benchmark-research

Reverse-engineer how selected companies in [INDUSTRY] actually grew, sold, priced, implemented, expanded, and defended their position — then extract reusable patterns. NOT generic company summaries.

Works best for questions like:

• How did the fastest-growing companies in this industry actually scale?
• What sales motion / pricing model / initial wedge worked?
• What implementation / deployment model made customers successful?
• What expansion mechanics created retention and NRR?
• What recurring laws or archetypes appear across the best companies?

Entry-point flow (on trigger)

When the trigger phrase fires, run a short intake before any research:

1. Industry / scope. What industry? Why this one? What counts as "fast growth"? What benchmark questions matter? (Phase 0)
2. Mode. Full programme (15–25 companies, weeks of work) or MVV (8 companies, lighter — see below)?
3. Target company for the private application layer. Is there a "our company" perspective for private recommendations? If yes → public layer + private memo. If no → public benchmark only.
4. Project location. Where does research-project/ live? It is NOT inside agency engagement/ — it has its own artefact tree (see "Folder structure" below) and outlives any single session.

Capture these in research-project/README.md and proceed by phase.

Core principle

Strict separation into four layers — do not mix them:

1. Source harvest — structured primary-source archive
2. Company synthesis — one reverse-engineering memo per company
3. Meta-synthesis — cross-company laws, archetypes, patterns
4. Application — private recommendations (what to copy / adapt / reject / sequence)

Public benchmark can discuss lessons and boundary conditions. The private application layer says what THIS company should do. Keep them separate if the benchmark will be shared.

Folder structure

research-project/
├── README.md                      # scope, decisions, current phase
├── source-harvest-phase/
│   └── <company>/
│       ├── company.md
│       ├── TODO-sources.md
│       ├── people/<person>.md
│       └── sources/<source>.md
├── synthesis-phase/
│   ├── <company>/
│   │   ├── research-brief-<company>-playbook.md
│   │   ├── <company>-playbook-analysis.md
│   │   └── <company>-playbook-slides-outline.md
│   ├── meta-synthesis/
│   │   ├── meta-synthesis-industry-benchmark.md
│   │   ├── growth-laws.md
│   │   ├── sales-laws.md
│   │   ├── archetypes.md
│   │   ├── pattern-matrix.md
│   │   ├── company-comparison-table.md
│   │   └── cross-insights.md
│   └── application/
│       ├── application-memo.md
│       ├── priority-matrix.md
│       ├── do-copy-dont-copy.md
│       └── sequencing-roadmap.md
└── publishing-layer/              # optional

This is not an agency engagement/ directory — it lives outside the agency whitelist. A benchmark programme spans many sessions; each session advances one or more phases.

Evidence labels (use everywhere)

If a claim matters, label it. Do not let elegant writing blur evidence quality. This is the same discipline reality-checker / skeptic enforce on agency artefacts.

Source priority

Prefer primary / operator sources. Order:

1. Founder / CEO / executive interviews
2. Podcasts, talks, conference appearances
3. Company-written material — product docs, launch posts, customer stories, trust docs, pricing pages
4. Operator posts from people who built the motion
5. Job descriptions, org charts
6. Customer case studies / quotes
7. Partner pages / ecosystem announcements
8. Investor memos / analyst reports (clearly labelled)
9. News articles only if they contain original facts, quotes, or data

Avoid overweighting: recycled funding news · SEO summaries · unsourced listicles · vague market maps · press with no operator detail.

Company selection

Score each candidate on:

• Growth velocity — scaled unusually fast?
• Relevance — in or near the target industry?
• Evidence availability — enough sources to reverse-engineer the motion?
• Distinctiveness — adds a unique mechanism, not just another copy?
• Transferability — others can learn from it, even if not copy directly?

Tiers:

• Tier 1 — Core benchmark. Directly relevant, strong evidence, clear motion.
• Tier 2 — Boundary case. Not perfectly comparable but teaches a mechanism.
• Tier 3 — Context only. For contrast, not a core example.

Do not include a company just because it is famous. Include it because it teaches a mechanism.

Phase 1 — Source harvest

Goal: structured evidence archive, not a giant narrative report.

