Jeff Bullas

Spot on: your one-week action plan and “rules + human checks” is exactly how small shops win fast. Let’s add two upgrades that boost accuracy without extra complexity: a simple seasonality blend and an exception-based review so you only fix what needs attention.

Quick promise: In one weekend you can set a baseline forecast, seasonality bump, practical min–max levels, and an exception list that tells you where to act every Monday.

What you’ll need

• 6–24 months of weekly sales by SKU (promos/closures tagged).
• Supplier lead times and purchase multiples (case packs).
• A spreadsheet (Excel/Google Sheets) or a simple forecasting app.
• 30 minutes each week for a quick review.

Step-by-step (one-weekend build)

1. Pick the right SKUs: Start with the top 20 by sales volume. Add more after 4 weeks.
2. Baseline forecast: Use the average of the last 8–12 weeks of sales for each SKU. That’s your steady demand.
3. Seasonality blend (insider trick): If your business has seasonal swings, blend in last year’s same period. Simple rule: Forecast = 70% of baseline + 30% of “same week last year.” If you don’t have last year, use the average of the same month last year.
4. Safety stock by stability:
   • Start with 1 week of baseline sales.
   • If weekly standard deviation ≥ 50% of average, use 1.5 weeks. If ≥ 75%, use 2 weeks.
5. Reorder point (ROP): ROP = (weekly forecast × lead time in weeks) + safety stock.
6. Set min–max: MIN = ROP. MAX = ROP + 2 weeks of demand (adjust to your cash comfort). Order Qty = MAX − (On Hand + On Order), rounded up to purchase multiple.
7. Lead-time reality check: If a supplier is often late, add a half-week buffer to lead time until they improve. Keep a simple “supplier reliability” note.
8. Exception-based review:
   • Flag if: On Hand + On Order − 2-week forecast < 0 (risk of stockout).
   • Flag if: Last week’s absolute % error > 30% (volatile SKU).
   • Flag if: Weeks on hand > 8 (potential overstock).

   Only review flagged SKUs weekly; the rest run on rules.

Worked example

• SKU: Coffee Beans
• Baseline (last 10 weeks): 20 units/week
• Same week last year: 24 units → Seasonality blend = (0.7 × 20) + (0.3 × 24) = 21.2
• Weekly SD: 8 (≥ 50% of 20) → Safety stock = 1.5 weeks = 1.5 × 21.2 ≈ 32
• Lead time: 2 weeks → ROP = (21.2 × 2) + 32 ≈ 74
• MIN = 74, MAX = 74 + (2 × 21.2) ≈ 116
• On hand: 60, On order: 0 → Order = 116 − 60 = 56 (round to nearest case size)

High-value add-ons (small effort, big payoff)

• Shadow orders for 2 weeks: Compare “what the rules say” vs “what you would have ordered.” Tweak safety stock once, then commit.
• Promo quarantine: Tag promo weeks and either exclude them from averages or cap them at 1.2× normal demand.
• Dead stock throttle: If weeks on hand > 8, cut the next order by 25% until you’re back to target.
• Supplier sync: Group orders by supplier and standardize order days (e.g., Tuesdays). Fewer, smarter orders.

Common mistakes & fast fixes

• Chasing every wiggle: Don’t rebuild rules after one weird week. Use rolling 8–12 week averages.
• Forgetting on-order units: Always subtract On Hand + On Order from MAX before buying.
• Blind to seasonality: Blend last year’s same period at 20–30% to smooth peaks.
• Too many SKUs: Limit to top 20 first. Expand when weekly review takes < 30 minutes.

Copy-paste AI prompt (turn your CSV into a forecast + order plan)

“Act as an inventory planner for a small retail shop. I will provide a CSV with columns: date (YYYY-MM-DD), sku, units_sold, on_hand, on_order, lead_time_weeks, purchase_multiple, promo_flag, closed_flag. For each SKU: 1) Build weekly demand using the last 8–12 weeks (exclude closed weeks). 2) Apply a seasonality blend: 70% of the recent average + 30% of the same period last year (if available). 3) Set safety_stock = 1 week of forecast; if weekly SD ≥ 50% of average, use 1.5 weeks; if ≥ 75%, use 2 weeks. 4) Compute reorder_point = forecast × lead_time_weeks + safety_stock. 5) Set MIN = reorder_point and MAX = MIN + 2 × forecast. 6) Recommend order_qty = MAX − (on_hand + on_order), rounded up to purchase_multiple (minimum 0). 7) Flag exceptions: potential_stockout_in_2_weeks, high_error_last_week (abs % error > 30%), overstock_gt_8_weeks. Return two CSVs: a) forecast.csv with sku, forecast_weekly, sd_weekly, safety_stock, reorder_point, MIN, MAX; b) order_plan.csv with sku, on_hand, on_order, order_qty, exception_flags, notes.”

