Methodology
How we classify, grade and score
01Principles
- 01
Evidence before confidence.
Every published figure is an observation tied to a source, a date and a grade.
- 02
Unknown is a legitimate value.
Missing data is shown as Unknown. It is never converted to zero or estimated silently.
- 03
Grades attach to metrics, not companies.
A company can have a verified valuation and a self-reported headcount.
- 04
Zero employees is an optimisation constraint, not dogma.
Some work should stay human. We measure the design, not ideology.
- 05
Organisation-level analysis.
We analyse workflows and operating models, never individual employees.
- 06
Payment never influences rank.
No company can pay for inclusion, classification, score or position.
02Classification
Each company receives exactly one class, based on its employee position and how recurring work is executed. Only ZE-Strict and ZE-Operational companies are zero-employee companies. Ultra-Lean is a reference class.
Founders only. No employees and no recurring human workforce.
- Founders only; employee count is zero.
- No recurring human workforce (no standing contractor team).
- Occasional legally or professionally necessary services (accountant, lawyer, auditor) are permitted and disclosed separately.
No employees. Limited specialists may exist; software and AI do most recurring execution.
- Employee count is zero.
- Limited contractors or professional specialists may exist.
- Software and AI perform the majority of recurring operational execution.
A small human team with exceptional AI and software leverage.
- Has employees, but very few relative to revenue or valuation.
- Tracked as a reference class: not zero-employee, but informative about leverage.
An interesting candidate with insufficient evidence, or one that does not currently qualify.
- Evidence is insufficient to classify, or the company does not meet another class.
- Below the materiality threshold: less than $100k supported annualised revenue and no disclosed outside funding. Zero employees is trivially easy before a business has customers.
- Re-reviewed when new evidence appears.
03Evidence grades
Grades attach to individual metric observations. A derived value (for example revenue per human) inherits the weakest grade of its inputs. Click any grade badge on the site to see its source, source type, dates, URL and notes.
| Grade | Meaning | Typical sources |
|---|---|---|
| AExternally verified | Audited, regulatory or transaction-level evidence. | Regulatory filings, audited accounts, public-company disclosures, an acquirer's official announcement stating the price. |
| BStrongly corroborated | Several credible independent sources, or third-party verified data. | Two or more outlets with their own reporting; payment-processor-verified dashboards run by an independent operator; reporting based on reviewed documents. |
| CFirst-party | Founder or company statements, including dashboards they control. | Founder posts, company websites, interviews in which the founder supplies the figure. |
| DWeak / unverified | A secondary claim without traceable primary support. | Aggregators, listicles, blogs repeating figures without attribution. |
| ?Unknown | No usable evidence. The value is not published. | Missing data stays missing. It is never converted to zero. |
Source independence
A news article that repeats a founder's own figure is first-party evidence (C), not corroboration. A payment-processor-verified dashboard counts as independent only when its operator is independent of the company; a founder's figures on a dashboard the founder owns are first-party. Grade B requires at least one independent source; the build fails if a B-graded observation has none, and if an A-graded observation lacks a filing, audit or transaction source.
- Regulatory filingIndependent
- Audited financialsIndependent
- Transaction announcementIndependent
- Independent pressIndependent
- Press interviewFirst-party
- Founder statementFirst-party
- Company websiteFirst-party
- Social postFirst-party
- Public dashboardFirst-party
- Verified dashboardIndependent
- Podcast / videoFirst-party
- DatabaseIndependent
- Secondary blogFirst-party
04Headline values
- Latest supported value. Cards and tables show the most recent observation graded A–C. Grade D claims appear only in a dossier's evidence table.
- Revenue. We show annual revenue, run rate or MRR exactly as stated. Comparisons annualise MRR as MRR × 12 and label it as derived.
- Valuation. For private companies we report the latest supported private valuation (priced round, reported round, or acquisition price) with its type. We never use “market cap” for private companies and never infer a valuation from revenue.
