aaron

Quick win: Good point — keeping onboarding simple is the single best lever to reduce confusion and speed revenue. Here’s a direct, practical way to use AI to build repeatable client onboarding documents.

The problem: onboarding docs are inconsistent, take too long to produce, and leave clients unsure of next steps.

Why this matters: clean onboarding reduces client questions, accelerates project starts, and improves retention. That directly affects cashflow and capacity.

What I’ve learned: structure beats creativity for onboarding. A predictable template + AI for drafting saves hours and makes reviews trivial.

1. What you’ll need
   • One example client onboarding document (even a rough Word/Google doc).
   • A list of standard intake fields (name, scope, timelines, deliverables, billing).
   • An AI text tool (ChatGPT or similar) and a place to store templates (Google Docs, Notion, or your CRM).
2. Step-by-step to a working system
   1. Define the 6 core sections: Welcome, Scope, Timeline, Deliverables, Client responsibilities, Next steps + signature.
   2. Create a short template with placeholders: {ClientName}, {StartDate}, {Deliverable1}.
   3. Use AI to draft the content for each placeholder from a short intake form.
   4. Review & standardize tone (one reviewer, 10–15 minutes).
   5. Save the final document as a template and automate population (manual copy-paste to start; add automation later).

AI prompt you can copy-paste (paste into your AI tool and replace bracketed items):

“Create a concise client onboarding document for [ServiceName] for a small business. Use a friendly professional tone. Sections: Welcome (1 short paragraph), Scope (bulleted list of deliverables based on: [Deliverable1]; [Deliverable2]), Timeline (start date: [StartDate], milestones: [Milestone1]), Client responsibilities (3 clear bullets), Billing & payment (terms: [Terms]), Next steps (3 actions with due dates). Use placeholders where applicable.”

What to expect: first drafts in seconds; final document ready after a 10–15 minute human review.

Metrics to track

• Time from contract signature to project start (goal: reduce by 30% within month 1).
• Number of clarification emails after onboarding (goal: reduce).
• Percent of clients completing onboarding checklist within 7 days.

Common mistakes & fixes

• Too much text — fix: limit each section to 1–3 bullets.
• Unclear responsibilities — fix: use direct language and deadlines.
• No review step — fix: require a single 10-minute human approval before sending.

One-week action plan

1. Day 1: Pick one recent onboarding doc and list standard fields.
2. Day 2: Build the template with placeholders.
3. Day 3: Run the AI prompt and create 3 sample drafts for different client types.
4. Day 4: Review and finalize one template.
5. Day 5: Start using for all new clients; track time-to-start and questions.
6. Day 6–7: Tweak language based on client feedback and lock the template.

Your move.

Quick win: Good question — starting with Notion keeps your team in one place and removes a lot of friction.

The problem: Editorial work is slow because briefs, drafts, edits and publishing notes live in different places and rely on manual handoffs.

Why it matters: Faster, repeatable editorial workflows reduce publish time, increase output quality and free senior editors for strategy instead of admin.

Lesson from practice: I’ve seen teams cut time-to-publish by 40–60% by standardising a Notion database, adding one AI step for first drafts, and automating the simple handoffs.

1. What you’ll need
   • Notion account (workspace + database)
   • Notion AI (recommended) or an OpenAI key + Zapier/Make for automation
   • Simple brief template and 30 minutes of setup time
2. Setup — build the database (20–40 minutes)
   1. Create a Notion database called Editorial Calendar.
   2. Add properties: Title, Status (Idea, Assigned, AI Draft, Editing, Ready), Assignee, Publish Date, Word Count target, AI Brief, AI Draft, Editor Notes, Link.
   3. Create two page templates: Brief template and Publish checklist.
3. Integrate AI — two simple paths
   1. Notion AI: Open the page, paste your brief, use Notion AI to generate a draft into the AI Draft property.
   2. Zapier/Make + OpenAI: Create a zap triggered by Status = “AI Draft”. The zap sends the AI Brief to OpenAI and writes back the output to the AI Draft field.
4. Daily use
   1. Owner fills Brief template (3–5 bullet points: angle, audience, keywords, tone, CTA).
   2. Mark Status = AI Draft. AI creates first draft. Editor rounds to polish, sets Status to Ready or Editing.