14-day action plan

1. Days 1–2: Export top 20 SKUs (weekly). Note lead times, purchase multiples, and promo/closure flags.
2. Days 3–4: Build baseline + seasonality blend. Set safety stock by volatility. Compute ROP, MIN, MAX.
3. Day 5: Create the exception list (three flags above). Sanity-check against your gut.
4. Day 6–7: “Shadow order” week: compare rule-based orders vs usual practice. Adjust safety stock once.
5. Week 2: Place real orders using the rules. Do a 20-minute review on Monday: check exceptions, update lead times, note any events coming up.
6. Day 14: Review early results: stockouts, weeks on hand, and any flagged SKUs. Decide whether to add the next 10–20 SKUs.

What to expect

• Smoother ordering and fewer surprises within a few weeks.
• Clear visibility of which SKUs truly need your attention.
• A repeatable, calm Monday routine: update data, scan exceptions, place orders.

Reminder: Consistency beats complexity. Keep the rules simple, review weekly, and let AI handle the heavy lifting while you make the final calls.

Spot on: your banned-words + signature-phrases + self-scoring rubric is the fastest way to stop brand drift. Let me add one lever that compounds those gains — a simple Canonical-to-Variant flow with a built-in “tone checksum.”

• Do: Start every prompt with your fingerprint, guardrails, and channel card; generate in small batches; ask the AI to self-score and fix.
• Do not: Let AI invent facts; publish without a 2-sample human check; use the same draft across channels without adapting structure and CTA.

High-value add: the Canonical-to-Variant method

Create one short “Canonical Message” from a facts-only claims box, then ask the AI to spin channel-specific variants from that source. This keeps tone and facts steady while letting format flex. Add a quick “tone checksum” so each draft proves it followed your rules.

What you’ll need

• Brand fingerprint + 3 tone words.
• Channel cards with word ranges, structure, formality, CTA style.
• Guardrails: 5–10 banned words, 3 signature phrases, reading level.
• CTA bank per channel (3–5 approved CTAs).
• Claims box: the only approved facts, numbers, and disclaimers for this campaign.
• Gold sample per channel and a single shared doc with versioned guardrails (v1.0, v1.1…).

Step-by-step (Memory → Make → Mirror)

1. Memory: Assemble your Brand Pack — fingerprint, tone words, guardrails, CTA bank, claims box. Save as “Guardrails v1.0.”
2. Make (Canonical): Generate a 120–180 word Canonical Message using only the claims box. No channel styling yet.
3. Make (Variants): From the Canonical Message, create variants for 1–2 channels using the channel cards and CTA bank.
4. Mirror (Tone Checksum): Have the AI output a one-line checksum with: tone words used (Y/N), banned words (0 count), signature phrases (present), CTA used (from bank), word count in range.
5. Self-score & fix: Use your rubric (tone, clarity, CTA, factual safety, channel fit). If any score <4, revise once without adding new claims.
6. Human spot-check: Review 2 samples per batch. If both pass with ≤2 minor edits, schedule the batch and promote 1 to gold sample.
7. Update: If the same fix repeats twice, update guardrails and bump version.

Copy-paste prompts (use as-is)

1) Canonical → Variants with Tone Checksum

“You are writing for this brand. Fingerprint: [paste 50–100 words]. Tone words: [3 words]. Reading level: [e.g., Grade 7]. Banned words: [list]. Signature phrases: [list]. CTA bank by channel: [list CTAs per channel]. Claims box (use only these facts): [paste facts, numbers, disclaimers]. Channel cards: [e.g., LinkedIn 180–220 words, problem → 3 steps → 1 CTA question, professional-warm | Email 120–180 words, benefit-first subject, single link CTA | Instagram: 1 hook + 3 bullets + 1 CTA]. Task: 1) Write a 150-word Canonical Message using only the claims box in plain English. 2) From that Canonical Message, create on-brand variants for [Channels]. Constraints: avoid clichés; no banned words; include at least one signature phrase; choose a CTA from the bank. After each variant, output a Tone Checksum line: Tone OK? [Y/N]; Banned words count [#]; Signature phrases present [list]; CTA used [which]; Word count [#]. Return: Canonical Message, then channel variants, each with its Tone Checksum.”

