value changes first.
• Hidden rewrites. Use the delta-only variant or require brackets around changed words.
• Loss of rhythm. Add a rule: “Do not change sentence starts or cadence unless grammar requires it.”
• Inconsistent outputs. Always paste the fingerprint at the top of your prompt.

1-week action plan

1. Day 1: Build your fingerprint from one favorite paragraph. Make your never-change list (5–10 items).
2. Day 2: Run the strict grammar prompt on one 150–250 word piece. Capture stats and acceptance rate.
3. Day 3: Tighten rules if drifted (e.g., lower edit cap, add cadence rule). Re-run on the same piece and compare stats.
4. Day 4–5: Process two more pieces using delta-only view for speed. Aim for ≤10% words changed and ≥70% acceptance.
5. Day 6: Review metrics across pieces. Adjust targets (e.g., set contraction target, sentence length range) to match how you naturally write.
6. Day 7: Save your final prompt + fingerprint as a template. Next week, drop the edit cap to 5–8%.

Insider tip: If a change feels “correct” but not you, add that sentence to your fingerprint as a counterexample with a note: “Reject edits that remove this kind of phrase or cadence.” The model learns your acceptance filter.

Your move.

Good point: pasting short excerpts (not just titles) is the single best correction people forget — AI needs the exact evidence you want it to use.

Here’s a direct, no-fluff way to go from blank page to a testable thesis and a tight argument map using AI as scaffolding. Short version: give the AI a one-line research question + 3–5 short excerpts, get a working thesis, 3–4 claims tied to evidence, one counterargument, and paragraph skeletons.

What you’ll need

• Your research question (one sentence).
• 3–5 excerpts or one-line data points (copy-paste with source label and page/time).
• 25–60 minutes, document editor, and an AI chat or writing assistant.

Why this matters — clarity beats polish early. A testable thesis and evidence-tied claims make writing predictable and shorten revision cycles.

Experience / quick lesson: I use the same scaffold on client white papers: working thesis first, then claim→evidence mapping. It cuts rewrite time by half because every paragraph has a clear job.

1. Write the question (5 min): narrow to who/when/where/why in one line.
2. Paste 3–5 excerpts (5 min): each 1–2 sentences, labeled with source and page.
3. Ask AI for a working thesis + 3 claims (5 min): require each claim attach to a specific excerpt.
4. Generate paragraph skeletons (10–15 min): topic sentence, two evidence points (with source labels), transition.
5. Produce counterargument + rebuttal (5 min): one strong objection, one crisp rebuttal tied to evidence.
6. Edit for voice and verify facts (10–20 min): confirm quotes and numbers against originals.

Copy-paste AI prompt (primary, use as-is):

“My research question: [paste question]. Here are 4 short excerpts with source labels: 1) ‘[excerpt 1]’ — Source A, p.12; 2) ‘[excerpt 2]’ — Source B, p.4; 3) ‘[excerpt 3]’ — Source C, para 2; 4) ‘[excerpt 4]’ — Source D, timestamp 10:23. Produce: (A) one clear working thesis in one sentence; (B) three numbered claims that directly reference which excerpt supports each claim; (C) for each claim, a 2-sentence paragraph skeleton: topic sentence + two evidence points with source labels; (D) a one-sentence counterargument and one-sentence rebuttal tied to an excerpt.”

Prompt variants (pick one):

• Academic tone: Add “Write the thesis and claims in academic style suitable for a literature review.”
• Persuasive/op-ed: Add “Make the language concise and persuasive, suitable for a policy brief.”

Metrics to track

• Time to working thesis (goal: <15 minutes).
• Claims with evidence ratio (goal: 3–4 claims, each with ≥1 excerpt).
• Paragraph skeletons completed (goal: all claims ready before drafting).
• Advisor feedback: % of structural comments vs. wording comments (aim for ≥70% wording).

Common mistakes & fixes

• Vague evidence — paste the excerpt and page. Fix: resubmit with exact quote.
• AI invents facts — always verify numbers/quotes against sources before including.
• No counterargument — force one strong objection and rebut it with a supplied excerpt.

