aaron

Good starter: the thread title is clear and focused—exactly what we need to build practical prompts.

Hook: If you want AI to teach math step-by-step without handing over the final answer, you need prompts that force guided reasoning, checkpoints, and progressive hints.

Problem: Most prompts either give full solutions (kills learning) or only restate the problem (useless). You want an explanation that teaches the process, surfaces common mistakes, and preserves the student’s effort.

Why it matters: Teaching that keeps students engaged improves retention, diagnostic insight, and reduces rework. That translates to higher pass rates, less one-to-one tutoring time, and measurable learning gains.

Experience / lesson: I’ve used scaffolded prompts to convert AI from an answer machine to a tutor: chunk problems, require reasoning steps, ask for formative checks, and only reveal the final answer on request.

1. What you’ll need: the problem statement, target student level (e.g., middle school algebra), desired teaching style (concise, Socratic, worked-example), and allowed hint count.
2. How to do it (step-by-step):
   1. Start with a role and objective: “You are a patient math tutor.”
   2. Specify output structure: “List assumptions, then 4 numbered steps, then two short practice questions.”
   3. Prevent answer leaks: “Do not provide the final numeric result unless the student types ‘SHOW ANSWER’.”
   4. Include formative checks: “After each step, pose one quick question for the student to confirm understanding.”
   5. Set hint policy: “Provide up to 3 hints; each hint reveals progressively more info.”
3. What to expect: a stepwise explanation, short checkpoint questions, controlled hints, and optional practice items. Expect initial tuning—adjust phrasing to avoid partial answers leaking.

Copy-paste prompt (use as baseline):

“You are a patient math tutor for a [student_level] student. Given the problem: ‘[paste problem here]’: (1) State any assumptions and definitions in one sentence. (2) Provide 4 clear, numbered instructional steps that teach the method—after each step, include a single yes/no checkpoint question for the student. (3) Do not calculate or reveal the final answer under any circumstance unless the student types ‘SHOW ANSWER’. (4) Offer up to 3 hints labeled HINT 1/2/3; each hint must be progressively more revealing. (5) End with two short practice problems that use the same method but different numbers.”

Prompt variants:

• For exams: add “Use exam-style concise language; include one common trap.”
• For younger learners: add “Use a friendly voice and simple vocabulary. Use analogies.”
• For advanced: add “Include one brief proof or justification.”

Metrics to track (KPIs):

• Student comprehension rate (pre/post quiz improvement %)
• Hint usage rate (hints per session)
• Answer reveal rate (% of sessions where ‘SHOW ANSWER’ used)
• Time-to-understand (minutes per concept)

Mistakes & quick fixes:

• If AI gives final answers: tighten “Do not reveal” clause and add a penalty line: “If final answer is given, restart.”
• If explanations are too long: request “concise steps, max 2 sentences each.”
• If hints reveal too much: make hints strictly progressive and limited to 3.

1-week action plan (practical):

1. Day 1: Pick 5 representative problems; run baseline prompt; capture outputs.
2. Day 2: Tweak prompt phrasing (role, checkpoints); re-run and compare clarity.
3. Day 3: Test with 3 students; record hint and reveal usage.
4. Day 4: Measure comprehension with a short quiz; record improvement.
5. Day 5: Optimize wording based on failures; reduce leaks.
6. Day 6: Create a prompt variant for exam prep and one for remedial help.
7. Day 7: Consolidate best prompt, document KPIs, and roll out to learners.

Your move.

Nice callout: The 5-minute quick win is exactly right — fast tests remove guesswork. I’ll add what to measure, exact steps that non-technical people can follow, and a worked example with targets so you know where to stop, iterate, or scale.

Why this matters: AI speeds copy and segmentation, but product–market fit is demonstrated by repeatable revenue signals: people paying, returning, and referring. If you don’t set KPI thresholds before you test, you’ll waste time debating results instead of acting.