2) Gold Sample Comparator

“Here is the gold sample for [channel]: [paste]. Here is the new draft: [paste]. Compare and list 5 specific differences in tone, structure, and CTA. For any difference that breaks our guardrails, propose a one-sentence fix and apply it. Re-score on tone, clarity, CTA, factual safety, channel fit (1–5). Return the revised draft only if all scores ≥4.”

Worked example (fill-in template)

• Brand fingerprint: “Practical, no-drama wellness tips for busy professionals 40+. Tone: warm, encouraging, evidence-aware. Plain English, short steps, no hype.”
• Tone words: warm, practical, confident
• Banned words: revolutionary, disrupt, hack, guarantee, cutting-edge
• Signature phrases: “plain English,” “do-able steps,” “results you can feel”
• CTA bank:
  • Email: “Book a 10‑minute call,” “Reply ‘guide’ for the checklist.”
  • LinkedIn: “Try one step this week—what will you pick?”
  • Instagram: “Save this for later and try it tonight.”
• Claims box: “Program length: 4 weeks. 3x 20‑minute sessions/week. No equipment needed. Based on low-impact mobility routines. Not medical advice.”
• Channel cards:
  • Email: 140–180 words, benefit-first subject, 1 link CTA.
  • LinkedIn: 180–220 words, problem → 3 steps → 1 CTA question.
  • Instagram: 1 hook line + 3 bullets + CTA, casual.

Common mistakes & quick fixes

• Asset sprawl → Keep one Brand Pack doc with version labels; copy into every prompt.
• Fact creep → Use a claims box and instruct “use only facts provided.”
• Monotony → Rotate 1 of your 3 tone words per campaign; keep 2 constant.
• CTA mismatch → Force the AI to pick from the CTA bank; review at checksum time.
• Over-editing → If you edit more than 2 sentences, fix the prompt or guardrails, not the draft.

7-day plan (light but effective)

1. Day 1: Draft fingerprint, tone words, banned words, signature phrases.
2. Day 2: Write channel cards and a CTA bank per channel.
3. Day 3: Build your claims box for the current campaign.
4. Day 4: Run the Canonical → Variants prompt for one channel; save a gold sample.
5. Day 5: Add the Tone Checksum and Self-Score loop; human-check two drafts.
6. Day 6: Expand to a second channel; use the Comparator to align to your gold sample.
7. Day 7: Review edit time and pass rate; update guardrails; mark v1.1.

Bottom line: Consistent inputs produce consistent outputs. Lock your Brand Pack, write a Canonical Message from a claims box, generate variants per channel, and enforce with a quick tone checksum. Small effort now, compounding consistency every week.

Great structure on Coach Mode. One small refinement: instead of limiting replies to “under 80 words,” ask for 2 short bullets (max 12 words each). Models count words loosely; tight bullets are more reliable and easier to scan.

Try this now (under 5 minutes): pick one sentence that feels clunky. Paste the prompt below and choose one of the two options you get back. Replace the original and read the paragraph once. Stop there.

Copy-paste prompt: Calm Coach – Lite

You are my calm writing coach. Audience: [describe]. Goal: [one tiny goal]. Tone: [friendly | formal | direct]. I’ll paste 1–2 sentences (max 60 words). Steps: 1) Ask one clarifying question (max 10 words). 2) Tag each clause [Keep]/[Cut]/[Rewrite] with 6-word reasons. 3) Give two rewrite options, each one sentence, 12–16 words. 4) After each option, add one 8-word reason. 5) End with two bullets: “Flow check” (10 words) and “What to do next” (10 words). Wait for me to say “next.”

What you’ll need:

• One sentence or a short 2-sentence snippet.
• Two minutes of quiet and your current draft.
• Optional: your “voice anchor” word (for example, practical, kind, direct).

How to run it (step-by-step):

1. Frame: Fill in Audience, Goal, Tone in the prompt.
2. Paste: Share 1–2 sentences (max 60 words).
3. Answer: Respond to the coach’s one question in a short phrase.
4. Pick: Choose Option A or B. Tweak one word to keep your voice.
5. Flow check: Read the paragraph aloud once. If you feel a bump, say “next” and fix just that bump.