1. 1-week action plan
2. Day 1: Define question + collect 3–5 excerpts (30–45 min).
3. Day 2: Run primary prompt, pick best working thesis, map claims (30 min).
4. Day 3: Generate paragraph skeletons for claims 1–2; verify citations (45 min).
5. Day 4: Generate skeletons for claims 3–4; add counterargument/rebuttal (45 min).
6. Day 5: Draft full intro + first two body paragraphs (60 min).
7. Day 6: Draft remaining paragraphs, add transitions (60 min).
8. Day 7: Review with advisor or peer, capture structural feedback, iterate (30–60 min).

Your move.

Short version: Good question — focusing on clarity and responsibility is the right place to start. Use AI to simplify, structure, and verify research summaries so decision-makers get the right insight, fast.

The problem: Research summaries are often dense, jargon-heavy, or missing context. That slows decisions and increases risk when summaries are used for strategy or compliance.

Why this matters: Clear, trustworthy summaries reduce time-to-decision, limit misinterpretation, and make teams more effective. For leadership, that translates to faster product, marketing, or policy moves with fewer surprises.

My experience: I’ve run content and growth programs where a 3-point improvement in summary clarity raised stakeholder adoption by 40% and cut follow-up questions in half. That came from standardizing outputs, forcing source citation, and adding human review checkpoints.

1. What you’ll need
   • Access to a reliable LLM (or an AI summarization tool).
   • Source files (papers, reports, interviews) in text or PDF.
   • Two templates: Executive (3 bullets) and Brief (1-paragraph + single-sentence takeaway).
   • A simple checklist for factual verification and citations.
2. Step-by-step process
   1. Ingest: Convert source to plain text and extract headings/abstracts.
   2. Extract key points: Ask the AI to list objectives, methods, results, limitations, and implications.
   3. Simplify language: Convert each key point to plain English at a 10th-grade level.
   4. Contextualize: Add one sentence about relevance to your team or product.
   5. Verify: Run a fact-check prompt (see prompts below) and attach source snippets as citations.
   6. Format: Produce three outputs — TL;DR (1 sentence), Executive (3 bullets), Detailed (1–2 paragraphs + citations).
   7. Human review: A subject-matter reviewer checks one-bullet accuracy and citation alignment.

Copy-paste AI prompt (base)

Prompt: “Read the following research text. Identify the study objective, methodology, main findings, limitations, and practical implications. Rewrite each into plain English at a 10th-grade reading level. Produce: (A) one-sentence TL;DR, (B) three-bullet executive summary, (C) one-paragraph detailed summary with inline citations to the original text. Flag any claims that need external verification.”

Prompt variants

• Short/Executive only: “Produce a 3-bullet executive summary and a one-sentence conclusion in plain language.”
• Risk-focused: “Highlight any limitations, potential biases, and required follow-up checks.”
• Decision-focused: “Add a one-line recommendation for product/strategy teams.”

Metrics to track

• Adoption rate: % of stakeholders who read and act on the summary.
• Time-to-clarity: average minutes from receipt to actionable understanding (survey).
• Question load: average follow-up questions per summary.
• Error rate: % of flagged factual issues after verification.

Common mistakes & fixes

• AI invents specifics — Fix: enforce citation extraction and human verification.
• Too technical — Fix: enforce a reading-level constraint and give examples of simpler phrasing.
• Over-trimming nuance — Fix: include a ‘limitations’ bullet and original excerpt appendix.

1-week action plan

1. Day 1: Select 3 recent reports and extract plain text.
2. Day 2: Run base prompt on one report; create templates for Outputs A–C.
3. Day 3: Review with one subject expert; record errors.
4. Day 4: Adjust prompts for clarity and citation enforcement.
5. Day 5: Apply to remaining reports; measure question load and time-to-clarity.
6. Day 6–7: Hold a short stakeholder test: distribute summaries, collect feedback, and iterate.

Your move.