Core approach (what you’ll need):

• AI access (ChatGPT or similar)
• One-page landing builder (simple tool or template)
• Email tool and payment processor (Stripe/PayPal)
• Survey or booking tool (Typeform/Calendly or built-in form)

Step-by-step (do this):

1. Pick 2 audience segments to test (A and B). Document their core pain in one sentence each.
2. Run the AI prompt (below) to generate 3 headlines, a 120-word blurb, and a 5-question presale survey per segment.
3. Build one landing page per segment with: headline, blurb, price, limited spots, payment button, and the survey link.
4. Drive 100–300 visits per page via email + niche posts or a $50–150 ad test.
5. Track conversions and book 10–20 short calls with paid or committed leads. Use calls to refine pricing and objections.

What to expect: Within 2 weeks you’ll have clear conversion rates and interview themes. Expect noisy data early — focus on paid conversion and repeat interest.

Metrics to track (KPIs & targets):

• Visitor → Opt-in: target 3–8%
• Opt-in → Paid pre-sale: target 8–20% (higher is better)
• Refund requests within 14 days: target <10%
• Retention (first month attendance/engagement): target ≥60%
• Cost per paid lead (if ads): keep under expected LTV / 3

Mistakes & fixes:

• Mistake: Using free opt-ins as the only signal. Fix: Require payment or refundable deposit.
• Mistake: One-segment testing. Fix: Run at least two segments in parallel and compare.
• Mistake: No pre-defined success thresholds. Fix: Set KPI targets before you launch.

Checklist — Do / Don’t:

• Do: Require money or a refundable deposit.
• Do: Book short calls with paid signups.
• Don’t: Treat AI-generated testimonials or engagement as proof of fit.
• Don’t: Over-survey — keep questions tight.

Worked example (copy-pasteable targets):

• Segment: Mid-career marketers. Traffic goal: 200 visitors. Opt-ins (5%) = 10. Paid pre-sales target (15% of opt-ins) = 1–2 paid customers. If you hit ≥2 paid customers at $7/mo with ≤$50 ad spend, iterate and scale. If 0 paid → pivot copy/segment.

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

“You are a marketing copywriter. I run a paid weekly newsletter/short online course for [audience: e.g., mid-career marketers who want to run growth experiments]. Write: 1) three headline variants; 2) a 120-word landing page description emphasizing outcomes and including price and limited spots; 3) a 5-question pre-sale survey focused on pain, current solutions, willingness to pay, and ideal outcomes; 4) a 3-line follow-up email to send after sign-up asking to book a 15-minute call. Keep tone direct, credible, and non-salesy. Also suggest two objections and one short FAQ item to address each objection.”

7-day action plan (exact next steps):

1. Day 1: Choose 2 segments and run the AI prompt for each.
2. Day 2: Build two one-page landing pages + payment + survey.
3. Day 3: Send to email list and post in 3 niche places; start a $50 ad test split between pages.
4. Day 4–6: Collect traffic, run headline A/B, book calls with paid/committed people.
5. Day 7: Review KPIs vs targets, summarize interview themes, decide: iterate, pivot segment, or scale ad spend.

Your move.

Smart call-out: your page-locked facts + two-pass workflow is the backbone. Here’s how to add quality gates and KPIs so you ship a reliable, decision-ready one-pager in 45–60 minutes, every time.

Try this now (under 5 minutes)

• Grab any single page from a report with a key chart or table.
• Paste it into your AI tool using the prompt below.
• Outcome: two clean number boxes with page tags and one-line implications for CFO and COO. Paste them into your template’s top-right corner.

Copy-paste prompt: “From the text I paste next, output exactly: 1) Two labeled number boxes that capture the single most decision-relevant metric each (include timeframe and baseline), 2) One-line implication for a CFO and one for a COO, 3) Exact page reference in brackets for each number. Do not invent figures. If a page ref is missing, label [NO PAGE].”

The snag

Speed without quality drifts into vague, untrusted summaries. The failure modes: missing baselines, source bias, and implications that overreach the evidence.

Why it matters

Executives act on clarity and confidence. A one-pager that pairs page-locked numbers with a single, dated recommendation shortens time-to-decision and increases adoption of the next step.