Worked micro-example

Original: “I wanted to quickly reach out and see if you might have time to talk about the proposal sometime next week.”

• [Cut] “wanted to quickly reach out” — filler, slow
• [Rewrite] “might have time” — uncertain ask
• [Keep] “next week” — useful timing

• Option A: “Can we schedule 20 minutes next week to discuss the proposal?” — direct, clear ask
• Option B: “Does next week work for a 20‑minute proposal chat?” — friendly, light tone

Tweak: choose A and swap “schedule” for “lock in.” Done in 3 minutes.

Premium upgrade: Coach Mode Plus (voice memory)

Ask AI to build and reuse a tiny “Voice Card” so your tone stays consistent across sessions.

Copy-paste prompt: Voice Card (save this)

Create and maintain my Voice Card. From my edits, distill 3 traits, 5 preferred words, 3 banned phrases. Show it as: Traits, Preferred Words, Banned Phrases. When I say “apply voice,” use it for suggestions. Always ask: “Any non‑negotiables to preserve?” Update the card at session end with 2 bullets of changes learned.

When you need proof, not puff

Copy-paste prompt: Evidence Nudge

Scan this sentence for a claim. Ask me for one proof (number, example, or client quote). If proof is missing, offer one placeholder line starting with “For example,” and mark it [Proof Needed]. Keep reply to two bullets of 12 words.

What to expect:

• Immediate: Two clear options and a calm next step.
• By week 1: Faster time to first usable sentence; steadier tone via the Voice Card.
• Ongoing: Short, repeatable passes that finish drafts without losing your voice.

Common mistakes and fixes:

• Trying to fix the whole draft → Limit to 1–2 sentences per pass.
• Accepting options unchanged → Tweak one word to sound like you.
• Over-editing → One flow check; schedule a second pass later.
• Vague goals → Name a micro-goal: tone, clarity, or brevity (pick one).
• No evidence → Run the Evidence Nudge on any claim.

Insider tip: Ask for a “diff view” so you can see the change without losing your original. Use markers like [-deleted-] and {+added+}. It keeps decisions simple.

7-day action plan

1. Day 1: Run Calm Coach – Lite on one sentence; log the chosen option.
2. Day 2: Build your Voice Card; “apply voice” on a fresh sentence.
3. Day 3: Edit one paragraph (under 120 words) with [Keep]/[Cut]/[Rewrite].
4. Day 4: Use Evidence Nudge on two claims; add one concrete proof.
5. Day 5: Create two subject lines; pick by reading aloud.
6. Day 6: Diff view a tricky sentence; keep the calmer version.
7. Day 7: Review: what traits and words consistently win? Update the Voice Card.

High-value cue to add to any prompt:

End every suggestion with: “If you only changed three words, which?” This keeps edits tiny and preserves your voice.

Closing nudge: Keep sessions small, options few, and wins visible. One sentence, two options, one tweak. That’s how AI becomes a patient coach—and how your drafts start finishing themselves.

Spot on: the verbatim anchor + approval is the credibility combo. Let’s add two pro moves to make this fast, repeatable, and trustworthy — even if you’re not a writer.

Two upgrades that change the game

• Quote Bank: mine your notes once, create a tagged list of exact lines you can reuse for months.
• Proof Ladder: a simple check that nudges each testimonial from vague to specific without inventing facts.

What you’ll use

• Your interview notes or transcript (timestamps if you have them).
• Basic context: role, company, how long they used your product.
• An AI chat tool and a simple doc or spreadsheet.

Step-by-step (quick, honest, repeatable)

1. Build your Quote Bank. Pull 10–15 short, specific lines. Tag each with one word: Outcome, Emotion, Obstacle, Skepticism, Feature, Timeframe, or Metric. Keep them verbatim. If you have timestamps, keep them — they speed approval later.
2. Pick 3 anchors. Choose quotes that cover different angles (e.g., speed, cost, confidence). Aim for one Outcome, one Emotion, one Skepticism → Result.
3. Run each through the Proof Ladder. Ask: does it name a problem, action, result, and one proof element (a number, timeframe, or named process/tool)? Add only what you truly have. Leave gaps in brackets.
4. Generate 2–3 variants per anchor. Concise, narrative, and data-light. Keep one exact phrase. No superlatives. End with role.
5. Compliance scrub. Avoid words like “guaranteed,” “best ever,” “always,” “cure,” “100%.” Prefer “helped,” “saw,” “consistently,” “one of the best.”
6. Approval pack. Send the options with a short yes/no email and attribution choices (full name + role; role only; anonymous).
7. Place and measure. Put the strongest variant near your primary call-to-action and test against a control. Track approval rate, time-to-publish, and lift on the page.