Per company, populate source-harvest-phase/<company>/:

• company.md answers: what the company does · initial wedge · why it matters for this benchmark · known milestones · major open questions.
• TODO-sources.md lists: missing high-value interviews · gated pages worth revisiting · missing transcripts · ambiguous claims needing verification · important-but-unsourced claims.
• sources/<source>.md per source: title · source type · date · URL · who is speaking · why this source matters · key facts · direct quotes · what it reveals about growth/GTM/pricing/product/adoption · caveats.
• people/<person>.md per person: role · relevance to growth story · notable claims/quotes · source references · career pattern if relevant to GTM credibility.

Extract from each source: growth milestones · wedge & initial use case · buyer persona · sales motion · founder-led sales clues · pilot/POC structure · design partner details · pricing & packaging · contract size hints · implementation/deployment model · trust / compliance / security / integration details · customer success motion · expansion / retention mechanics · org buildout & key hires · partner / channel motion · moat & defensibility · non-transferable advantages.

Done criteria: company.md + TODO-sources.md exist · multiple strong source notes exist · key people notes where relevant · major GTM / product / pricing / trust questions have at least partial evidence · remaining gaps explicitly listed.

The reusable harvest prompt: references/source-harvest-prompt.md.

Phase 2 — Company synthesis

Goal: strategy-grade reverse-engineering memo per company.

For each company, produce in synthesis-phase/<company>/:

• research-brief-<company>-playbook.md (use template: references/company-launch-template.md)
• <company>-playbook-analysis.md
• <company>-playbook-slides-outline.md

Memo structure (11 sections)

1. Executive summary — what is the company · why in benchmark · core growth interpretation · sourced vs inferred
2. Core motion — initial wedge · target buyer/user · why urgent · category framing
3. Growth system decomposition — acquisition channels · conversion mechanisms · trust mechanisms · implementation/activation path · expansion loop
4. Unit economics & commercial logic — pricing model · budget replaced/captured · ACV / contract size estimates · gross margin/COGS logic · payback · expansion logic
5. Sales cycle reverse engineering — who bought · who used · who blocked · who championed · pilot/POC structure · proof points · procurement/security path · founder-led vs sales-led vs partner-led
6. Implementation / deployment model — onboarding sequence · integration requirements · services / forward-deployed / solutions role · customer success model · time to value · scale-up path
7. Why the company won — timing · wedge strength · product advantage · distribution advantage · credibility/trust advantage · ecosystem/partner advantage
8. Benchmark relevance & transferability boundaries — what this case demonstrates · what is broadly transferable · what should not be copied blindly · what depends on unique market structure
9. Factor analysis — McKinsey-style table: factor · evidence · strength · transferability · caveats
10. Risks & fragilities — what could break the model · competitive threats · margin/pricing risks · channel risks · customer concentration · regulatory/trust risks
11. Final benchmark assessment — tier recommendation · why included/excluded · what future research should verify

Reusable synthesis prompt: references/synthesis-prompt.md.

Phase 3 — Cross-company meta-synthesis

After enough company memos exist (recommended: ≥6), create:

• meta-synthesis-industry-benchmark.md
• growth-laws.md (or .yaml)
• sales-laws.md
• archetypes.md
• pattern-matrix.md
• company-comparison-table.md
• cross-insights.md

Answer: what patterns repeated · what was rare vs common · what commercial logic appeared repeatedly · what sales tactics repeated · what archetypes emerged · what mechanisms were true exceptions · what changed by subcategory / buyer / market maturity.

Growth laws

Structural patterns in how companies scaled. Per law: id · name · short_description · body · prevalence · key_examples · anti_pattern · caveats.

Categories to consider: wedge clarity · prestige-first adoption · domain-expert GTM · proof before scale · labor-budget or revenue-budget pricing · expansion flywheel · high-touch implementation as moat · trust architecture · platform expansion after wedge · ICP discovery filter · auditability as GTM mechanism.

Sales laws

Patterns in commercial-motion execution. Same structure as growth laws.

Categories to consider: founders sell first · demo against the customer's own work/data · willingness-to-pay as qualification · paid pilot structure · objector-to-champion conversion · practitioner pull · reference hierarchy · multi-year contracts as switching cost · economic buyer pre-close.

Pattern matrix

Company × pattern table. Cells: confirmed / partial / absent / unknown — do not force binary labels where evidence is mixed. Use partial and unknown liberally.

Example fields: company · tier · wedge_clarity · prestige_first · domain_expert_gtm · proof_before_scale · labor_budget_pricing · expansion_flywheel · founder_sells_first · demo_against_own_work · paid_pilot · practitioner_pull · trust_architecture · implementation_moat · evidence_quality (high/medium/low) · notes.