Field-tested upgrade: add QA gates, source weighting, and a results dashboard

1. Stand up your evidence schema (3 minutes). Decide once how you store facts: Item | Value | Unit | Timeframe | Baseline | Source [page]. Anything missing a page is excluded.
2. Weight your sources (2 minutes). Simple scale: 3 = primary data/multi-study consensus, 2 = reputable analyst/model, 1 = single-model projection/press. Note the score next to each fact.
3. Pass 1 — Extract (15–25 minutes). Use your extraction prompt to fill the evidence bank with page tags and source weights. No interpretation.
4. Triage to decision deltas (5 minutes). Star the top three figures that change a decision (growth rate, share shift, cost delta). Ensure each has timeframe and baseline.
5. Pass 2 — Synthesize (15–20 minutes). Draft headline; 3–5 fact→implication bullets; two number boxes; one top risk; confidence band with reason. Aim for 350–450 words.
6. QA Gate (5–7 minutes). Run the red-team + math check prompts below. Fix anything flagged. If any headline figure lacks a page or baseline, it’s out.
7. Ship and log (3 minutes). Add version label (v1), owner/date on the recommendation, and log KPIs: build time, errors corrected, acceptance.

Copy-paste prompts (add to your toolkit)

• Evidence schema extractor: “Extract only verifiable items into this schema: Item | Value | Unit | Timeframe | Baseline | Source tag [page]. Include direct quotes if they carry conclusions. No interpretations. Flag any missing page as [NO PAGE].”
• Decision-delta synthesis: “Using only the evidence list with page tags, draft: 1) three one-line headlines stating what changed, magnitude, and why it matters; 2) 3–5 bullets where each has one sentence of fact (with [page]) and one sentence of executive implication; 3) two number boxes (label + value + timeframe + baseline + [page]); 4) top risk; 5) confidence (High/Med/Low) with one-line rationale. No new facts.”
• QA red-team: “Identify any claims without page tags, any numbers not present in the evidence, any implication that exceeds the evidence. Return a list of fixes with the exact sentence to change and the matching evidence item.”
• Math and sanity check: “Validate all arithmetic and timeframes (CAGR ranges, percentage-point vs percent). List mismatches, corrected values, and the evidence lines used.”

What to expect

• First run lands at 60–80 minutes; with the schema + QA gate, you’ll settle at 45–60 minutes.
• Expect to drop 10–20% of ‘interesting’ lines that lack pages or baselines. Trust goes up as fluff goes down.

KPIs that show progress

• Time-to-first-draft: target ≤60 minutes.
• Evidence coverage: ≥95% of figures have page tags and baselines.
• Red-team findings per draft: ≤3 after week two.
• Executive acceptance rate: ≥80% approved or forwarded within 48 hours.
• Action adoption: ≥70% of recommendations started by the named date.

Common mistakes and fast fixes

• Mixing forecasts with actuals. Fix: label each figure as Actual or Forecast and include timeframe.
• Percent vs percentage points confusion. Fix: force the math check prompt to flag ‘pp’ vs ‘%’ and correct.
• Implied causality from correlation. Fix: change “drives” to “coincides with” unless the source states causation.
• Overloaded visuals. Fix: replace with two number boxes; cap visuals at two items.
• AI-invented citations. Fix: “No page, no page.” Drop it or locate the page yourself.

1-week plan to operationalize

1. Day 1: Implement the evidence schema and source weights. Run extraction on one report section. Log time.
2. Day 2: Complete extraction, triage decision deltas, and draft the one-pager. Apply QA gate and math check.
3. Day 3: Send to one exec. Track acceptance and requested edits. Update your template.
4. Day 4: Repeat on a second report. Aim for ≤60 minutes. Measure red-team findings.
5. Day 5: Add persona toggle (CFO/COO) implications. Standardize number box formats.
6. Day 6: Create a simple KPI log (time, coverage, findings, acceptance, adoption).
7. Day 7: Review KPIs, codify “no page, no page” and “one verb + date” as non-negotiables.