- Employees exclude founders. Contractors are tracked separately. A claim of zero employees is labelled “claimed” unless independently corroborated.
- Humans = founders + employees. Revenue per human uses this denominator; contractor dependence is scored separately.
- Monthly revenue is not MRR. Monthly totals that include one-time sales are recorded separately from recurring revenue. Cumulative “since launch” totals and forward-looking projections are recorded but never used as a revenue headline.
- Time consistency. A derived value never combines observations describing periods more than 12 months apart. Revenue from one year is not divided by headcount from another; the result is shown as Unknown.
- Period, not publication date. “Latest” means the latest period described. A 2026 interview about 2023 revenue describes 2023.
- Post-acquisition figures describe a different organisation. They stay in the evidence table but are excluded from headline facts.
05Freshness
- A record not re-verified within 180 days of the dataset date is marked stale.
- A metric observation older than 365 days is shown in the stale state.
- Every dossier shows its last verification date and next scheduled review.
- Staleness is computed against the dataset release date, not the reader's clock, so pages are reproducible.
06Zero-Employee Score
The score (ZES-0.1) is a deterministic model, not an objective scientific law. Nine dimensions are each mapped to 0–100 by a published response curve and combined by fixed weights. No language model is involved in computing it.
| Dimension | Weight | Measured as | Response curve (input → points) |
|---|---|---|---|
| Recurring workflow automation | 20% | Recurring workflows automated | 0→0 · 100→100 |
| Human-headcount dependence | 15% | Employees excluding founders | 0→100 · 1→85 · 3→65 · 10→35 · 25→12 · 50→0 |
| Founder intervention | 15% | Founder hours/week on recurring execution | 0→100 · 5→90 · 10→75 · 20→50 · 40→15 · 60→0 |
| Agent decision autonomy | 15% | Highest production autonomy level | 0→0 · 4→100 |
| Contractor dependence | 10% | Recurring contractor FTE | 0→100 · 0.5→75 · 1→55 · 2→30 · 5→0 |
| Revenue per human | 10% | Annual revenue per core human (founders + employees) | $50k→0 · $100k→15 · $250k→35 · $500k→50 · $1000k→70 · $5000k→90 · $10000k→100 (log scale) |
| AI & software economics | 5% | AI, software & infrastructure cost as % of revenue | 0→100 · 10→100 · 20→80 · 35→50 · 50→20 · 70→0 |
| Approval burden | 5% | Human approvals per week | 0→100 · 10→90 · 25→70 · 50→45 · 100→15 · 200→0 |
| Operational resilience | 5% | Two-week founder absence | 0→0 · 1→35 · 2→70 · 3→100 |
Unknown inputs
Dimensions with unknown inputs are excluded, not scored as zero. We publish a point score only when at least 60% of the model weight is observable; the point score is the weighted mean over known dimensions. We always publish the evidence-supported range: the lowest and highest score consistent with what is known.
Confidence
Confidence is certainty-weighted coverage. Measured calculator inputs count fully and estimates count 0.6. For companies, inputs inherit their evidence grade: A 1.0, B 0.85, C 0.6, D 0.3. ≥ 0.75 is High, ≥ 0.5 Medium, below that Low.
