Purpose

This recipe generates a personalized, evidence-based dossier for retailers showing operational distress signals — inventory write-downs, supply chain disruption, workforce instability, or financial deterioration. The output is a 2-page document that names specific signals with dates, quantifies business impact using industry benchmarks, and offers 3 graduated action options — ready for delivery to the decision-maker identified during enrichment. [src1, src2]

Prerequisites

Enriched company profile available from signal-library/retail/enrichment-mapping/2026 — Retail Signal Enrichment Mapping
Signal detection report with scored signals available from signal-library/retail/detection-rules/2026 — Retail Detection Rules
Decision-maker contact (name, title, email/LinkedIn) from enrichment output
Claude or GPT API key — from Anthropic Console or OpenAI Platform
PDF generation tool — Puppeteer, WeasyPrint, or equivalent (free)

Constraints

Every claim must be falsifiable. Each data point requires a specific source, date, and verifiable figure. A recipient must be able to fact-check any claim within 5 minutes. [src1]
Diagnostic tone only. The dossier reads like a medical report: observation, interpretation, options. Never use phrases like "we can help you," "our solution," or "schedule a demo." [src2]
Maximum 2 pages plus optional appendix. Distressed retailer executives are time-constrained; dossiers exceeding 2 pages see 40-60% lower read-through rates. [src3]
All industry benchmarks must cite the study, year, and sample size. Unsourced benchmarks are indistinguishable from hallucination. [src4]
All personalization fields must be populated from enrichment data. Never deliver a dossier with placeholder text. Partially personalized outreach performs worse than generic outreach. [src5]

Tool Selection Decision

Which path?
├── Output format = PDF AND review mode = auto-deliver
│   └── PATH A: Automated PDF — Claude API + WeasyPrint + email delivery
├── Output format = PDF AND review mode = human review
│   └── PATH B: Reviewed PDF — Claude API + WeasyPrint + review queue
├── Output format = HTML email AND review mode = auto-deliver
│   └── PATH C: Automated HTML — Claude API + HTML template + email API
└── Output format = Markdown AND review mode = human review
    └── PATH D: Draft Markdown — Claude API + Markdown output + review queue

Path	Tools	Cost	Speed	Output Quality
A: Automated PDF	Claude API + WeasyPrint	$0.02-0.10/dossier	2-5 min	High — requires confidence threshold (score 8+)
B: Reviewed PDF	Claude API + WeasyPrint	$0.02-0.10/dossier	5-15 min	Excellent — human catches edge cases
C: Automated HTML	Claude API + HTML template	$0.02-0.08/dossier	1-3 min	High — inline rendering, no attachment friction
D: Draft Markdown	Claude API	$0.02-0.05/dossier	1-2 min	Good — fastest for review-heavy workflows

Execution Flow

Step 1: Assemble Input Data

Duration: 1-2 minutes per dossier · Tool: Python/Node.js script

Collect all required inputs from the enrichment and detection pipeline outputs into a single structured context object for the LLM. Required fields: company name, domain, revenue, decision-maker name and title, at least 2 signals with source, date, and data point.

Verify: All required fields populated — company name, at least 1 decision-maker, at least 2 signals with source+date+data_point. · If failed: If fewer than 2 signals, return to detection pipeline and lower threshold, or flag company for monitoring rather than outreach.

Step 2: Generate Executive Summary

Duration: 1-2 minutes · Tool: Claude/GPT API

Generate a 1-paragraph executive summary that names specific signals with dates and hooks the reader with quantified impact. Structure: Name 2-3 specific signals with dates → state what the pattern typically costs using industry benchmarks → close with a quantified impact statement.

Example output: “In Q3 2025, Acme Retail Corp disclosed $45M in inventory write-downs with days-inventory-outstanding rising from 95 to 132. In the following 30 days, 6 urgent supply chain roles appeared on LinkedIn and Indeed — 3x the normal hiring rate. Glassdoor supply chain team ratings dropped 0.8 points in Q4. Among the 127 retailers McKinsey tracked with this compound pattern, 12-18% margin erosion followed within 2 quarters.”

Verify: Summary contains at least 2 specific data points with dates. No promotional language detected. · If failed: Regenerate with stricter system prompt constraints if promotional language appears.

