Methodology
DataEI research is built on a transparent, repeatable, agent-augmented methodology. Every benchmark, briefing, and finding traces back to a documented source, a defined rubric, and analyst review. No black boxes.
Method is one half of the answer. The other half is what money cannot do to a finding — see Research Integrity for the sponsorship, subscription, and disclosure policy that sits behind everything on this page.
This method powers the DataEI Diamond — a rotatable, tier-free map of how product families fit against the tradeoffs buyers actually weigh.
The DataEI Research Method
Five phases, from raw data collection to published benchmark. AI agents handle scale; analysts handle judgment.
AI agents continuously collect public vendor data, financial signals, product positioning, and market activity across the $2M–$50M SMB segment. The MCP server orchestrates collection across hundreds of sources. Output: raw research dataset, continuously updated.
Vendors submit to the Research Briefing Program. Analysts run structured briefings to gather inside perspective, validate positioning, and capture market context that public data misses. Output: briefing notes, validated vendor profiles.
Agents cross-reference collected data against briefing notes, public filings, and historical benchmarks. Conflicts are flagged. The MCP server runs the Leverage Index calculator and segment analyzers. Output: draft synthesis, flagged conflicts.
DataEI analysts review every synthesis before publication. We apply 20+ years of technology-market experience to challenge findings, resolve conflicts, and pressure-test conclusions. Agents propose; analysts decide. Output: reviewed findings, editorial decisions.
Validated findings publish as Leverage Index benchmarks, research briefs, and briefing reports. Every published figure links to its source data, methodology step, and analyst reviewer. Output: published benchmark, source-traced.
The Leverage Index — Inputs and Weighting
The Leverage Index measures how effectively a company converts headcount into output. It is a 0–100 composite of four scored inputs plus one bounded analyst adjustment. Below is the full input list, what each one actually measures, where the data comes from, and what it is worth. You should be able to reconstruct our arithmetic and argue with it.
| Input | What we measure and how it's sourced | Weight |
|---|---|---|
| Revenue per employee | Trailing-twelve-month revenue divided by average full-time-equivalent headcount over the same period. Contractors and fractional staff count at 0.5 FTE where disclosed. Scored as a percentile rank inside the company's sector and size band, never as an absolute dollar figure. Sources — Company filings, company-disclosed financials, Research Briefing Program submissions, and analyst-verified estimates. Estimated figures are marked as estimates on the company record. | 35% |
| Automation & agent adoption depth | A 0–5 rubric applied separately to each of our six coverage areas, then averaged and rescaled to 0–100. A capability scores above 2 only if it is in production, used at least weekly, and displacing or avoiding a named role — not piloted, not licensed-and-idle. Sources — Briefing submissions, company disclosure, analyst interviews, and agent-collected public signals (job postings, stack detection, changelogs, release notes) with collection timestamps. | 25% |
| Output velocity | Delivery throughput against sector and size-band peers: shipped releases per quarter, median commit-to-production cycle time where disclosed, and delivery rate against publicly stated roadmap commitments. Percentile-ranked within band. Sources — Public changelogs, release notes, repository and package-registry signals, product announcements, and briefing submissions. | 25% |
| Margin conversion | Gross margin, plus operating or net margin where disclosed, adjusted for sector and stage. This input exists to catch leverage that does not reach the P&L: output per person that is bought with spend rather than earned with systems. Sources — Filings, company disclosure, and briefing submissions. Where margin is undisclosed, the input is scored from sector-and-stage norms and marked as estimated. | 15% |
| Leadership time allocation | An analyst adjustment, not a weighted input. Applied when interview evidence shows founder or leadership time is concentrated on building leverage rather than executing work the company has failed to systematize. Bounded at ±5 points on the composite, applied only after the four scored inputs are calculated, and always disclosed on the company record with a one-line stated reason. Sources — Analyst interview and briefing notes only. Never inferred from public data, never applied without a recorded reason. | ±5 pts |
| Composite | Weighted sum of the four scored inputs, then the bounded adjustment, then clamped to the 0–100 range. | 100% |
Under-leveraged
Output scales with headcount. Growth requires hiring.
Developing
Leverage exists in pockets but has not reached the operating model.
Leverage-first
Output grows materially faster than headcount, and it reaches the margin.
Sample, Normalization, and Revision
A weighting is only as honest as the sample it runs on. Here is what the current dataset contains, how a company is compared, and what happens when we get something wrong.
- 47 companies in the active Q3 2026 benchmark dataset; target of 100 by Q4 2026.
- Coverage window is the trailing twelve months, restated each quarter.
- A company enters the dataset through a briefing, an analyst-initiated profile, or client-commissioned benchmarking. Paying for research does not change eligibility or score.
- Sample size per band is published alongside every score. We say n.
- Every input is percentile-ranked within a sector × size band, not against the dataset as a whole. A 15-person infrastructure company is not scored against a 400-person services firm.
- A band must contain at least 8 companies before any score in it publishes. Thin bands are held, not extrapolated.
- Inputs estimated rather than disclosed are marked. More than two estimated inputs makes the score provisional, and it is labeled that way wherever it appears.
- No input is imputed from a vendor's own marketing claims.
Scores are restated quarterly. When a figure changes, we publish the correction with its date and the reason — we do not silently edit a published number.
Any company can contest its score. The route is a briefing, and the standard is evidence: a score moves when a sourceable input changes, and not otherwise. Disagreement, escalation, and commercial relationship are not inputs. If we made a mistake, we say so in the correction.
The weighting above is v1 and fixed for the Q3 2026 dataset. If we change a weight, we version the methodology, publish what changed and why, and restate prior scores under the new weighting so the series stays comparable.
Source Transparency
Every DataEI finding links to its source. We don't publish unsourced claims.
- Public vendor data (sites, filings, product pages)
- Research Briefing Program submissions (vendor-confirmed)
- Analyst interviews and primary research
- MCP-collected market signals (with timestamp)
- Pay-to-play rankings or paid placement
- Publish findings without analyst review
- Use agent output as final word — agents propose, analysts decide
- Hide conflicts — we flag them in the synthesis
Where the money sits
A published weighting only matters if nobody can pay to move it. Research Integrity documents what sponsorship and subscription do and do not buy, and how every commercial relationship is disclosed.