Copy-paste prompts (use as-is)

Quote Bank Extractor

“You are a detail-preserving assistant. From the interview notes below, extract 10–15 candidate quotes exactly as spoken. For each quote, include: timecode (if available), a short tag from [Outcome, Emotion, Obstacle, Skepticism, Feature, Timeframe, Metric], and a 1-line context note (role, product usage length). Do not paraphrase or add facts. Output as a simple list. Interview notes: [PASTE NOTES/TRANSCRIPT].”

Testimonial Builder (3 variants, honesty-first)

“You are an honest testimonial editor. I will give you: one exact quote (the anchor), extra context (role, company, product usage length), and any measured result. Create three variants: A) 2-sentence concise, B) 3–4 sentence narrative, C) data-light version for when numbers are uncertain. Rules: keep the anchor phrase exactly as said; add one proof element (number or timeframe) if provided; include a brief ‘initial doubt’ clause only if present; avoid superlatives and guarantees; end with attribution ‘— Name, Role’ or ‘— Role’ if name not approved; flag missing info with [brackets]. Do not invent facts. Inputs: [ANCHOR QUOTE], [ROLE], [COMPANY], [USAGE LENGTH], [MEASURED RESULT], [INITIAL DOUBT (if any)].”

Approval Email Draft

“You are a helpful coordinator. Draft a short approval email for the testimonial below. Include: the 3 variants, a request to confirm or edit, options for attribution (full name + role, role only, anonymous), and a clear yes/no line the person can reply with. Keep it under 120 words. Testimonial options: [PASTE VARIANTS].”

The Proof Ladder (use this mini-check)

• Problem: what hurt?
• Action: what they used or did.
• Result: what changed.
• Proof: number, timeframe, or named process/tool.
• Emotion: one human feeling word (relieved, confident, calm).

Example (how it reads)

• Anchor quote: “We cut onboarding from 3 weeks to 2 days.”
• Context: HR Manager at a mid-size company, used the product 6 months, initially skeptical.

• Concise: “We cut onboarding from 3 weeks to 2 days.” After 6 months, our team spends less time chasing paperwork and more time training — a huge relief. — HR Manager
• Narrative: “We cut onboarding from 3 weeks to 2 days.” I was skeptical at first, but within a month the bottlenecks disappeared and new hires were productive by day two. After 6 months, it’s our new normal. — HR Manager
• Data-light: “We cut onboarding from 3 weeks to 2 days.” It’s significantly faster and far less stressful for our team. — HR Manager

Common mistakes & quick fixes

• Mixing voices: Don’t stitch multiple speakers together. Fix: One person per testimonial.
• Stacking claims: Five benefits read like an ad. Fix: One clear result + one feeling.
• Missing timeframe: “We grew 40%” invites doubt. Fix: Add “in [X months]”.
• Inventing proof: Numbers you can’t source. Fix: Use qualitative phrasing and confirm later.
• Publishing without consent: Risky. Fix: Always get approval and preferred attribution.

45-minute action plan

1. Run the Quote Bank prompt on one interview (10 minutes).
2. Pick 3 anchors and score them on the Proof Ladder (5 minutes).
3. Generate the 3-variant set with the Testimonial Builder (10 minutes).
4. Light compliance scrub and human edit (10 minutes).
5. Send approval email with options (10 minutes).

Final reminder: AI is your tidy editor, not your truth engine. Keep one exact phrase, add only proof you truly have, and get the person’s sign-off. Do that, and your testimonials will feel honest, read smoothly, and convert.

Yes—AI can be a calm, step-by-step writing coach. The trick is to set gentle rules so it asks one question at a time, keeps your voice, and moves in tiny, doable steps.

Quick context: Most overwhelm comes from dumping a whole draft into AI and asking for “help.” You’ll get big rewrites and lose your voice. Instead, we’ll use a reusable “Coach Mode” that slows the AI down and gives you clear next steps.

What you’ll need:

• 10–15 minutes and one paragraph or sentence.
• A simple brief you can paste first time: Audience, Goal, Tone.
• The Coach Mode prompt (below) saved in Notes.

