How AEAI is Calculated
Every number in the AEAI is traceable to a defined formula, a named data source, and a published signal. This document is the complete specification — open for peer review, criticism, and improvement.
1. Overview
The Australian Enterprise AI Index (AEAI) is a quarterly benchmark measuring the maturity, safety, and momentum of artificial intelligence adoption across Australia's enterprise sector. It produces a single national score (0–100) and six industry sub-scores, updated each quarter, enabling organisations to track progress relative to peers, identify governance gaps, and make evidence-based decisions about AI investment.
Three questions AEAI is designed to answer:
How mature is Australian enterprise AI, relative to last quarter and relative to peers?
Where are the systemic governance gaps that create regulatory and reputational risk?
Is investment translating into capability — or just into announcements?
AEAI draws on primary survey data, official government disclosures, ASX/investor filings, and verified incident reporting to construct a composite score that is rigorous enough for board-level reporting and readable enough for executive briefings. Scores are published in the third month of each quarter (March, June, September, December).
Quarterly
National score
6 industries
Industry sub-scores
16 sources
Data sources
6 stages
Signal pipeline
2. Scoring Architecture
Composite Formula
AEAI_national = (Adoption × 0.25) + (Governance × 0.25) + (Investment × 0.25) + (Incidents × 0.25)
Measures the breadth and depth of AI deployment across Australian enterprise — distinguishing announced intent from operational deployment.
Measures whether enterprises have the policies, structures, accountabilities, and audit mechanisms to deploy AI responsibly.
Measures financial commitment to AI as a strategic priority — distinguishing one-off procurement from sustained strategic investment.
Captures AI-related failures, harms, breaches, and adverse outcomes. The only dimension where higher raw signal = lower score.
What the score means — 0 to 100 scale:
Pre-adoption
AI activity is isolated, ungoverned, or not yet measurable
Experimental
Pilots underway without governance, investment strategy, or incident frameworks
Developing
Meaningful adoption exists; governance and investment are inconsistent
Maturing
Adoption is broad; governance frameworks in place; investment is strategic
Advanced
AI embedded across operations; governance is auditable; ROI is measurable
Leading
Globally competitive maturity; governance is proactive; incidents are rare
A score of 50represents the midpoint of "Developing" — where Australian enterprise AI currently sits in aggregate. Each dimension contributes equally (25%) — excellence in one cannot compensate for failure in another.
3. Dimension: Adoption
Measures the breadth and depth of AI deployment across Australian enterprise — distinguishing announced intent from operational deployment.
Data Sources
Sub-Metrics
Enterprise functions with AI in production (not pilot)
Share of employees using AI in standard workflow
Distribution across automation / augmentation / autonomous
Average time since first AI system reached production
Degree of AI integration with external data and APIs
4. Dimension: Governance
Measures whether enterprises have the policies, structures, accountabilities, and audit mechanisms to deploy AI responsibly.
Data Sources
Sub-Metrics
Published AI policy, ethics frameworks, and accountability structures
AI risk register, model inventory, and review cadence
Alignment with APRA CPS 230, ASIC guidance, OAIC Privacy Act requirements
Disclosure of AI system use, decisions, and audit trails
Human-in-the-loop controls for high-risk AI decisions
5. Dimension: Investment
Measures financial commitment to AI as a strategic priority — distinguishing one-off procurement from sustained strategic investment.
Data Sources
Sub-Metrics
AI-dedicated spend as percentage of total technology budget
AI/ML engineering and data science hiring rate
Multi-year AI roadmaps, board-level investment approvals
AI vendor contracts, acquisitions, joint ventures
AI-related patent filings and research program expenditure
6. Dimension: Incidents
Captures AI-related failures, harms, breaches, and adverse outcomes. The only dimension where higher raw signal = lower score.
Data Sources
Sub-Metrics
Frequency of disclosed AI failures per deployment (inverse-scored)
Proportion of high-severity vs low-severity incidents (inverse-scored)
ASIC, APRA, OAIC, or ACCC enforcement actions related to AI (inverse-scored)
Transparency and completeness of incident reporting (positive credit)
Time from incident discovery to resolution (positive credit)
Inverse Scoring Formula
Incidents_score = 100 − (weighted_incident_signal_average × 100)
A high-Adoption industry reporting zero incidents is treated with scepticism — the Incidents score is capped at 80 with a detection-gap flag.
7. Industry Coverage
AEAI tracks six industries chosen for their AI adoption velocity, regulatory significance, and economic weight. Banking, Insurance, and Government receive a 1.3× weight premium due to their greater systemic consequence.