Copy-paste AI prompt (use in Notion AI or your Zap):

Write a 600-word article for [audience: small-business owners over 40] about [topic]. Start with a clear benefit statement, use plain language, include three practical steps, a brief example, and a single call-to-action to book a free consult. Tone: confident, warm, non-technical. Target keywords: [keyword list].

Metrics to track

• Drafts generated per week
• Average time: brief→publish
• Average revision rounds per article
• Publish rate (pieces published ÷ briefs started)
• Engagement: opens, reads, or pageviews after 30 days

Common mistakes & fixes

• Vague briefs → Fix: enforce a 5-bullet brief template.
• Over-trusting raw AI output → Fix: require one editor pass before scheduling.
• Too many status columns → Fix: keep 5 core statuses only.

1-week action plan

1. Day 1: Create database + templates (30–40m).
2. Day 2: Add 5 example briefs and test Notion AI or Zap.
3. Day 3: Run one draft through end-to-end, note time saved and edits needed.
4. Day 4–5: Train your team on the brief template (15m session).
5. Day 6–7: Publish 1–2 pieces, record metrics and adjust briefs/automation.

Your move.

You’re asking the right question: not “shorter,” but “shorter without losing the caveats, tensions, and edge-cases.” That’s where most summaries fail.

The problem: generic summarization flattens hedging language, erases minority views, and blends evidence with opinion. Decisions made on that kind of output look confident but fragile.

Why it matters: nuance is where risk and opportunity hide—assumptions, uncertainty levels, and conflicting evidence. Preserve those, and you’ll make faster, safer calls.

What works in practice: a dual-pass approach (extract first, then compress) with guardrails—citations, confidence labels, counterpoints, and a coverage check. It’s fast, auditable, and repeatable.

• What you’ll need: a modern AI assistant, a way to get your report into clean text (with paragraph numbers), and 10 minutes for a verification pass.
• What to expect: a tight executive summary plus a “nuance map” capturing caveats, uncertainty, and opposing views; 60–80% time saved versus manual notes; a 5–10 minute review step remains essential.

Start here: copy-paste prompt

Use this on any report (paste the report below the prompt). Expect an executive summary and a documented nuance layer you can trust.

Prompt:

You are my Report Nuance Keeper. Your job is to produce a decision-ready summary that preserves nuance, not just brevity.

Process, in order:

1) Extractive pass — pull key sentences verbatim with paragraph numbers; 2) Abstractive pass — compress into plain English; 3) Nuance map — list assumptions, caveats, uncertainty levels, and minority/contradictory views; 4) Coverage check — confirm which sections were summarized.

Requirements:

– Keep hedging words (e.g., “may,” “likely,” “preliminary”).

– Label each claim with a confidence: High / Medium / Low.

– Provide 5–10 pull quotes with [para#] citations.

– Separate evidence from interpretation.

– Include counterpoints and what the report did NOT cover but should have (gaps).

– If a detail isn’t in the text, say “insufficient evidence.” Do not invent.

Output format:

1) Executive Summary (5–7 bullets). 2) Decision-Relevant Signals (What matters, Why, So-what). 3) Caveats & Uncertainties (with confidence labels). 4) Minority/Conflicting Views. 5) Pull Quotes with [para#] citations. 6) What’s Missing/Gaps. 7) If-Then-Else Implications (2–4). 8) Section Coverage Map (list sections with Covered/Partial/Missed).

Now I will paste the report content with paragraph numbers. Work step-by-step and keep citations in every section where applicable.