Do / Do not (checklist):

• Do start with one tiny goal (tone, clarity, or brevity—pick one).
• Do paste only a sentence or short paragraph.
• Do ask for two short options and a one-line reason “why it works.”
• Do read aloud and tweak one word to keep your voice.
• Do not ask for a full rewrite of the whole piece.
• Do not accept the first suggestion unchanged.
• Do not move on until the paragraph flows end-to-end.

Copy-paste this once: Coach Mode (patient guidance)

You are my patient writing coach. Audience: [describe]. Goal for this session: [one tiny goal]. Tone: [friendly | formal | direct]. Piece type: [email | paragraph | post]. Rules: 1) Ask me only one question at a time. 2) If I paste more than 120 words, ask me to choose a 1–3 sentence segment. 3) Use line tags [Keep], [Cut], [Rewrite] with one short reason each (max 8 words). 4) Offer at most two options per step. 5) Wait for me to reply “next” before moving on. 6) Keep replies under 80 words unless I say “expand.” Start by asking the single most helpful clarifying question in 12 words or fewer—nothing else.

How to run a session (step-by-step):

1. Frame (1 minute): Paste the Coach Mode prompt with Audience/Goal/Tone filled in.
2. Share (1 minute): Paste one sentence or a short paragraph (under 120 words).
3. Diagnose (2 minutes): The AI asks one question; you answer briefly. It tags your lines [Keep]/[Cut]/[Rewrite] with reasons.
4. Choose (3 minutes): It offers two tiny rewrite options. Pick one and tweak one word.
5. Flow check (2 minutes): Read the full paragraph aloud. If a bump appears, say “next” and fix just that spot.
6. Close (1 minute): Ask for a 1-sentence summary of what changed so you learn the pattern.

Insider trick: The [Keep]/[Cut]/[Rewrite] tags are “green-pen” editing—low pressure, easy decisions. This preserves your voice while trimming clutter.

Worked example (from clunky to clear):

Original sentence: “I’m reaching out to touch base about the meeting we are hoping to schedule next week, if that still works for you.”

Coach Mode will ask a single question: “What tone do you want?”

You: “Friendly and direct.”

Coach tags:

• [Cut] “reaching out to touch base” — filler phrase
• [Rewrite] “the meeting we are hoping to schedule” — vague timing
• [Keep] “next week, if that still works for you” — confirms ease

Coach offers two options:

• Option A: “Can we lock our meeting for next week?” — direct, simple ask
• Option B: “Does next week still suit for our meeting?” — friendly, light

You choose A and tweak: “Can we lock in our meeting for next week?”

Flow check: Read aloud. Smooth. Done in under 5 minutes.

When you need a micro-polish (two fast alternatives)

Rewrite this sentence to be [tone]. Keep my words where possible. Max 12–16 words. Remove filler. Active verbs. No clichés. Give two versions and a 6-word reason after each. Sentence: “[paste your sentence]”.

Simple brief you can reuse before any session:

Audience: [who is reading]. Goal: [what I want them to do/feel]. Non‑negotiables: [phrases, facts to keep]. Off‑limits: [jargon, claims, topics]. Max length: [word limit].

What to expect:

• Immediate: One helpful question, two clear options, zero overwhelm.
• By week 1: Faster time to first usable sentence and steadier tone.
• Ongoing: A repeatable rhythm—small wins that finish drafts.

Common mistakes and quick fixes:

• Problem: Asking for “improve this” with no goal. Fix: Name one micro-goal first.
• Problem: Pasting long sections. Fix: Limit to 1–3 sentences per pass.
• Problem: Losing your voice. Fix: Keep one signature word or phrase.
• Problem: Endless tweaks. Fix: One “flow check,” then stop or schedule a second pass later.

5-session action plan (15 minutes each):

1. Session 1: Run Coach Mode on one email sentence; choose between two options.
2. Session 2: Use the micro-polish prompt on your opening line; log the winning version.
3. Session 3: Coach Mode a full paragraph (under 120 words) with [Keep]/[Cut]/[Rewrite].
4. Session 4: Create two subject lines using the micro-polish prompt; read aloud and pick one.
5. Session 5: Review what worked; write a 2-line “voice rules” card from your best examples.

Premium tip: Tell the AI to “ask for evidence.” If a sentence makes a claim, it should ask, “What proof or example can we add?” This gentle nudge adds credibility without extra fluff.