Step 3: Generate Evidence Pack

Duration: 2-3 minutes · Tool: Claude/GPT API

Generate one evidence entry per signal with 5 fields: source (exact document/page), date, data point (specific number), context (1-2 sentence conservative interpretation), and verification path (how to independently confirm).

Verify: Each signal entry has all 5 fields populated. Verification paths are actionable (e.g., “Search SEC EDGAR for CIK 0001234567, Form 10-Q, filed 2025-11-15”). · If failed: Remove any signal lacking a verification path — it was likely inferred rather than observed.

Step 4: Generate Impact Analysis

Duration: 1-2 minutes · Tool: Claude/GPT API

Quantify business impact using industry benchmarks with full citation. Show the math: “At $380M revenue, 12-18% margin erosion = $45.6M-$68.4M annual impact.” State time horizon explicitly and include confidence qualifier. [src1, src6]

Available benchmarks:

Inventory distress: 12-18% margin erosion within 2 quarters (McKinsey State of Fashion 2025, n=127) [src1]
Supply chain disruption: 3-5% revenue loss per quarter of sustained disruption (Bain 2025, n=89) [src6]
Workforce instability: 2x turnover cost in affected function within 6 months (MIT Sloan 2024, n=214) [src7]
DIO above 120 days: 15-22% higher carrying costs (IHL Group 2025, n=350) [src4]

Verify: At least 1 dollar-denominated impact figure derived from company revenue + benchmark range. Full citation present. · If failed: If revenue data missing, state benchmark as percentage with note.

Step 5: Generate Recommended Actions

Duration: 1 minute · Tool: Claude/GPT API

Generate 3 graduated options:

Option 1 (Low commitment): Free 30-minute diagnostic call — verbal walkthrough of signal pattern and peer comparisons. No obligation.
Option 2 (Medium commitment): Paid supply chain assessment with specific price range, duration, and deliverables.
Option 3 (No action): Do-nothing trajectory modeling — factual projection of what happens if patterns continue for 2 more quarters. Not fear-mongering, a benchmark-based projection.

Verify: All 3 options present. Option 1 is genuinely no-cost. Option 3 cites a benchmark. No urgency language detected. · If failed: Strip urgency language and regenerate Option 3 only.

Step 6: Assemble and Format Dossier

Duration: 1-3 minutes · Tool: PDF generation (WeasyPrint/Puppeteer) or HTML template

Assemble generated sections into the final dossier format: Executive Summary + Evidence Pack on page 1, Impact Analysis + Recommended Actions on page 2, optional appendix with methodology note and benchmark references.

Output files:

distress_dossier_{company_slug}_{timestamp}.pdf — Formatted 2-page dossier ready for delivery
distress_dossier_{company_slug}_{timestamp}.json — Structured dossier data for programmatic processing
dossier_qa_checklist_{company_slug}.json — Quality assurance results from verification checks

Verify: PDF renders at 2 pages or fewer. All personalization fields populated. All source citations present. · If failed: Reduce evidence pack to top 2 signals and regenerate.

Output Schema

{
  "output_type": "distress_dossier",
  "format": "PDF + JSON",
  "sections": [
    {"name": "executive_summary", "type": "string", "required": true},
    {"name": "evidence_pack", "type": "array", "required": true},
    {"name": "impact_analysis", "type": "string", "required": true},
    {"name": "recommended_actions", "type": "array", "required": true},
    {"name": "appendix", "type": "object", "required": false}
  ],
  "personalization_fields": [
    {"name": "company_name", "type": "string", "required": true},
    {"name": "decision_maker_name", "type": "string", "required": true},
    {"name": "decision_maker_title", "type": "string", "required": true},
    {"name": "signal_dates", "type": "array", "required": true},
    {"name": "signal_data_points", "type": "array", "required": true}
  ],
  "expected_page_count": "2 + optional appendix",
  "sort_order": "sections in fixed order: summary, evidence, impact, actions",
  "deduplication_key": "company_name + generated_at"
}

Quality Benchmarks

Quality Metric	Minimum Acceptable	Good	Excellent
Falsifiability rate	> 80% of claims have verifiable source	> 90%	100%
Personalization completeness	All required fields populated	+ industry-specific context	+ competitor references
Benchmark citation accuracy	1 cited benchmark with source	2-3 benchmarks with n-counts	All benchmarks with full citation
Tone compliance	No “we/our/schedule” detected	+ no superlatives or urgency	+ reads as independent analyst report
Page length	2 pages or fewer	1.5-2 pages	1.5 pages + focused appendix
Signal-to-noise ratio	2+ signals with evidence	3+ signals, all falsifiable	3+ signals + trend analysis

If below minimum: Regenerate the failing section with stricter system prompt constraints. If falsifiability rate is below 80%, audit each claim against the input data and remove any that lack a specific source.