Bottom line

Keep the two-pass method, then bolt on QA gates, source weights, and KPIs. The result: a one-pager OS that is fast, verifiable, and consistently acted on.

Your move.

Good call on keeping visuals practical and aligned with the message — that’s the single biggest pitfall I see.

Hook: AI-generated art can make dry data memorable. Problem: too many non-technical presenters either overuse abstract images or ignore legal and brand consistency issues. Why it matters: the wrong image undermines trust and slows decisions; the right image speeds understanding and buy-in.

Short lesson from experience: treat AI art like a hired designer — brief it, edit the results, and document usage. That process turns novelty into business outcomes.

1. What you’ll need
   • A simple creative brief (1 paragraph per slide).
   • Access to one image-generation tool (e.g., an AI image generator) and basic image editor (Cropping, contrast, text overlay).
   • Your brand palette and one approved typeface.
   • Checklist for licensing and alt text.
2. How to do it — step-by-step
   1. Write a one-sentence objective for each visual: what should the audience think/decide after seeing it?
   2. Prompt the AI with that objective and style constraints (see copy-paste prompts below).
   3. Select 3 candidates, crop and apply your brand colors, add a 6–8 word caption that reinforces the takeaway.
   4. Confirm licensing, add alt text, and note the source in your slide notes.
   5. Test with one trusted colleague and iterate.

What to expect: 15–30 minutes per slide for your first run; 5–10 minutes per slide once you have templates and saved prompts.

AI prompt (copy-paste):

Create a high-contrast, professional image for a client presentation slide that communicates: “subscription growth acceleration.” Style: clean, minimal, brand palette: navy #0A2342 and accent #F5A623, flat vector style, single focal element, no faces, aspect ratio 16:9. Deliver 3 variants: abstract chart metaphor, symbolic icon with arrow, and simplified landscape with upward path. Provide a one-sentence caption for each.

Prompt variants:

• Executive brief: “Generate 3 professional slide images showing subscription growth using navy and orange, minimal style, 16:9.”
• Creative brief: “Produce 3 conceptual visuals: rising ribbon, staircase of blocks, arrowed path — flat colors, no text.”

Metrics to track

• Decision velocity: time from presentation to decision.
• Slide engagement: % of slides discussed vs skipped.
• Comprehension score: 1–5 rating from stakeholders post-meeting.
• Reuse rate: how often an image is used across decks.

Common mistakes & fixes

• Using images that don’t support the takeaway — Fix: write the objective first.
• Ignoring licensing — Fix: keep a usage log and choose permissive licenses or create internal license notes.
• Style mismatch with brand — Fix: apply a color overlay and consistent caption template.

1. 1-week action plan
   1. Day 1: Pick three slides you want to improve; write one-sentence objectives.
   2. Day 2: Generate 9 images (3 per slide) using the main prompt and 1 variant each.
   3. Day 3: Edit and apply brand colors; add captions and alt text.
   4. Day 4: Run a quick review with one stakeholder; capture feedback.
   5. Day 5: Finalize and log licensing; prepare the slide deck for the next presentation.

Your move.

Quick win: If you haven’t already, run the two-cell check: overall conversion rates for A vs B and one segment (new vs returning). If the lift is concentrated in one group, that’s your fastest clue about a driver.

Good point in your post — AI is for hypothesis generation, not magic causality. I’ll add a practical, outcome-oriented path to move from signal to confident next steps you can run in a week.

Why this matters

Finding where the lift actually comes from changes what you build next. Fix the channel/device that drove the lift and you scale; chase the wrong signal and you waste time and revenue.

What I’ve learned (short)

Most reliable wins come from a two-step loop: 1) detect concentrated lift with simple slicing, 2) validate with a focused follow-up (replicate or targeted experiment). AI speeds up step 1 and suggests plausible mechanisms for step 2.