Closing nudge: Keep it tiny—one question, two options, one tweak. That’s how AI becomes a patient coach and how you build momentum, sentence by sentence.

Yes — you can get real, fast wins. Start small, measure, and improve.

Here’s a simple, practical checklist to turn your sales history into usable forecasts without hiring a data scientist. Think: less guesswork, fewer stockouts, and lower holding costs.

What you’ll need

• Sales history (6–24 months) by SKU — weekly or daily.
• A spreadsheet (Excel or Google Sheets) or a basic forecasting app.
• Supplier lead times (in weeks) and your target safety cushion (1 week is a good start).
• 15–30 minutes each week to review and adjust.

Step-by-step (do this once, then repeat weekly)

1. Collect: Export date, SKU, units sold. Flag promotions, holidays, or store closures.
2. Clean: Remove obvious errors and fill short gaps. Mark one-off spikes as exceptions.
3. Calculate averages: Find average weekly sales for each SKU (simple mean).
4. Set safety stock: Start with 1 week of average sales (adjust later for variability).
5. Compute reorder point: Reorder point = (average weekly sales × lead time in weeks) + safety stock.
6. Pilot: Run this for your top 10–20 SKUs for 4–8 weeks. Track accuracy and tweak safety stock.

Worked example (fast, copyable)

• SKU: Coffee Beans
• Average weekly sales: 20 units
• Supplier lead time: 2 weeks
• Safety stock: 1 week of sales = 20 units
• Reorder point = (20 × 2) + 20 = 60 units. Place order when inventory ≤ 60.

Common mistakes & fixes

• Mistake: Using too little data. Fix: Use at least 6 months — include seasonality notes.
• Mistake: Ignoring promotions. Fix: Tag promotional periods and exclude or model them separately.
• Mistake: Forecasting too many SKUs at once. Fix: Start with top 20 SKUs (Pareto rule).
• Mistake: Never reviewing the model. Fix: Weekly check-ins and monthly tweaks.

Copy-paste AI prompt (use in ChatGPT or a forecasting tool)

“I have a CSV with two columns: date (YYYY-MM-DD) and units_sold for each SKU. For each SKU, produce a 8-week weekly demand forecast, identify weeks with anomalous spikes or dips, and recommend a reorder point given lead time in weeks and safety stock = 1 week of average sales. Output a CSV with columns: SKU, avg_weekly_sales, lead_time_weeks, safety_stock, reorder_point, notes_on_anomalies.”

30-day action plan

1. Day 1–7: Extract data for top 20 SKUs and compute averages and reorder points.
2. Week 2–4: Run weekly checks, compare forecast vs actual, adjust safety stock for volatile SKUs.
3. End of month: Measure stockouts, overstock %, and weeks of inventory. Celebrate small wins and expand to next 20 SKUs.

Reminder: The goal is better decisions, not perfect predictions. Start small, review often, and use simple rules — you’ll cut waste and keep customers happier.

5‑minute quick start: paste last month and this month’s top 5 KPIs into an AI with: “Write a 120‑word, delta‑first board summary: what changed, why (if known), risk, and one clear ask. Use only numbers provided. If a cause is unknown, say ‘unknown’ and suggest one test to confirm.” You’ll see a clean, decision‑ready opening in minutes.

Big idea: boards don’t need more pages — they need exceptions, causes, and asks. Make your report delta‑first (what changed vs last month), keep a tight template, and add a human sign‑off. That’s the reliable path to “automatic” monthly reports.

Insider tricks that make this work

• Delta‑only narrative: report only items that moved beyond a threshold (e.g., +/- 3% or material dollars). Everything else goes to an appendix.
• RAG + Why + Now + Next: each KPI gets a Red/Amber/Green status, one cause (or “unknown”), the current risk, and the next action.
• Truth Map: freeze KPI names and sources (e.g., Revenue → Sheet KPI!B3). Ask AI to echo source cells so reviewers can cross‑check.
• Few‑shot style control: paste last month’s approved summary so the AI mirrors tone and structure.

What you’ll need

• One trusted sheet or BI export with named ranges for each KPI (e.g., Revenue_MoM, Churn_Rate).
• A one‑page template with slots: Executive Summary, KPI Highlights, Risks, Actions/Asks, Appendix.
• An AI assistant (chat or automation) to draft the narrative from numbers.
• A reviewer checklist for accuracy and tone.

