BG Dashboard
Documentation

AI Prompt

Copied out of the code at build time, so it is what actually runs.

You are a senior ecommerce operator who has launched beauty and FMCG products
for twelve years. You are reading a product hypothesis at the very start of a
validation process.

WHAT YOU HAVE: the form below, and nothing else. No market file has been uploaded
yet — that is step 2. You therefore have NO figures for this category, this
product or its competitors.

WHAT THIS MEANS FOR YOU:
· Do not state a market size, a growth rate, a competitor's price, a search volume
  or any other number as if you knew it. You do not.
· Where a claim would need such a number, write it as something to verify in step 2
  instead of asserting it.
· Your judgement of the hypothesis is welcome and is what you are here for. Invented
  evidence for that judgement is not.

PRODUCT HYPOTHESIS
- Name: {name}
- Type: {type} · Market: {market} · Category: {category}
- Where the idea came from: {source}
- Target price: ${price_from}–${price_to}
- Description as written by the team:
{description}

SCORING — 100 points, two of them already computed:
· Completeness of the form: {completeness}/35 — computed, use as given.
· Price-range viability: {price_score}/10 — computed from the width of the range, use as given.

The remaining 55 are yours, and each has a test. Award the points only when the
test is met by the text above; award 0 when it is not. Half marks only where the
scale says so.

· Consumer need — 15 points. Does the description name a specific problem and who
  has it? 15 if both are there and the problem is one a person would recognise in
  themselves. 8 if a problem is named but vaguely, or the sufferer is "everyone".
  0 if the description is a list of features with no problem behind them.
· Uniqueness — 15 points. Is there something here a competitor could not put on
  their own listing tomorrow? 15 if the difference is in the product. 8 if it is
  in the positioning or the bundle. 0 if the difference is only adjectives.
· Go-to-market — 10 points. Does the text say where this sells, to whom, and why
  now? 10 for all three, 5 for one or two, 0 for none.
· Your own assessment — 15 points. Given twelve years of launches: how likely is
  this hypothesis to survive contact with real data? Say what would have to be
  true, and score how plausible that is. This is the one place your judgement
  stands alone, so make the reasoning visible.

OUTPUT — exactly this shape, nothing before or after it:

**SCORE: [X]/100**
_Breakdown: Completeness {completeness}/35 · Need [X]/15 · Uniqueness [X]/15 · Strategy [X]/10 · Price {price_score}/10 · Expert [X]/15_

**Hypothesis Assessment**
2–3 sentences. Lead with the single thing that most determines whether this works.

**What to refine**
Exactly three points, no more. Each names a weakness in the hypothesis as it is
written — not a risk of the market — and each starts with a short bold label of
two or three words, then one or two sentences: what is thin, and what would
settle it. Where a later step settles it, say which (step 2 for market evidence,
step 3 for the survey, step 4 for the economics). Raise nothing the description
above does not already put in doubt.

**What to pull in step 2** — be specific enough that somebody can act on it without asking you:
- Keepa Product Finder: category, price band to filter on (derived from the target
  price above), minimum monthly units worth looking at, and why those bounds.
- Helium10: 3–5 seed keywords a buyer of this product would actually type.
- Google Trends: the search terms and the timeframe, and what shape would support
  or kill this hypothesis.
- TikTok Shop: the hashtags and the category to look at.
- Reviews: which competitor products to pull reviews for, and the complaint you
  most expect to find.

**DECISION: GO to market validation** or **Needs refinement**
One sentence saying which, and what would change it.

Data sources

Rule: missing source → score = 0 for that criterion. The system never guesses.

How to export Keepa: keepa.com → Product Finder → set filters → Export CSV
How to export BA: Seller Central → Brand Analytics → Search Frequency Rank → export

Scoring criteria (6 dimensions)

Market Size & Growth  ×20
Competition           ×18
Search Demand         ×17
TikTok Signal         ×15
Product Requirements  ×15
Entry Barrier         ×15

JSON Schema

{
  "verdict": "GO" | "NO-GO" | "CONDITIONAL GO",
  "total_score": 0-100,
  "scores": {
    "market/comp/search/tiktok/product/barrier": {
      "score": 0-100, "note": "1 line", "key_fact": "specific number"
    }
  },
  "requirements": [
    {"icon":"⭐","metric":"Min. Rating","value":"≥4.3★","rationale":"..."},
    {"icon":"💬","metric":"Min. Reviews","value":"≥500","rationale":"..."},
    {"icon":"📦","metric":"Sales Target","value":"≥800/mo","rationale":"..."},
    {"icon":"💰","metric":"Price Window","value":"$14–$20","rationale":"..."},
    {"icon":"📈","metric":"Growth Rate","value":"≥8%/mo","rationale":"..."},
    {"icon":"🏆","metric":"BSR Target","value":"Top 500","rationale":"..."}
  ],
  "success_probability_min": 0-100,
  "success_probability_max": 0-100,
  "prob_title": "phrase",
  "swot": {
    "strengths/weaknesses/opportunities/threats": ["point with data", "..."]
  },
  "consumer_voice": ["buyer insight", "..."],
  "next_steps": [{"num":"01","title":"...","desc":"..."}],
  "review_insights": null | {
    "top_positives": [...], "top_pains": [...],
    "claim_suggestions": ["in buyer language"]
  }
}

Probability range rule: spread 10–20 pts. Score ≥65 → min ≥50. Score <40 → max ≤45.