Five-step implementation

1. Pre-process the report
   • Add paragraph numbers (e.g., [1], [2], …). If it’s a PDF, export to text and number each paragraph.
   • Mark major sections (Executive Summary, Methods, Results, Limitations, etc.).
   • Optional: highlight names, dates, and metrics you care about.
2. Run the dual-pass prompt
   • Paste the prompt and the numbered report.
   • Scan the output for structure and citations. If missing, reply: “Re-run with citations and coverage map intact.”
3. Stakeholder tailoring
   • Follow-up prompt: “Tailor the Executive Summary for CFO, Operations, and Legal. Keep contradictions and caveats visible for each.”
   • Expect three short, role-specific versions that emphasize cost, execution risk, or exposure.
4. Verification pass
   • Ask: “List each claim with its [para#] source or mark ‘insufficient evidence’.”
   • Spot-check 5 claims against the report. Correct any drift and re-run if needed.
5. Standardize
   • Save the prompt as a template. Create a 1-page SOP: inputs, steps, checks, and sign-off.
   • Batch process your top three recurring report types next week.

Metrics that keep this honest

• Time to first draft summary (minutes).
• Reviewer time (minutes) to accept or edit.
• Correction rate: number of edits per summary.
• Lost-nuance incidents: cases where a caveat/conflict was missed.
• Decision clarity score: 1–5 rating from the decision-maker.
• Coverage ratio: sections marked Covered vs Partial/Missed.

Common mistakes and fixes

• Flattened nuance: The model drops caveats. Fix: explicitly require hedges and a Caveats & Uncertainties section with confidence labels.
• Hallucinated facts: Fix: mandate [para#] citations and the “insufficient evidence” rule. Reject any claim without a source.
• Overlong outputs: Fix: set limits (e.g., “Executive Summary max 120 words; each section max 7 bullets”).
• Too generic: Fix: include 1–2 exemplar summaries from past work in your prompt as style guides.
• One-size-fits-all: Fix: run stakeholder-tailored summaries; nuance varies by role.

One-week rollout

1. Day 1: Pick three representative reports. Number paragraphs. Define your metrics and acceptable thresholds.
2. Day 2: Run the dual-pass prompt on Report 1. Note gaps. Tweak the prompt (especially the coverage map and confidence labels).
3. Day 3: Process Report 2 with the improved prompt. Add stakeholder-tailored outputs. Start tracking metrics.
4. Day 4: Build the verification step (claim-to-para map). Create a 1-page SOP.
5. Day 5: Pilot with two colleagues. Collect decision clarity scores and correction counts.
6. Day 6: Iterate. Add 2 exemplar outputs to the prompt. Lock your template.
7. Day 7: Scale to your next batch. Schedule a monthly review of metrics and missed-nuance incidents.

Insider trick: Ask for a “Section Coverage Map” and a “Nuance Map” every time. The first proves nothing was skipped; the second captures assumptions, uncertainty, and minority views in one view—this is where most AI summaries fail, and it’s where your decisions get safer.

Your move.

Build quizzes that prove mastery of your objectives—nothing else.

Quick correction: Don’t ask AI for “a quiz on topic X.” Start with measurable learning objectives and the evidence you want to see. AI can’t infer your standards; you have to supply them.

Why this matters: Aligned quizzes raise signal quality, speed up course iteration, and improve completion rates. Misaligned quizzes waste learner time and give you unreliable data.

What you’ll need: an AI assistant (any leading LLM), your learning objectives (with Bloom’s level), a short content scope, common learner misconceptions, and a spreadsheet or form tool to import items.

What to expect: A blueprint, a small but high-quality item bank, rationale-rich feedback, and data you can track (difficulty, discrimination, coverage). Expect to iterate once after a small pilot.

My lesson from many builds: Blueprint first, then generate within constraints, then pilot with a small group and adjust using simple item stats. That sequence avoids 80% of rework.