Step-by-step (what you’ll need and how to do it)

1. Gather: CSV with visitor ID, variant, outcome, and 3 attributes (device, new/returning, source).
2. Clean: remove duplicates, null variants, and impossible values.
3. Simplest checks: calculate conversion rate and N for A and B overall and for each attribute level.
4. Flag candidates: find segments where absolute lift > overall lift × 2 and N >= 100 (adjust threshold for your traffic).
5. Run quick stats: compute difference-in-proportion z-score or use your spreadsheet’s stats add-on to get p-value for that subgroup.
6. Use AI: paste the summary (counts, rates, p-values) into the prompt below to get prioritized hypotheses and checks.

Copy-paste AI prompt

“You are a data-savvy product manager. Here’s the summary: overall A: 12% (n=20,000), B: 14% (n=20,000). Mobile users A: 10% (n=8,000), B: 16% (n=8,200). Desktop users A: 13% (n=12,000), B: 12% (n=11,800). Provide: 1) three prioritized causal hypotheses for the mobile lift, 2) two quick validation checks I can run in data, 3) a follow-up experiment design to confirm causality (sample sizes and success metric).”

Metrics to track

• Conversion rate (overall and by segment)
• Absolute lift and relative lift
• Sample size per segment
• p-value / confidence interval
• Replication result (same metric in follow-up)

Common mistakes & fixes

• Small N in winning subgroup — fix: don’t act until N≥100–200 or replicate.
• Multiple slicing leading to false positives — fix: prioritize hypotheses and control FDR or pre-specify tests.
• Instrumentation errors — fix: check event fires and variant assignment logs.

1-week action plan

1. Day 1: run slices, compute rates & Ns.
2. Day 2: run simple stats and feed summary to AI prompt above.
3. Day 3: run two validation checks (balance & instrumentation).
4. Day 4: design targeted follow-up (replicate or narrow audience) with specified N.
5. Day 5–7: launch follow-up, monitor primary metric and segment performance.

Your move.

Quick win: In 5 minutes generate a neutral background image or a 10–15s music loop, export it, and save a one-line title plus three keywords. That single test proves the cycle: create → label → store.

The problem: AI makes creation fast, but most beginners skip the business steps—metadata, licensing, and compliance—so assets don’t sell or get removed.

Why it matters: If you treat AI content like an asset class and systematize metadata + licensing, you turn one-off experiments into steady income streams and defensible rights positions.

Experience / lesson: I’ve seen creators publish hundreds of assets; the ones that scale follow a repeatable checklist and log their tool terms. Simple edits and clear metadata multiply acceptance and downloads.

What you’ll need

• An AI tool that permits commercial use (confirm terms and save a screenshot of the T&Cs).
• Basic editor: photo editor for images; any DAW or audio editor for music.
• Accounts on 2–3 marketplaces (stock sites, music libraries) or your own storefront.
• Generation log: date, tool, prompt/settings, license terms.

Step-by-step (images)

1. Pick demand: backgrounds, neutral lifestyle props, patterns, textures.
2. Generate 6–12 variations; pick top 3 and do light edits (crop, color, retouch artifacts).
3. Strip recognisable faces, logos, brands. If a person appears, remove or get a model release.
4. Write metadata: title, 8–15 keywords, short usage notes (e.g., “web banner, royalty-free”).
5. Upload with the license you want (royalty-free vs rights-managed) and keep the original generation log attached.

Step-by-step (music)

1. Define mood, tempo, length, and use-case (loop, bed, sting).
2. Generate 4–8 short stems/loops; consolidate best takes and normalise levels.
3. Export WAV for libraries, MP3 for previews; include tempo/key and intended license in metadata.
4. Upload, choose sync/royalty-free options, and save your generation log.

Copy-paste AI prompts

• Image prompt (no people, high-res): “Create a high-resolution 4000x3000px seamless neutral studio background: soft warm light, subtle texture, desaturated teal and beige tones, minimal shadows, no text, no logos, commercial use allowed.”
• Music prompt (loopable): “Produce a 15-second instrumental loop, 95 BPM, mellow corporate ambient, soft synth pad, simple guitar pluck arpeggio, warm reverb, clean mix, royalty-free use—export as 44.1kHz WAV and mp3 preview.”

