How Bluescore Works

Theoretical Foundations

Bluescore is designed on a foundation of established frameworks across statistics, strategy, and innovation research:

• Composite indexing methodology — informed by the Balanced Scorecard (Kaplan & Norton, 1992)
• Statistical normalization — z-score standardization (Altman, 1968) applied within each cohort
• Technology adoption modeling — informed by Diffusion of Innovations (Rogers, 1962)
• AI assessment principles — aligned with the NIST AI Risk Management Framework (2023)

These frameworks inform the design of Bluescore. They are not the methodology itself, but they anchor it in recognized academic and institutional traditions.

Cohort-Relative Scoring

Direct comparison between a three-person AI startup and Microsoft is not meaningful. A global ranking would reward scale and punish early-stage focus, because a seed-stage company cannot compete on revenue, headcount, or press volume against an incumbent. We therefore compute Bluescore within cohorts of true peers, so each score reflects relative strength against a comparable competitive set rather than against the entire 250-company universe.

Reading a Bluescore in context. A score of 89 for Anthropic means it sits in the 89th percentile among AI Native companies — not among all 250 companies in the index. This makes cross-cohort score comparisons intentional rather than accidental: a 92 in Big Tech and a 92 in Applied AI describe two different competitive realities.

The global AI250 leaderboard reflects these cohort-relative scores, surfacing the strongest momentum signals across all peer groups simultaneously rather than collapsing structurally different companies onto a single linear scale.

Confidence Interval

Every Bluescore is accompanied by a confidence interval — a 0–100 data-completeness score that tells you how much signal we have behind the number, not how certain the model is about the outcome.

The interval is derived from the fraction of the 46 raw signals we have populated for a given company. More disclosed data, richer public footprints, and broader press coverage produce a higher score. Sparse private companies score lower because fewer inputs are available.

• Not a grade. A lower confidence interval does not mean a company is worse. It means the evidence behind its Bluescore is thinner.
• No company reaches 100. Some signals are inherently unavailable — private revenue, undisclosed headcount, or unannounced partnerships — so the interval has a built-in ceiling.
• Use it as a transparency signal. Enterprise buyers should weigh the confidence interval alongside the Bluescore when deciding how much to trust a single score.

Data Sources & Inputs

Bluescore synthesizes signals from 200+ authorized data sources organized into seven categories:

• Industry research and analyst reports
• Company press releases and official communications
• Technical publications (arXiv, conference proceedings, peer-reviewed journals)
• Patent filings and intellectual property records
• Regulatory and financial disclosures (SEC filings, public funding announcements)
• News and trade press coverage
• Open technical signals (GitHub activity, technical documentation)

Sources include both publicly accessible feeds and authorized commercial data providers. Our research agents retrieve and synthesize relevant content from these sources daily, using a Retrieval-Augmented Generation (RAG) architecture that grounds each component score in cited evidence.

The Bluescore Formula

Bluescore is a weighted composite of five principal components, scaled to a 0–100 range and percentile-ranked within each cohort:

Bluescore_cohort = percentile_0_100( Σ wi × PCi )

where:
  PCi              = principal component score i (1 through 5)
  wi               = component weight (20% each, reviewed quarterly)
  percentile_0_100 = rank within the company's cohort, not the full AI250

Each company's score reflects its relative position within its cohort. Weights are reviewed quarterly by our Board of Advisors and adjusted only when supported by sustained predictive evidence.

The specific calibration of signals within each component is proprietary, reviewed under our quarterly methodology versioning process, and subject to change as new evidence emerges.

Algorithm Cards

How each principal component is measured

Cadence & Governance

• Daily refresh. Scores are recomputed daily as new signals arrive from our source feeds. Every company profile displays the date of its most recent Bluescore update.
• Quarterly methodology review. Our Board of Advisors reviews the methodology each quarter, including component weights, source inclusion, and any structural changes.
• Versioned methodology. Every change to the methodology is logged with a version number, change rationale, and effective date. Prior versions remain accessible for historical reference.
• Dynamic ranking. The AI250 itself is dynamic. Each quarter, companies may be promoted into the list or relegated to our Watchlist based on sustained signal movement.

Limitations

We disclose what Bluescore is, and what it is not:

• Bluescore is a relative score, not an absolute measure of value. A score of 75 reflects position within the AI250 cohort, not an objective rating in isolation.
• Private-company signals are sparser than public-company signals. Where disclosed data is limited, Bluescore relies more heavily on public technical and market signals, which may understate or overstate certain components.
• New companies require a stability window. Newly added companies typically take one full quarterly cycle before their Bluescore stabilizes against accumulated signal.
• Bluescore is not a recommendation. It is a structured signal designed to inform research, not replace independent due diligence.

Ask the Agent

Pro FeatureEvery Bluescore on this platform can be defended on demand.

Pro subscribers can click "Why this Bluescore?" on any company profile to receive an instant, source-cited explanation of the specific signals and component contributions driving that score. Each explanation cites the underlying evidence and links back to the source material.

This is how we replace the traditional analyst-call model: instant, source-cited, reproducible explanations available 24/7 for any company in the AI250.

Board of Advisors

Versioning

Current methodology version:

v3.0

Last updated:

May 29, 2026

Change history:

View changelog →