Error Handling

Error	Likely Cause	Recovery Action
LLM generates promotional language	System prompt not restrictive enough	Add explicit negative examples: “NEVER write: 'we can help', 'our solution', 'schedule a demo'”
Impact analysis lacks dollar figures	Revenue data missing from enrichment	Use percentage-only impact with note; flag company for manual revenue lookup
Evidence pack has unverifiable claims	LLM hallucinated data points not in input	Cross-check every data point against input JSON; remove unverifiable claims
PDF exceeds 2 pages	Too many signals or verbose generation	Limit evidence pack to top 2 signals by composite score; reduce executive summary to 3 sentences
Personalization fields contain placeholders	Enrichment data incomplete	Do not generate dossier — return to enrichment pipeline to fill gaps
LLM refuses impact projections	Safety filter triggered by financial predictions	Reframe as “industry benchmark range” rather than “prediction”; cite study explicitly
Benchmark data is outdated	Source study older than 2 years	Flag as “based on {year} data” and note recalibration recommended

Cost Breakdown

Component	Per Dossier	50 Dossiers/Month	500 Dossiers/Month
LLM API (Claude Sonnet)	$0.02-0.05	$1-2.50	$10-25
LLM API (Claude Opus)	$0.05-0.10	$2.50-5	$25-50
PDF generation (WeasyPrint)	$0	$0	$0
Email delivery (SendGrid)	$0	$0 (up to 100/day)	$15/mo
Total (Sonnet + free tools)	$0.02-0.05	$1-2.50	$10-25

Anti-Patterns

Wrong: Leading with product pitch

Dossier opens with “Our firm specializes in retail supply chain optimization” or closes with “We'd love to schedule a demo.” Result: recipient classifies the dossier as spam and never reads the evidence. Diagnostic credibility is destroyed in the first sentence. [src2]

Correct: Lead with their data, not your capabilities

Open with specific signals the recipient will recognize (“Your Q3 10-Q disclosed $45M in inventory write-downs”). The evidence sells the conversation — not the sender.

Wrong: Including speculative or unfalsifiable claims

Dossier states “Your supply chain is likely underperforming” or “Most retailers in your position struggle with inventory.” These claims cannot be verified and sound like generic sales copy. [src1]

Correct: Every claim cites source, date, and data point

“Your DIO increased from 95 to 132 days between Q2 and Q3 2025 (SEC 10-Q, filed 2025-11-15, page 23).” The recipient can verify this in 2 minutes.

Wrong: Overloading with signals

Including 6-8 signals to appear thorough. Result: the dossier exceeds 2 pages, buries the key insight, and the executive stops reading after page 1. [src3]

Correct: Top 2-3 signals that form a compound pattern

Select the 2-3 signals that together tell a coherent story. Inventory write-down + urgent supply chain hiring + declining Glassdoor scores form a narrative. Adding unrelated signals dilutes the message.

Wrong: Generic benchmark without attribution

“Companies in your situation typically lose 15% of their margins.” No source, no sample size, no year. Indistinguishable from a hallucinated claim. [src4]

Correct: Full benchmark citation

“Among 127 retailers tracked by McKinsey (State of Fashion 2025), those showing compound inventory + workforce distress signals experienced 12-18% margin erosion within 2 quarters.”

When This Matters

Use this recipe when the signal detection pipeline has identified a retailer showing inventory, supply chain, workforce, or financial distress signals and the enrichment pipeline has resolved the signals to a specific company with decision-maker contacts. This is the asset generation step between enrichment (upstream) and outreach delivery (downstream). Without the dossier, outreach lacks the evidence-based positioning that differentiates signal-driven selling from cold outreach.

Distress Dossier Template