Metrics to track

• Assets uploaded per week; acceptance rate by platform.
• Downloads/views per asset and conversion to sales.
• Average revenue per asset and time-to-first-sale.

Common mistakes & fixes

• Bad metadata → low discoverability. Fix: spend 5–10 minutes per asset on 8–15 targeted keywords.
• Ignoring tool T&Cs → takedown risk. Fix: save time-stamped T&C screenshot and your generation log.
• Poor audio levels → rejections. Fix: normalise and check LUFS for platform guidelines.

One-week action plan

1. Day 1: Run the 5-minute quick win; save log and T&Cs screenshot.
2. Day 2–3: Produce 10 image variations + metadata; upload to 1 stock site.
3. Day 4–5: Produce 10 music loops; export WAV + preview MP3; upload to 1 music library.
4. Day 6: Review acceptance rates and tweak prompts/edits.
5. Day 7: Calculate time-per-asset and set a weekly target (e.g., 5 assets/week).

Your move.

—Aaron

Quick note: Good point about prioritizing headline variation and testing — that’s exactly where Quality Score moves fastest.

Why this matters: Google’s Quality Score is driven by expected click-through rate (CTR), ad relevance, and landing page experience. Better headlines and descriptions increase CTR and perceived relevance, which lowers cost per click and improves position.

What I’ve learned: AI speeds up ideation and variant generation, but you must pair AI output with disciplined testing and landing-page alignment. AI gives volume; you decide the strategy.

What you’ll need:

• List of target keywords and top-performing existing ads
• Access to your Google Ads account for experiment setup
• Simple spreadsheet or Google Sheet to track variants and metrics
• An AI tool that can generate multiple ad versions (ChatGPT, Claude, etc.)

Step-by-step process:

1. Gather: Export top 20 keywords by volume and current top 10 ads with CTR and Quality Score.
2. Seed AI: Use the prompt below to generate 8 headline variants and 4 description variants per keyword group.
3. Filter: Keep variants that include the exact keyword, a benefit/promise, and a clear CTA. Limit to 3 headlines + 2 descriptions per ad group for A/B testing.
4. Launch experiments: Use responsive search ads (RSA) or ad variations to rotate evenly for 2 weeks.
5. Analyze: After 2 weeks, pick winners by CTR lift and conversion rate; roll winners to expanded testing with landing page tweaks.

AI prompt (copy-paste):

“You are an ad copywriter. For the keyword set: [insert keywords separated by commas], write 8 short headlines (max 30 characters each) and 4 descriptions (max 90 characters each). Each headline should include one of the keywords exactly once, a clear benefit, and a call-to-action. Keep tone professional, trust-building, and tailored to buyers over 40. Include variants that emphasize speed, price, and guarantee. Return as a numbered list labeled Headlines and Descriptions.”

Metrics to track:

• Quality Score (overall and per keyword)
• CTR by ad variant
• Conversion rate and cost per conversion
• Impression share and average CPC
• Landing page bounce rate and load time

Common mistakes & fixes:

• Keyword stuffing – fix: match intent, use exact keyword once in headline, vary phrasing in other headlines.
• Too many variants live at once – fix: test 3×3 matrix (3 headlines x 2 descriptions) per ad group.
• Ignoring landing page – fix: ensure headline messaging matches landing page H1 and CTA.
• Not segmenting by device – fix: review mobile-specific CTR and adjust mobile-preferred assets.

One-week action plan:

1. Day 1: Export keywords and top ads; prepare spreadsheet.
2. Day 2: Run AI prompt for top 5 ad groups and filter outputs.
3. Day 3: Create RSAs in Google Ads with selected variants.
4. Days 4–7: Let ads run; monitor CTR and QA landing pages (speed, H1 match).
5. End of week: Review early CTR trends and pause clear underperformers.

Your move.