Inputs
- Founders
- How many founders actively operate the company? People with ownership who work in the business. Passive investors are excluded. [people]
- Employees
- How many employees are there, excluding founders? People on payroll (full- or part-time), counted per person, not FTE. Contractors go below. [people]
- Recurring contractors
- How much recurring contractor time do you use, in full-time equivalents? Recurring freelancers and agencies doing ongoing execution, as FTE (40 h/week = 1.0). Exclude one-off professional services like an annual audit. [FTE]
- Annual revenue
- What is your annual revenue or current annualised run rate? Trailing-twelve-month revenue, or current MRR × 12. In USD. [USD / year]
- Recurring workflow automation
- What share of recurring workflows run end-to-end without a human executing steps? List workflows that recur at least monthly (billing, onboarding, support triage, reporting, deploys…). Count those executed by software or agents, with humans only approving or handling exceptions. [% of recurring workflows]
- Founder operational hours
- How many founder hours per week go to recurring operational execution? Hours spent doing recurring work by hand: answering tickets, running reports, manual deploys, invoicing. Exclude strategy, product decisions, sales conversations and approvals. [hours / week (all founders)]
- Agent decision autonomy
- What is the highest autonomy level your AI agents operate at in production? Count only agents running in recurring production workflows, not experiments. [level 0–4]
- AI & software cost
- What do AI, software and infrastructure cost as a share of revenue? Monthly model/API spend, SaaS subscriptions and hosting, divided by monthly revenue. Excludes payment processing fees. [% of revenue]
- Approval burden
- How many times per week must a human approve an automated action before it proceeds? Approval interruptions (refunds, outbound emails, deploys, payouts). Approval of high-risk actions is healthy; the score measures load, not existence. [approvals / week]
- Operational resilience
- What happens if every founder is unreachable for two weeks? Assume no emergencies beyond normal variance. Judge customer-facing quality, not just uptime. [level 0–3]
07Race to $1B rules
Verified definition. A company is a verified $1 billion zero-employee company when it is classified ZE-Strict or ZE-Operational, its zero-employee position is graded A or B, and a valuation or acquisition price of at least $1 billion is graded A or B.
- Only ZE-Strict and ZE-Operational companies are ranked.
- Ranking uses the latest supported (A–C) private valuation or acquisition price, highest first.
- Companies without a supported valuation are listed unranked, ordered by annualised revenue. We do not impute valuations for bootstrapped companies.
- Progress toward $1B = latest supported valuation ÷ $1B.
08Publication & submissions
A company is published only when sources graded A–C establish both its team position and at least one of revenue or valuation. Otherwise it stays in the research queue: its name may be listed, but no metrics are published.
Materiality. To be classified ZE-Strict or ZE-Operational, a company also needs at least $100k of supported annualised revenue or disclosed outside funding. Zero employees is trivially easy before a business has customers. Companies below the threshold go to the Watchlist. The build enforces this rule.
- Submitted: by the public or found in our own research.
- Research: primary sources collected, claims recorded as dated observations.
- Evidence graded: each observation is graded A–D.
- Classified: a class is assigned, with a written rationale.
- Decision: publish, watchlist, or reject.
Submissions are never published automatically. Submit a company or submit a correction.
09Independence
Rankings, classifications and scores are never paid for and never influenced by commercial relationships. Our paid products (the Zero-Employee Architect and the future Portfolio Control Plane) use the same public methodology. Corrections with evidence are applied whoever submits them.
10Limitations
- Most private-company figures are first-party (grade C). We label them but cannot audit them.
- Contractor use is rarely disclosed, so “no employees” can hide substantial human work.
- The dataset is small and selected for evidence quality, not randomly sampled. It does not estimate prevalence.
- Score weights reflect judgement. They are published so you can disagree with them precisely.
11Changelog
- 18 Oct 2026methodologyDownloads now withhold observations, timeline events and headline values that rest on a source whose terms forbid redistribution (TrustMRR). They stay on the site with attribution, labelled on each dossier. Export schema 1.1 adds withheld and withheldFacts.
- 16 Oct 2026releaseInitial public dataset (v0.1.0): 10 published dossiers and 6 candidates in the research queue. Evidence grades A–D are attached per metric; scoring model ZES-0.1.
- 16 Oct 2026methodologyPublished the classification rules (ZE-Strict, ZE-Operational, Ultra-Lean, Watchlist), the materiality threshold, and the verified definition of a $1B zero-employee company.
- 16 Oct 2026methodologyHeadline rules: monthly revenue is kept separate from MRR; cumulative totals and projections are never used as revenue; derived values may not combine periods more than 12 months apart.
- 16 Oct 2026addedPolsia enters the Race to $1B at rank 1 on a founder-stated $250M valuation (grade C). SEC Form D filings (grade A) show funding that differs from press reports.