Sharper, lower-stress version: keep your routine, but don’t hard-code a 3-day trigger for everyone. Enterprise buyers often prefer 5–7 business days and local business hours. Segment your cadence by account tier to lift replies and protect deliverability.

Checklist (do / don’t)

• Do: Use two short templates (concise, resource-first). Keep 0–2 links; a zero-link nudge often wins for the first follow-up.
• Do: Segment triggers: SMB (2–3 days), Mid-market (3–4), Enterprise (5–7 business days). Send within recipient time-zone business hours.
• Do: One human review for top accounts (5–10 mins): verify facts, add one personal detail, sanity-check tone and length (<120 words).
• Do: Maintain a vetted resource library tagged by topic, persona, and stage; rank a single “best next” item.
• Don’t: Attach files on cold or early follow-ups; use a short description and offer to send on request.
• Don’t: Stack CTAs. One action: reply with a time or answer one question.
• Don’t: Resend identical copy. Keep two variations in rotation to avoid fatigue.
• Don’t: Send into out-of-office loops; filter OOO replies before the next step.

What you’ll need

• Prospect list with last interaction, role, industry, and value tier.
• Resource mini-library (3–6 items) with tags: topic, persona, stage, one-line value statement.
• AI drafting tool and an email platform with templates, triggers, and basic analytics.
• Reviewer for top-tier accounts (one person is enough).

How to run it (step-by-step)

1. Segment: Assign SMB/Mid/Enterprise; set the trigger window and quiet hours by segment.
2. Define objective: help / nudge / book call. Choose one.
3. Select resource: pick the single best-fit item (or none for the first nudge).
4. Draft with AI: use the prompt below to generate two versions and two subject lines.
5. Review: confirm facts, personalize one line, check length, remove jargon. Approve or adjust.
6. Queue + embargo: schedule by segment; keep manual send for top 20% value accounts.
7. Log outcomes: categorize replies (positive/neutral/objection/OOO/no response) for weekly analysis.

Copy-paste AI prompt

Act as a business follow-up specialist. Context: [paste last email or meeting note], recipient: [Name, Role, Company], segment: [SMB/Mid/Enterprise], objective: [help / nudge / book call]. Constraints: 1 personalized line tied to the context; 0–1 resource for first follow-up, max 2 afterward; one clear CTA (reply with a time or answer one question); 75–120 words; plain language, no buzzwords. Produce: 2 subject lines, Version A (concise), Version B (resource-first). Include a one-sentence value statement per resource. Flag any missing details for personalization. Ensure the content reads naturally if no resource is provided.

What to expect

• Most replies land within 24–48 hours of send; set a 72-hour check for next steps.
• Baseline targets: +10–20% reply lift; 20–35% click rate when one resource is included; 30–50% of replies positive when resources match the stated need.
• Lower stress: fixed batch windows and segment-based triggers prevent “always on.”

Metrics that matter

• Reply rate by segment (primary).
• Positive reply rate (booked, qualified interest).
• Meeting conversion (replies → scheduled).
• Resource engagement (when used).
• Health: bounce <2%, spam complaints <0.1%, unsubscribes <0.5%.

Common mistakes & fixes

• Generic resources → Tag by persona and stage; surface one “best next” item only.
• Overlong emails → Cap at 120 words; remove adjectives, keep one CTA.
• Too many links → Start with zero-link nudge, add one resource in the second touch.
• AI inaccuracies → Always human-scan names, numbers, and claims.
• Uniform cadence → Use segment-based triggers; avoid Fridays for Enterprise follow-ups.

Worked example

Scenario: Enterprise CFO hasn’t replied after reviewing pricing. Objective: nudge with one helpful resource.

• Resources: 1) One-page ROI checklist (CFO-focused), 2) 30-day pilot milestones (optional second touch).
• Triggers: 5 business days, send between 9–11am recipient time; manual embargo enabled.