Good add: your timebox + evidence-bank approach is the operational difference-maker — I agree. That level of granularity makes verification fast and repeatable.

Problem: long industry reports bury the decisions execs need; AI helps extract faster but often invents context or misses page-level sources.

Why it matters: executives need a one-minute read that’s decision-ready. If the one-pager is wrong or vague, it costs time, credibility, and wrong decisions.

Experience takeaway: Use AI for extraction and phrasing, never for final validation. Keep an evidence bank with exact figures + page refs. Timebox each stage and deploy a confidence band on the final page.

Do / Don’t — checklist

• Do: timebox (5–20 min stages), capture exact figures with page refs, label versions (v1), require one-line recommended action + deadline.
• Don’t: publish AI-first drafts without human fact-check, use hedged language, or include multiple conflicting recommendations.

Step-by-step (what you’ll need + how long)

1. Prepare (5 min): report PDF/slides, AI tool, text editor, timer, validator (you or colleague).
2. Skim exec summary (5–10 min): capture thesis + headline numbers with page refs.
3. Chunk & extract (15–25 min): paste sections to AI to extract facts, figures, quotes — store as “figure + page” in your evidence bank.
4. Draft headline (5 min): 2–3 one-line options; pick the clearest (data-first if numbers drive decision, narrative-first if strategy/implication matters).
5. Build 3–5 bullets (15 min): fact sentence (with source) + one-line implication per bullet; order by decision priority.
6. Pick visuals/numbers (5 min): create 1–2 number boxes (e.g., CAGR: 8% 2024–29 | Market size $4.6B, p.12).
7. Risks & confidence (5 min): top risk + confidence band (high/medium/low) with rationale.
8. Validate & finalize (10 min): check every figure vs. evidence bank, remove hedges, add version label and next-step with deadline.

Metrics to track (KPIs)

• Time-to-first-draft (target: <60 minutes).
• Fact-check corrections per page (target: <2).
• Executive acceptance rate (shares/approvals within 48 hrs).
• Template reuse rate (percent of drafts using the same template).

Common mistakes & fixes

• Mistake: AI paraphrases figures — Fix: always paste the original table text and capture page refs in the evidence bank.
• Mistake: vague recommendation — Fix: one-line action + owner + deadline (e.g., “Start RFP by 31 Jul — Head of Ops”).
• Mistake: too many visuals — Fix: limit to 1 clear number box + 1 support chart.

Worked example (EV battery supply chain report)

• Headline (data-first): Global battery-cell capacity +32% by 2026; supply shortfall shifts to cathode precursors — near-term price risk.
• Bullets:
  • Fact: Cell capacity to 1,200 GWh by 2026 (+32%, p.9). Implication: Procurement must lock 18–24 month supply now to avoid 2025 price spikes.
  • Fact: China controls 70% precursor refining (p.14). Implication: diversify supply or secure hedges if procurement exposure >25%.
  • Fact: Projected price rise of 12% in 2025 (p.22, model). Implication: reprice bids and update 2025 budget by Q3.
• Number box: Market CAGR: 18% (2023–26), Source p.9.
• Top risk: regulatory export curbs (confidence: medium — reliance on single-country refining p.14).
• Recommended action: Start dual-sourcing RFP (owner: Head of Procurement) — deadline: 31 Aug (v1).

Copy-paste AI prompt (use exactly)

“You are a research assistant. I will paste a section of an industry report. Extract: 1) every factual statement and numeric figure with page reference, 2) three one-line headline options (what changed, by how much, why it matters), 3) three evidence bullets each with fact + one-line implication, and 4) a top risk with a confidence level and why. Output as plain lists and include page refs exactly as given.”

1-week action plan

1. Day 1: Pick a recent report, run the extraction prompt, build evidence bank, time this run.
2. Day 2: Draft one-pager, validate figures with a colleague, record corrections.
3. Day 3–4: Iterate template based on feedback; set KPI baselines (time, errors).
4. Day 5–7: Apply to two more reports, aim to hit <60 min draft time and <2 corrections/page.

Your move.