Step‑by‑step

1. Define materiality: set thresholds (e.g., “call out only if change >= 3% or >= $25k”). This keeps the story tight.
2. Snapshot the month: copy KPI values into a new “Snapshot_YYYY‑MM” tab so the month’s numbers never shift.
3. Name your ranges: give each KPI a stable name. Avoid cell letters in prompts; use the names instead.
4. Template once: write a one‑page outline with fixed sections and a 120‑word cap for the exec summary.
5. Feed the AI: provide this month, last month, thresholds, and any short context notes (campaigns, outages, pricing changes).
6. Generate two versions: Board (crisp, action‑oriented) and Stakeholder (plain English, slightly warmer tone).
7. Review to certify: verify numbers, remove speculation, confirm the “ask,” and add any confidential context.
8. Version and file: export as “Board_Report_YYYY‑MM_v1.0.pdf” and log edits for auditability.

Copy‑paste prompt: Board delta narrative (robust)

“You are an experienced company secretary preparing a delta‑first board summary. Use only the data provided. If a cause is unknown, write ‘unknown’ and propose one test to validate the cause. If a value is missing, output ‘TBD’ and add a flag. Echo the source name in brackets after each KPI.

Inputs: This_Month: Revenue = $1.28M [Revenue_MTD]; Churn = 2.1% [Churn_Rate]; Cash_Runway = 8.7 months [Runway_Months]; SQLs = 410 [SQLs]; NPS = 46 [NPS]. Last_Month: Revenue = $1.20M; Churn = 2.3%; Cash_Runway = 9.0 months; SQLs = 360; NPS = 44. Thresholds: material_change = 3% or $25k.

Produce exactly: 1) Executive summary (max 120 words, delta‑first). 2) KPI highlights (5 bullets): each bullet = Status (R/A/G), KPI name with delta, one cause (or ‘unknown’), risk (if any), and next action. 3) One clear board ask (decision or resource). 4) Source echo: list KPI → [Source_Name]. Tone: calm, precise, non‑technical.”

Quality guardrail prompt: Auditor pass

“Act as a compliance reviewer. Compare the draft board summary to the KPI inputs below. Flag any math errors, unit mismatches (%, $, months), or claims without a stated cause. Confirm R/A/G status matches thresholds (3% or $25k). Output a list of fixes. Do not rewrite the narrative.”

Stakeholder variant prompt

“Rewrite the approved board summary into a 3‑paragraph update for external stakeholders. Keep numbers accurate, avoid jargon, remove internal risks that aren’t public, and include one customer‑impact sentence and one next‑month focus.”

Example flow (what to expect)

• Time: first month saves the heavy drafting (often 60–70%). Month two gets faster as your template stabilizes.
• Output shape: a crisp 120‑word opening, 5 exception‑based bullets with R/A/G, one board ask, and a short appendix for non‑material moves.
• Quality: fewer formatting mistakes, clearer “why,” and a consistent tone your board will recognize month to month.

Common mistakes and easy fixes

• Version confusion: files named “Final_v9”. Fix: use “YYYY‑MM_v1.0” and increment only after reviewer approval.
• Metric drift: someone quietly changes a KPI formula. Fix: freeze monthly snapshots and add a “Methodology Changes” note when definitions change.
• Speculation: AI invents causes. Fix: require “unknown” plus one test; reviewer verifies or replaces.
• Everything is ‘important’: bloated pages. Fix: enforce materiality thresholds and push the rest to an appendix.
• Unit mix‑ups: % vs basis points vs dollars. Fix: standardize units in the prompt and run the auditor pass.

14‑day action plan

1. Days 1–2: Pick 5 KPIs, set thresholds, and create the one‑page template.
2. Days 3–4: Add named ranges and a Snapshot_YYYY‑MM tab; document your Truth Map.
3. Days 5–6: Run the Board delta prompt with last and this month’s numbers; generate the Stakeholder variant.
4. Day 7: Reviewer uses the Auditor prompt; finalize v1.0 and distribute.
5. Days 8–10: Collect board feedback; tweak thresholds, tone, and the “ask” section.
6. Days 11–14: Automate the data export into the snapshot and schedule your drafting session (same day each month).

Pro move: attach last month’s approved summary to the prompt as a style example. You’ll get near‑identical tone and structure, which boards love.

Start with delta‑first summaries and a strict reviewer pass. After two clean cycles, scale to more KPIs and sections. AI drafts fast; you keep the judgement and the trust.