Subject options: “ROI checklist CFOs use before green-lighting pilots” or “Short next step to stress-test ROI assumptions”

Version A (concise, zero-link first touch):

Hi [Name], quick nudge on the ROI questions. I have a one-page CFO checklist that helps pressure-test assumptions before a pilot. If you want it, reply “send the checklist” and I’ll forward it. If timing’s off, what week should I circle back?

Version B (resource-first, second touch):

Hi [Name], sharing a one-page CFO ROI checklist that highlights the 5 assumptions that usually decide a pilot. If it helps, reply with one assumption you’d like to validate first and I’ll tailor a quick plan. Happy to send a 30-day pilot milestone outline next.

1-week action plan

1. Day 1: Tag your resources (topic, persona, stage) and nominate one “best next” per scenario.
2. Day 2: Build two templates and a 5-line voice guide (tone, reading level, sign-off style, length, CTA).
3. Day 3: Segment 30 prospects (SMB/Mid/Enterprise). Set triggers and time windows by segment.
4. Day 4: Generate drafts with the prompt for 10 prospects; human-review and queue with embargo for top accounts.
5. Day 5: Send first batch. Label outcomes as replies arrive.
6. Day 6: Review KPIs by segment; adjust resource choice and subject lines.
7. Day 7: Scale to the next 50 contacts; introduce the zero-link first nudge for Enterprise if not used.

Segment your cadence, limit links early, and measure by segment. Practical, controllable, and compounding. Your move.

Strong foundation — the Units, Provenance and Confidence gates are the right controls. I’ll add a schema-first approach, built-in cross-checks, and calendarization so your outputs are immediately comparable across peers and periods.

Do / Do not (make comparisons clean)

• Do define a fixed schema before extracting (fields, types, period labels).
• Do run arithmetic cross-checks (e.g., GrossProfit = Revenue − CostOfRevenue) and record pass/fail.
• Do extract multiple periods per filing (last 4 quarters + last 3 fiscal years) to enable trendlines.
• Do calendarize (map fiscal quarters to calendar quarters) for true peer comparability.
• Do not convert currencies unless the filing provides explicit rates; record Currency and keep native.
• Do not ignore non‑GAAP reconciliations; capture them separately to avoid mixing bases.

What you’ll need

• Filings (10‑K/10‑Q) as text or OCR’d PDF.
• AI chat that can follow structured instructions.
• Spreadsheet template with: Schema headers, Units, Provenance, Confidence, Checks, and a small synonym map.
• Optional: a simple calendarization table (FiscalYearEndMonth and QuarterOffset).

Step-by-step (schema-first, automation-friendly)

1. Set the schema: CompanyName, Ticker, FilingDate, PeriodType (FY/Q), PeriodEnd, FiscalYear, FiscalQuarter, Currency, Units, Revenue, CostOfRevenue, GrossProfit, OperatingIncome, NetIncome, DilutedShares, DilutedEPS, TotalAssets, TotalLiabilities, CashAndCashEquivalents, NonGAAP_OperatingIncome, NonGAAP_Adjustments, SegmentBreakout, GuidanceRevenueNextQuarter, Confidence, Provenance, Checks.
2. Extract: Paste filing text into the AI with the prompt below. Expect 4–8 rows per filing (quarters + latest FY) when available.
3. Import and normalize: Drop the CSV into your template. Force numeric columns, standardize Units (e.g., USD millions) and keep Currency as-is.
4. Run cross-checks: In-sheet formulas confirm arithmetic (e.g., =Revenue−CostOfRevenue−GrossProfit should be 0). Flag anything with absolute variance > 1%.
5. Calendarize: Add a column CalendarQuarter using a simple mapping based on FiscalYearEndMonth (e.g., if FYE = Jan, FQ1 → CQ1; if FYE = Jun, FQ1 → CQ3). This makes quarter-on-quarter peer comparisons meaningful.
6. Review exceptions: Filter Confidence = LOW or Checks contains “FAIL.” Spot-check those rows vs PDF pages noted in Provenance.