Step-by-step

1. Define outcomes: Convert each objective into a measurable statement with Bloom’s level (e.g., “Apply: Calculate net present value for a 5-year project”). Add 2–3 common misconceptions per objective.
2. Create a blueprint: For each objective, set weight, item types allowed (MCQ, scenario, short answer), and target difficulty mix (e.g., 30% easy, 50% medium, 20% hard).
3. Generate items with AI: Use the prompt below to produce 3–5 items per objective with rationales and distractors tied to misconceptions.
4. Quality screen: Manually check for clarity, single-best correct answer, and realistic distractors. Remove anything that can be answered by clue-hunting rather than understanding.
5. Pilot: Give 10–20 learners the quiz. Capture item-level responses and completion time.
6. Analyze and revise: Flag items that are too easy or too hard and those that fail to discriminate (everyone gets them right or wrong). Refine stems or distractors.
7. Deploy: Import items into your LMS or form tool with tags (objective, Bloom level, difficulty). Set up automatic feedback using the AI-generated rationales.

Copy-paste AI prompt (master)

Role: You are an expert assessment designer. Create an objective-aligned quiz that measures the specific learning outcomes below.

Inputs:

– Learner profile: [e.g., mid-career managers, non-technical]- Content scope: [what’s in-bounds / out-of-bounds]- Learning objectives (ID, verb, Bloom level): [e.g., OBJ1 Apply: Calculate NPV; OBJ2 Analyze: Compare financing options]- Common misconceptions per objective: [list 2–3 each]- Item types allowed: [e.g., MCQ 4 options, scenario MCQ, short answer]- Difficulty mix targets: [Easy 30%, Medium 50%, Hard 20%]- Tone and context: [professional, real-world business scenarios]

Output format for each item (strict):ItemID | ObjectiveID | BloomLevel | Difficulty(E/M/H) | ItemType | Stem | Options(A–D) or ExpectedAnswer | CorrectAnswer | Rationale(why correct) | DistractorLogic(why each wrong) | Tags(objective, topic, skill) | EstimatedTime(sec)

Tasks:

1) Produce a 20-item bank aligned to the blueprint and difficulty targets.2) Ensure each distractor is plausible and tied to a listed misconception.3) Include short, learner-friendly rationales.4) Vary scenarios and numbers to prevent cueing.5) End with a coverage summary: items per objective, Bloom distribution, difficulty distribution.

Variants you can run

• Scenario-heavy version: Emphasize real-world vignettes with only one defensible best answer. Keep stems under 90 words.
• Short-answer + rubric: Ask for 3–5 model short answers and a 3-point rubric (Exceeds/Meets/Below) with criteria tied to the objective.
• Misconception mining: Feed anonymized learner emails or FAQs and ask the AI to extract common errors, then regenerate distractors using those.
• Adaptive sets: Request three parallel forms (A/B/C) with the same blueprint but varied numbers and contexts for retesting.

One-click prompt: scenario items only

Create 10 scenario-based MCQs aligned to these objectives: [paste objectives with Bloom level]. Use realistic business contexts. For each item, output: ItemID | ObjectiveID | Bloom | Difficulty | Stem | A–D | Correct | Rationale | DistractorLogic | Tags | Time. Make one and only one best answer. Tie every distractor to a listed misconception. Target difficulty mix: [X%/Y%/Z%]. End with a coverage summary.

Insider trick: Ask the AI to generate items in two passes—first generate, then “adversarially critique” each item for ambiguity, cueing, and alignment, and revise. This boosts item quality without more of your time.

Metrics to track (KPIs)

• Objective coverage: ≥95% of objectives represented; weight matches blueprint.
• Item difficulty (p-value): Easy ~0.75, Medium ~0.5, Hard ~0.3 after pilot.
• Discrimination: Items where high scorers outperform low scorers by ≥30% keep; otherwise revise.
• Time on task: Median time per item within ±20% of estimate.
• Reliability: For 15+ items, aim KR-20/Cronbach’s alpha ~0.7+ (approximate using your LMS analytics).
• Rationale usefulness: ≥80% of learners rate feedback as clear in a one-question pulse.