How to optimize

• Change criteria weights: find const CRITERIA=[...] in code, edit w: values (sum must = 100)
• Add a criterion: add entry to CRITERIA array and corresponding field to JSON schema
• Stricter SWOT: add "Every SWOT point must cite a specific number or source" to prompt
• Reduce timeout risk: lower trim limits (currently Keepa 30 rows, Keywords 25, BA 20)

Van Westendorp

4 open-number questions → compute medians:
  Optimal Price = (Bargain + Getting Expensive) / 2
  Acceptable Range = Bargain .. Getting Expensive

Example (highlighter survey, n=55):
  Too cheap $5 · Bargain $15 · Getting expensive $20 · Too expensive $30
  → Optimal: $17.50  ·  Acceptable range: $15–$20

Price Elasticity chart

For each price point ($8, $10, $12, $14, $16, $18, $20, $22, $25, $30):
  % acceptable = respondents where (too_expensive >= price) / total

Rendered as SVG bar chart with:
  - Green shaded zone = VW acceptable range
  - 80% threshold dashed line
  - Your price highlighted in blue

Score formula

base = pi>=60 ? 80 : pi>=40 ? 55 : pi>=20 ? 30 : 20
bonus = price in VW optimal zone ? +10 : 0
final = min(100, base + bonus)

Purchase Intent = (answers 4 or 5 on 1–5 scale) / total × 100%
PASS threshold: PI >= 60%

AI "Extract Insights" (optional button)

Analyzes open-text responses ("What made you choose this design?")
Returns JSON: { "likes": [...], "dislikes": [...], "claim_ideas": [...] }
max_tokens: 2000

Output blocks

• Funnel — Total / Completed / Valid respondents
• Purchase Intent circular gauge — threshold at 60%
• Van Westendorp — 5 price anchors (too cheap → too expensive)
• Price Elasticity SVG curve with green optimal zone shading
• Product Insights — statement agreement bars + ✨ Extract Insights AI button
• Design Preference — bar chart by option, winner in gold
• Audience Profile — Gender, Age, Openness to new brands, Top purchase factors

How to optimize

• Change PI threshold: edit pi>=60 ? 80 in run5()
• Add NPS: parse column with "likely to recommend" keyword
• Richer claims: add "Suggest 3 Amazon bullet points" to Extract Insights prompt

Fee schedules (Q4 actual data)

AMAZON:
  FBA fee — Small $2.68 · Standard $3.09 · Large $3.16 /unit
  Referral — price <$10: 8%  /  price ≥$10: 15%  ← not flat!
  Storage — ~$0.012/unit

TIKTOK SHOP:
  FBT delivery — $4.03/unit (avg Q4)
  Referral fee — 3.9% of price
  Creator commission — avg 8.8%, range 0–16.5%

SHOPIFY:
  Delivery — Economy $3.10 · Standard $6.20 /order
  Shopify Payments — 4.5% of price

P&L formula

Example: Amazon, price $16.99, COGS $4.50, Standard, PPC $2.50, Discount $1.00

  FBA:     $3.09
  Referral: $16.99 × 15% = $2.55   ← price ≥$10
  Storage:  $0.012
  PPC:      $2.50  (manual)
  Discount: $1.00  (manual)

  Total cost: $4.50 + $3.09 + $2.55 + $0.01 + $2.50 + $1.00 = $13.65
  Profit:     $16.99 − $13.65 = $3.34
  Margin:     19.7%
  Monthly:    $3.34 × 2,000 units = $6,680

Verdict thresholds

margin ≥ 25%  → ✅ VIABLE
margin 10–24% → ⚠️ MARGINAL
margin < 10%  → ❌ NOT VIABLE + Hard Stop (cumulative capped at 35)

AI analysis prompt (optional)

Returns JSON with: verdict, verdict_reason, optimal_price, price_floor,
price_ceiling, price_position, cost_wins (3 margin improvements),
volume_needed, timeline_months, key_risk
max_tokens: 4000

How to optimize

• Update fees: find const FEE_SCHEDULES={...} in code
• Change VIABLE threshold: edit margin>=25 in calcUE()
• Change MARGINAL threshold: edit margin>=10 (also controls Hard Stop)
• Add CAC field: add "Customer Acquisition Cost" input to Amazon fields section

PPTX — 9 slides

How the download works: click "📥 Download Presentation (.pptx)" → data sent to parent file via postMessage → PptxGenJS generates file → browser downloads it directly. No Claude needed.

Adding product photo to Slide 1: open in PowerPoint → click the gray placeholder → insert your product image.

How to optimize

• Stricter GO: add "Only GO if cumulative ≥ 65" to prompt
• Add a slide: find exportPPTX() in combined file, add var sl9=pres.addSlide()
• Change slide colors: edit DARK, LIGHT, ACC, GRN, RED, GOLD in exportPPTX()
• Logo on slides: s1.addImage({path:'url',x:0.3,y:5.2,w:1.5,h:0.3})