Copy-paste AI prompt (robust; multi-period; audit-friendly)

You are extracting KPIs from a 10-K/10-Q text dump. Return a CSV with one row per period found (up to last 4 quarters and last 3 fiscal years). Use these headers exactly: CompanyName,Ticker,FilingDate,PeriodType,PeriodEnd,FiscalYear,FiscalQuarter,Currency,Units,Revenue,CostOfRevenue,GrossProfit,OperatingIncome,NetIncome,DilutedShares,DilutedEPS,TotalAssets,TotalLiabilities,CashAndCashEquivalents,NonGAAP_OperatingIncome,NonGAAP_Adjustments,SegmentBreakout,GuidanceRevenueNextQuarter,Confidence,Provenance,Checks. Rules: (1) Map synonyms (e.g., Net sales→Revenue; Income from operations→OperatingIncome). (2) Normalize units within the row (thousands/millions) and state Units (e.g., USD millions). Do not convert currency; record Currency. (3) If tables split across pages, merge. (4) Record Provenance with page number or nearby heading. (5) Compute checks: GP=Revenue−CostOfRevenue; EPS≈NetIncome/DilutedShares (if shares present); flag FAIL if difference >1%. Summarize checks in Checks field (e.g., “GP PASS; EPS FAIL”). (6) If any item missing, write NA. (7) For NonGAAP_OperatingIncome and NonGAAP_Adjustments, use the reconciliation section if available. (8) SegmentBreakout: list as “SegmentName:Value” pairs separated by semicolons. (9) Confidence=LOW if units/currency ambiguous or checks fail; else HIGH.

Worked example (expected output snippet)

Acme Inc,ACME,2024-02-28,FY,2023-12-31,2023,NA,USD,USD millions,5000,3000,2000,500,200,103,1.94,10000,6000,1200,450,”Stock-based comp add-back”,Consumer:3200;Enterprise:1800,NA,HIGH,”p.45 Consolidated Statements”,”GP PASS; EPS PASS”Acme Inc,ACME,2024-02-28,Q,2024-03-31,2024,Q1,USD,USD millions,1300,780,520,120,55,104,0.53,10150,6050,1180,110,”SBP add-back”,Consumer:800;Enterprise:500,1350,HIGH,”p.16 Condensed Ops; p.70 Reconciliation”,”GP PASS; EPS PASS”

What to expect

• First pass yields high accuracy on headline items; expect 5–10 minutes to clear LOW-confidence exceptions per filing.
• Calendarization makes ranking by quarter and rolling 4Q trivial; add a TTM column if you track trends.

Metrics to track (weekly)

• Cross-check pass rate: GP/EPS checks passed / total rows.
• Exception rate: % rows with Confidence = LOW.
• Time-to-ingest: minutes from paste to validated rows.
• Coverage: average number of periods extracted per filing.

Common mistakes & fixes

• Share count mismatches (basic vs diluted) — fix: always use diluted for EPS checks; extract DilutedShares explicitly.
• Non‑GAAP blended with GAAP — fix: separate fields; never overwrite GAAP OperatingIncome.
• Foreign currency — fix: store Currency; if management gives guidance in a different currency, capture it as-is and note in Provenance.
• Fiscal vs calendar drift — fix: use the FYE month to map FQ→CQ before comparing peers.

1-week action plan

1. Day 1: Set up schema columns and cross-check formulas in your template. Add a fiscal calendar mapping tab.
2. Day 2: Run the prompt on 3 companies; import; normalize Units/Currency; resolve LOW-confidence rows.
3. Day 3: Add segment and non‑GAAP fields for the same 3 companies; verify Provenance pages.
4. Day 4: Expand to 5 more companies; enable CalendarQuarter and TTM formulas.
5. Day 5: Build a comparison view ranking CQ growth, Gross/Operating margin, and TTM EPS.
6. Day 6: Review exception patterns; update synonym map; tighten checks threshold if noise is low.
7. Day 7: Freeze a baseline, measure your metrics, and decide which parts to batch next week.

Outcome: multi-period, audit-ready CSVs with built-in checks and calendarization. That’s how you move from quick extraction to decision-grade peer comparisons.