Common mistakes and quick fixes

• Mistake: Topic-based items without measurable verbs. Fix: Rewrite objectives with Bloom levels and evidence of mastery.
• Mistake: Vague stems or more than one defensible answer. Fix: Use the adversarial critique pass and force single-best answers.
• Mistake: Weak distractors. Fix: Base them on real misconceptions; require a “why wrong” note for each option.
• Mistake: Single difficulty level. Fix: Set and check the Easy/Medium/Hard mix explicitly.
• Mistake: No pilot data. Fix: Run a 10–20 person pilot; keep only items with acceptable difficulty and discrimination.

1-week action plan

1. Day 1: List objectives with Bloom levels and 2–3 misconceptions each.
2. Day 2: Build the blueprint (weights, item types, difficulty mix).
3. Day 3: Run the master prompt. Generate 30–40 items. Auto-critique and revise.
4. Day 4: Human review. Trim to the best 20–25 items. Load into your LMS or form.
5. Day 5: Pilot with 10–20 learners. Capture item-level responses and time.
6. Day 6: Analyze KPIs. Revise or replace low-performing items. Generate Form B for re-tests.
7. Day 7: Finalize, schedule, and turn on automatic feedback from rationales. Set a 2-week review checkpoint.

Your move.

Smart question—focusing on clearer and shorter messages is the fastest, lowest-risk way to get measurable gains in response rates and decision speed.

Here’s the move: use AI as your on-demand editor to apply BLUF (Bottom Line Up Front), strip fluff, and keep your tone. Done right, you’ll cut word count by 40–70% and get quicker, more decisive replies.

The problem: Most messages bury the ask, hedge with soft language, and force the reader to reconstruct context. That drags decisions, creates follow-up loops, and kills momentum.

Why it matters: Clear, short messages drive higher reply rates, faster approvals, fewer clarifications, and better client confidence. That’s pipeline velocity.

Lesson: A simple system—BLUF + guardrails + constraints—beats “rewrite this” every time.

What you’ll need:

• An AI assistant (ChatGPT/Copilot/Gmail/Outlook add-ins).
• A reusable rewrite prompt (below).
• Your tone guide: formal/informal, direct/warm, industry terms to keep/avoid.
• Privacy habit: remove names, prices, and confidential details before pasting.

How to do it (six steps):

1. Define the outcome. Who’s the reader, what’s the single decision/next step, and what channel (email, Slack, SMS)? Expect: 1–2 minutes.
2. Map the intent. Paste the original and ask AI to list the objective, key points, and missing context. You’ll spot gaps fast.
3. Rewrite with BLUF. Force a one-sentence top-line, 3 bullets max, a clear CTA with deadline, and a length cap.
4. Compress and de-jargon. Set a reading grade target (6–8), cut filler, convert walls of text to bullets.
5. Guardrails. Preserve facts, numbers, commitments, and any legal phrasing. Add your human line (“Thanks for the quick look”).
6. Template it. Save the prompt as a shortcut (text expander or email template) so the whole team can be consistent.

Copy-paste prompt (keeps tone, gets results):

• Prompt: “You are my executive communications editor. Rewrite the message below to be clear, concise, and action-oriented for a busy [role: CFO/Client/Partner]. Use BLUF (first sentence states the ask), keep my warm-but-direct tone, preserve all facts and numbers, and remove filler. Constraints: max 120 words, 3 short bullets, one clear CTA with a specific deadline and response method. Target reading level: grade 7. Output two options: (A) formal concise, (B) friendly concise. Then list any missing info as questions. Here is the message: [PASTE TEXT]”

Add-on prompts (use as needed):

• “Compress to 80 words without losing the ask or numbers. Keep my tone.”
• “Convert to SMS-length, same CTA, no jargon.”
• “Tone shift: from tentative to confident, respectful. Keep facts intact.”
• “Create a reusable template from this final version with [placeholders] for names, dates, and amounts.”

What to expect:

• Drafts in 10–20 seconds.
• 40–70% word-count reduction while keeping meaning.
• Clear CTA that cuts back-and-forth emails by 20–40% within two weeks.

Metrics to track (weekly dashboard):

• Average word count per message (target: down 40%+).
• Reading grade level (target: 6–8).
• Response time to key emails (target: -30%).
• Reply rate to CTAs (target: +15–25%).
• Clarification replies per thread (target: -30–50%).
• Time spent drafting per message (target: -50%).

