Methodology

How Bluescore Works

A transparent methodology for AI market intelligence.

The Bluescore ranks companies with sufficient public data in the Montclare AI250. It is a weighted average of 6 dimension scores, rescaled to 0–100 and ranked 1–250. There are no hidden factors and no machine-learning model deciding the rank: every score can be traced back to the signals behind it.

Methodology v4.0 · updated October 18, 2026

Framework

How Bluescore Is Built

1 · Data Inputs

News & Trade Press
Research Papers
Hugging Face Releases
Funding Announcements
Hiring & Job Posts
GitHub & Patents
Company & Product Pages

2 · Signals

Research agents record each signal; funding is checked against cited sources and news against the right company

3 · 6 Dimensions (0–10 each)

D1 Business Model & Revenue
21%
D2 Product Depth
21%
D3 Data & Model Moat
19%
D4 Talent & Research
16%
D5 Customer Adoption
13%
D6 Investment Momentum
10%

4 · Composite Bluescore

Bluescore (0–100)

Weighted average of the 6 dimensions, ranked 1–250 across all companies

Design Principles

  • Traceable. Anyone can follow a rank back to the dimension scores and signals behind it.
  • Stable. A rank moves when the company changes, not when a single day's reading wobbles.
  • Comparable. AI-native startups, public Big Tech and AI adopters are scored on one scale.

The 6 Dimensions

Each dimension is scored 0–10 from the signals below and carries a fixed weight:

DimensionWeight
D1 Business Model & Revenue
ARR signal, paying customers, pricing and enterprise tiers
21%
D2 Product Depth
model and product releases, open-source models, API integrations
21%
D3 Data & Model Moat
patents, research papers, model releases
19%
D4 Talent & Research
headcount, engineering share of roles, GitHub activity
16%
D5 Customer Adoption
news reach, G2 reviews and rating, web traffic, market share
13%
D6 Investment Momentum
total funding, last round, 6-month stock change (public companies)
10%

A seventh dimension, Responsible AI, is defined but not yet scored. Until a reliable source exists, its weight is shared across D1–D6 rather than filled with a placeholder.

How a Score Is Built

A company needs real values for at least 6 of 18 signals to be ranked; below that it's listed as Insufficient data instead of getting a score built from averages.

Acquisitions. Flag on announcement, remove on close. When a signed deal to acquire a listed company is announced, it stays ranked and is marked Pending acquisition; when the deal closes, it leaves the list on the next daily update.

  1. 1

    Collect

    Eight research agents and our data feeds record each company's raw signals: funding, hiring, technical output, news, partnerships, revenue, leadership and competitive position.

  2. 2

    Coalesce

    For each signal we take the latest known value within its lookback window: 14 days for most signals, 21 for the weekly funding fields, and 45 for total funding.

  3. 3

    Smooth

    Signals estimated by AI are smoothed with a trimmed mean (drop the highest and lowest values, average the rest) over a window holding about seven readings: 7, 14, 21 or 28 days depending on how often the signal is refreshed.

  4. 4

    Attribute

    Whole-company figures such as revenue, patents and papers are attributed to the AI share: AI Native 100%, Applied AI 60%, AI Adopter 25%, Big Tech 20%.

  5. 5

    Normalize

    Each signal is z-scored across all active companies (log-scaled for counts), mapped to 0–10 with the normal CDF, and ranked globally. Missing inputs take the population median.

  6. 6

    Weight & rank

    The weighted sum of the 6 dimensions becomes the Bluescore (0–100), and companies are ranked 1–250. Rank change is shown over 7 days (7d).

The Bluescore Formula

Bluescore = 10 × Σ wi × Di

where:
  Di = dimension score i, 0–10 (D1 through D6)
  wi = dimension weight (21%, 21%, 19%, 16%, 13%, 10%; sums to 100%)

Rank = position by Bluescore across all 250 companies

Each signal's 0–10 score is its z-score across all companies, passed through the normal distribution, so a company at the population average scores 5. Weights change only through a versioned methodology update (see Versioning below).

Cohorts

Every company sits in one of four cohorts. Cohorts give context: the Rankings page shows each company's rank within its cohort next to its overall rank. Both come from the same Bluescore, so a 70 means the same thing in every cohort.

AI Native

Pure-play AI companies founded to build AI products and infrastructure.

Big Tech

Large incumbent technology firms with significant AI divisions. Their whole-company figures are discounted to the AI share.

Applied AI

Companies applying AI to physical or specialized domains: robotics, defense, biotech and industrial automation.

AI Adopter

Established software and enterprise firms building AI into existing products. Their whole-company figures are discounted to the AI share.

Model Releases

The release count used in D2 and D3 is built from named releases, not a single estimate. Each major model or product release is recorded in a release ledger, and the count is every recorded release from the last 12 months. A release seen only once drops out after 7 days. For open-weight labs, model families published on Hugging Face set a floor: we use the larger of the two counts, never the sum.

Cadence & Governance

  • Daily scores. Bluescores and ranks are recomputed every night. Signals refresh on a schedule matched to how fast they change: news daily, funding weekly, and the rest every two or three days.
  • Checked inputs. Funding totals are written only when a cited source names the company, and each news match is confirmed before it counts.
  • Reviewed changes. Weights and formulas change only through a reviewed update, tested by replaying recent nights to measure the effect on ranks before it ships.
  • Versioned methodology. Each methodology release carries a version number and an effective date.

Theoretical Foundations

Bluescore draws on established frameworks across statistics, strategy and innovation research:

  • Composite indexing methodology — informed by the Balanced Scorecard (Kaplan & Norton, 1992)
  • Statistical normalization — z-score standardization applied across all 250 companies
  • 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.

Limitations

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

  • Many signals are AI estimates. Several signals are estimated by research agents rather than read from live data. Smoothing reduces day-to-day noise, but not systematic bias.
  • Insufficient public data. Companies with fewer than 6 of 18 real signal values stay listed as Insufficient data until more information is available.
  • Real changes take time to show. Smoothing means a genuine change can take one to two weeks to show fully in rank.
  • Bluescore is relative. A score reflects a company's position among the AI250, not an absolute measure of value.
  • Bluescore is not a recommendation. It is a structured signal designed to inform research, not replace independent due diligence.

Versioning

Current methodology version:
v4.0
Last updated:
October 18, 2026
What changed:
A transparent 6-dimension composite uses smoothing sized to how often each signal refreshes, with model releases counted from a release ledger.
Feedback