National score formula
AEAI_national = weighted_mean(industry_scores, weights=[1.3, 1.3, 1.3, 1.0, 1.0, 1.0])8. Data Sources
Sixteen sources across three confidence tiers. Tier 1 (Primary) carries a 1.5× weight multiplier, Tier 2 (Secondary) 1.0×, and Tier 3 (Tertiary) 0.6×.
9. Signal Processing
Raw signals undergo a six-stage processing pipeline before contributing to dimension scores.
Ingest
Signals are ingested with mandatory fields: source name, tier, dimension, industry, headline, URL, and timestamp.
Classify
Each signal is scored on two axes: Sentiment (−1.0 to +1.0) and Magnitude (0.0 to 1.0) by the analyst team.
Weight
Contributions are adjusted by source tier multiplier (1.5× / 1.0× / 0.6×), magnitude, and a recency discount (0.95× per month of age).
Normalise
Weighted aggregates are normalised 0–100 via min-max anchoring to historical min/max. Incidents dimension is then inverted: Score = 100 − raw.
Confidence Intervals
Bootstrap resampling (500 iterations, 90% CI) per dimension per industry. Low signal count (n < 5) widens CI and flags low-confidence.
Aggregate
Dimension scores combine into industry scores (4 × 25%). Industry scores combine into national score via weighted mean (Banking/Insurance/Government 1.3×).
Worked Example — From Raw Headline to Score Contribution
Source: CBA Annual Technology Report | Tier: 1 (Official, weight: 1.5×) Headline: "Commonwealth Bank deploys real-time AI fraud detection across 10M+ accounts" Dimension: adoption | Industry: banking | Timestamp: 2026-04-15T09:00:00Z
Positive keyword matches: deploy (+), launch (+), expand (+) sentiment = +0.60 | magnitude = 0.80 (high-significance announcement)
tier_multiplier = 1.5 | recency_discount = 0.95 (1 month old) weighted_contribution = sentiment(0.60) × magnitude(0.80) × tier(1.5) × recency(0.95) = 0.684
With 42 signals in Banking/Adoption this quarter: Σ(weighted_contributions) = 28.4 | signal_count = 42 Raw aggregate = 28.4 / 42 = 0.676 → min-max normalised → 74.5 (final Banking Adoption score) This signal's contribution: Δ +0.37 points
Bootstrap resampling: 500 iterations, n=42 signals 90% CI: 74.5 ± 3.2 points (sufficient signal count — n ≥ 5) Low-confidence threshold: n < 5 would flag with ⚠ on the dashboard
Banking/Adoption = 74.5 | Banking/Governance = 71.2 | Banking/Investment = 79.1 | Banking/Incidents = 46.4 Banking overall = (74.5 + 71.2 + 79.1 + 46.4) / 4 = 67.8 National weight: 1.3× (APRA-regulated, high AI adoption) Contribution to national score: 67.8 × 1.3 / Σ(industry_weights) = 10.1 of 61.4
10. Limitations and Caveats
No index is perfect. These are the known limitations of the AEAI methodology. We document them openly to prevent misinterpretation of the scores.
Government AI activity is materially under-reported. Utilities is the least visible industry. Interpret Utilities scores with lower confidence than Banking or Retail.
Signals favour organisations that publicly disclose AI activity. Smaller and unlisted companies are underrepresented. AEAI reflects the visible enterprise AI landscape — the actual median is likely lower.
Survey respondents may overstate maturity (social desirability) or understate incidents (reputational protection). Cross-validation against public data reduces but cannot eliminate this bias.
AEAI is a retrospective measure, not a real-time dashboard. A typical 6–10 week lag exists between signal occurrence and Index publication.
Methodology improvements may cause quarter-on-quarter deltas to reflect methodological changes as much as genuine market movement. Material changes are flagged in the Changelog.
AEAI does not sharply distinguish traditional ML from generative AI in all sub-metrics. A dedicated generative AI sub-index may be introduced in a future version.
11. Changelog
Published June 2026
- Initial publication of the AEAI framework
- Four dimensions established: Adoption, Governance, Investment, Incidents
- Six industries defined with weighting (1.3× for Banking, Insurance, Government)
- Scoring range 0–100 with min-max normalisation anchored to pilot dataset
- Incidents dimension inverse scoring introduced
- Six-stage signal processing pipeline defined
- Confidence interval methodology (bootstrap, n=500, 90% CI)
- 16 data sources catalogued across three tiers
Next scheduled release: Q3 2026 (September 2026)