Common mistakes and fixes:

• Sounds robotic: Tell the AI “keep my voice; vary sentence length; include one natural courtesy line.”
• Ask is buried: Force a one-line BLUF and a single explicit CTA with a deadline.
• Loss of nuance: Add “preserve qualifiers and commitments; do not change numbers/dates.”
• Wrong tone for senior execs: Use “formal concise (no emojis), confident, assume expertise.”
• Privacy leakage: Redact names, deal terms, and attachments; summarize sensitive parts instead of pasting.

One-week rollout:

1. Day 1: Baseline. Export last 20 sent emails. Measure word count, response time, and clarification replies.
2. Day 2: Install your AI tool of choice. Save the core prompt as a shortcut. Define your tone (3 do’s, 3 don’ts).
3. Day 3: Pilot on 3 message types: stakeholder update, client ask, internal request. A/B send: AI-rewritten vs. original to a small, low-risk group.
4. Day 4: Review results. Keep the version with highest reply rate and shortest response time.
5. Day 5: Create two templates (formal, friendly). Add placeholders. Share with your team.
6. Day 6: Automate. Add a text-expander snippet or email template button. Create a “TL;DR + CTA” macro.
7. Day 7: Roll out. Train the team in 30 minutes: BLUF, constraints, guardrails. Set weekly metrics and ownership.

Bottom line: yes—AI can rewrite your messages so they’re shorter, clearer, and more effective. Start with the prompt above, enforce BLUF, and track the numbers. You’ll feel the speed by next week.

Your move.

Quick take: Good point — focusing on measurable results and KPIs is exactly where this starts. AI can reliably enforce a content style guide at scale, but only if you design the process around outcomes, not tools.

The problem: Large teams drift — voice, punctuation, legal phrasing and formatting vary across writers. That costs time, confuses customers, and weakens brand trust.

Why it matters: Consistency reduces edits, speeds publishing, and makes your content convert better. Small improvements compound across hundreds of assets.

What I’ve learned: AI is best used as an automated editor and audit engine — not a magic fix. You get predictable results when you pair a clear, machine-readable style guide with an integration that fits your workflow.

1. What you’ll need
   • One canonical style guide (single doc with examples).
   • An AI assistant (off-the-shelf model or editor in your CMS/Slack).
   • A checklist for pre-publish checks and a small sample of labeled examples.
2. Step-by-step setup
   1. Consolidate: Create a 2–3 page style summary (voice, do/don’t, legal phrasing, headline rules, formatting examples).
   2. Train: Feed 20–50 example documents (good and bad) to the AI or set as reference prompts.
   3. Integrate: Add an AI “style check” step in your workflow—either a CMS pre-publish check or a Slack command for drafts.
   4. Automate: Configure the AI to return three things: violations, corrected text, and a one-line rationale.
   5. Audit: Weekly sample audit (10–20 pieces) to finetune the guide and examples.

Copy-paste AI prompt (use as-is)

Act as our content style enforcer. Given the style guide below and the content below it, list all style violations, return a corrected version that follows the guide, and give a one-sentence reason for each change. Output: compliant_text, violations (short list). Style Guide: {{PASTE_STYLE_GUIDE_HERE}} Content: {{PASTE_CONTENT_HERE}}

Metrics to track

• Initial compliance rate (%) — percent of pieces passing AI check without edits.
• Edit time reduction — average minutes saved per asset.
• Post-publish correction rate — percentage of published pieces needing fixes.
• Content velocity — assets published per week.

Common mistakes & fixes

• Vague guide — fix: add concrete examples and absolute rules (e.g., never use passive voice in headlines).
• Over-reliance on AI — fix: keep human review for legal/sensitive content.
• Poor training samples — fix: curate 20 high-quality examples to teach the model.