Archetypes

Clusters of companies with similar growth/commercial logic. Per archetype: name · definition · member companies · common wedge · buyer · pricing model · sales motion · implementation model · expansion mechanism · risks · what this archetype teaches.

Do not create a new archetype for every one-off exception. If only one company fits, mark it as a boundary case unless the pattern is strategically important enough to preserve as a future hook.

Cross-insights

Important patterns that may not be universal laws. Per insight: title · summary · evidence · companies involved · confidence level · implications · caveats.

Good cross-insights sound like:

• "Consumption-metered expansion may be a third NRR mechanism."
• "Prestige logos work differently in developer tools than in executive-procured software."
• "Compliance is not a back-office requirement; it is part of GTM."
• "Services-heavy deployment can be a moat when it creates workflow depth."

Phase 4 — Application layer (private)

ONLY after company synthesis and meta-synthesis are complete.

Answers: what should WE copy? · what adapt? · what NOT copy? · what sequence first? · what preconditions must be true before copying? · what would be dangerous cargo-culting?

Deliverables:

• application-memo.md
• priority-matrix.md (Pattern · Source companies · Why it worked there · Applicability · Preconditions · Risk if copied blindly · Recommended action)
• do-copy-dont-copy.md
• sequencing-roadmap.md

Keep application recommendations strictly separate from the public benchmark if the benchmark will be shared. Public pages must not leak private company-specific advice.

Quality gates

Before calling a company memo complete:

• Does it explain the growth machine, not just describe the company?
• Are evidence and inference clearly separated?
• Are pricing and economics addressed?
• Is sales cycle reverse-engineered?
• Is implementation / deployment model addressed?
• Is expansion logic explicit?
• Are non-transferable advantages identified?
• Are open questions listed?

Before calling meta-synthesis complete:

• Are laws based on multiple companies, not one anecdote?
• Are exceptions labelled as exceptions?
• Are boundary cases separated from core patterns?
• Are prevalence estimates honest?
• Do archetypes group mechanisms, not vibes?
• Does the pattern matrix reflect evidence quality?

Before publishing:

• Are public and private layers separated?
• Are source references preserved?
• Are unverified estimates labelled?
• Are generated pages checked for private leakage?
• Are stale claims removed after refactors?

Anti-patterns

1. Generic company summaries. Bad: "Company X is a leading platform that helps teams…" Good: "Company X won by starting with workflow Y, proving metric Z in pilot, then expanding via mechanism A."
2. Funding-news overweighting. Funding is not a growth machine. Only use funding news if it includes original metrics, customer facts, or operator quotes.
3. Unlabeled inference. If inferred, say it is inference.
4. Overfitting one company into a law. One company creates an emerging insight, not a universal law.
5. Confusing public pricing with actual economics. Public tiers may be entry points; real revenue may come from usage, enterprise contracts, services, volume, or expansion.
6. Mixing product adoption with revenue mechanism. "Users like it" ≠ "this is how the business scales."
7. Treating implementation as boring. In many enterprise categories, implementation IS the moat.
8. Cargo-culting prestige customers. Prestige-first works when peer trust matters; not in all markets.
9. Publishing private recommendations. Public = benchmark lessons; private = company-specific advice.
10. Not updating cross-synthesis after new company research. Every strong new memo should trigger a review of laws, matrix, archetypes, cross-insights.

Suggested sequence

Minimum Viable Version

Lightweight starter:

1. Pick 8 companies.
2. For each, collect 5–10 strong sources.
3. Write one memo per company using the standard structure.
4. Build one pattern matrix.
5. Extract 5 growth laws + 5 sales laws.
6. Write a 5-page application memo.

Enough to be useful. Full version expands to 15–25 companies and a publishable site.

Agent activation prompt

Use as the system / task instruction for an agent doing the research:

You are building a benchmark research corpus for [INDUSTRY]. Your job is to reverse-engineer how selected companies actually scaled. Do not write generic company summaries. Always separate evidence from inference. Prefer primary/operator sources. Maintain the project folder structure. For each company, first build a source archive, then a synthesis memo. After multiple company memos exist, extract cross-company laws, archetypes, and comparison matrices. Keep private application recommendations separate from public benchmark research. When uncertain, write TODO / open question instead of inventing certainty.

Per-phase reusable prompts live in references/:

• references/source-harvest-prompt.md
• references/synthesis-prompt.md
• references/company-launch-template.md