7-day action plan

1. Day 1: Draft 2-page style summary.
2. Day 2: Collect 20 example docs (10 good, 10 bad).
3. Day 3: Configure AI prompt and test on 5 articles.
4. Day 4: Set up the CMS or Slack pre-publish check.
5. Day 5: Run a 20-item audit and record compliance rate.
6. Day 6: Tweak style guide based on audit feedback.
7. Day 7: Train editors, set KPI targets for month 1.

Expect initial compliance in the 60–80% range; aim for 90% within 30 days with iteration.

— Aaron Agius. Your move.

Quick win: Framing best- and worst-case revenue scenarios is exactly the right place to apply AI — it turns messy assumptions into clear, testable projections you can act on.

The problem: Most small businesses either guess revenue or spend days building spreadsheets that don’t reflect uncertainty. That leads to bad decisions on hiring, marketing, and cash.

Why it matters: Know the range of possible outcomes, the drivers that move the needle, and the probabilities. That gives you predictable runway, prioritised investments, and clear KPIs tied to cash.

What I’ve learned: Simple, repeatable scenario models beat complex, fragile forecasts. Use AI to structure assumptions, generate scenarios, and translate them into monthly revenue curves you can track.

What you’ll need:

• Last 12 months of revenue by month (CSV or spreadsheet)
• Top-line assumptions: conversion rate, traffic, average order value (AOV), churn, CAC
• A spreadsheet (Excel or Google Sheets)
• Access to an AI assistant (ChatGPT or similar)

Step-by-step (do this):

1. Collect: export monthly revenue and units for last 12 months.
2. Define drivers: list 5 inputs that most affect revenue (traffic, conv%, AOV, churn, pricing).
3. Use the AI prompt below to generate three scenarios (worst/base/best) with monthly revenue for 12–36 months, and include probability weights and sensitivity notes.
4. Paste outputs into your spreadsheet. Add formulas to calculate cash burn, cumulative revenue, and runway.
5. Validate: compare month 1–3 with actuals; ask AI to revise assumptions where variance >10%.

Copy-paste AI prompt (primary):

“I will give you monthly revenue for the last 12 months and five key inputs: traffic, conversion rate, average order value, churn rate, customer acquisition cost. Create three scenarios (worst, base, best) for the next 24 months. For each scenario provide: monthly revenue, monthly customers (or transactions), assumptions for each input, probability weight (sum to 100%), and a 3-bullet explanation of the main drivers. Format as a CSV table with columns: Month, Scenario, Revenue, Customers, Traffic, ConversionRate, AOV, Churn, CAC, Probability. Use realistic ranges and note where assumptions differ between scenarios.”

Prompt variants:

• Conservative: Ask AI to be conservative on growth and assign 60% probability to base.
• Monte Carlo-lite: Ask AI to simulate 500 runs using random draws from input ranges and return the median, 10th and 90th percentiles per month.

What to expect: A table you can paste into Sheets, plus a short narrative explaining drivers and recommended monitoring thresholds.

Metrics to track:

• Probability-weighted expected revenue (monthly)
• Worst-case revenue (10th percentile)
• Revenue variance between best/base/worst
• Customer Acquisition Cost and Payback Period
• Cash runway under worst-case

Common mistakes & fixes:

• Mistake: Garbage inputs. Fix: Use the last 12 months of real data as baseline.
• Mistake: Ignoring correlations (e.g., traffic and conversion tied). Fix: Ask AI to note correlations and adjust ranges.
• Mistake: Overconfidence in a single scenario. Fix: Use probability weights and track percentiles.

1-week action plan:

1. Day 1: Export 12 months of revenue and list 5 drivers.
2. Day 2: Run the primary AI prompt, paste output into Sheets.
3. Day 3: Validate month 1–3 vs actuals; adjust inputs.
4. Day 4: Build simple dashboard: expected revenue, worst-case, runway.
5. Day 5: Identify 2 tactical moves (cut cost or boost traffic) for worst-case protection.
6. Day 6: Run a sensitivity check (change conv% by ±2%) and note impact.
7. Day 7: Present scenario summary and decision triggers to stakeholders.

Your move.