How we score readiness

Methodology.

Five pillars, explicit weights, a benchmark with a floor, and every recommendation traceable back to a published framework. Reproducible from the numbers below.

PillarsScoringTiersFormulasBenchmarksFrameworksIntegrityFAQ

Five pillars

The things that actually predict AI success.

Weights reflect what blocks production AI programmes most often โ€” governance and data carry the heaviest load.

  • Strategy & Leadership

    20%

    Ambition, exec sponsorship, investment posture, first use cases.

  • Data Readiness

    25%

    Catalogue, quality, lineage, governance and access policy.

  • Infrastructure & Technology

    15%

    Cloud footprint, MLOps, RAG, sandbox and observability.

  • Talent & Skills

    15%

    Specialists, broader AI literacy and clear career paths.

  • Governance & Risk

    25%

    AUP, tool inventory, oversight, vendor review and regulation.

Question scoring

Every answer maps to 0 โ€“ 4.

  • Likert 1-5

    0 โ€“ 4 integer

    Confidence statements and maturity self-assessments.

  • Yes / No

    0 or 4

    Binary presence-of-control questions.

  • Percent buckets

    0 โ€“ 4 โ†’ 0-10%, 10-30%, 30-60%, 60-85%, 85%+

    Adoption, coverage and frequency questions.

  • Multi-select

    Capped at 4

    Framework awareness, training topics, tool inventory.

Maturity tiers

Shorthand for your overall score.

  • Nascent

    0 โ€“ 25

    Starting out. No exec sponsor, no policy, no plan. The first move is the first move.

  • Emerging

    26 โ€“ 50

    Pockets of activity. Pilots exist but donโ€™t talk to each other. Data and governance are the bottlenecks.

  • Scaling

    51 โ€“ 75

    Moving beyond pilots. A proper runway exists. Most gaps are solvable in a quarter.

  • Integrated

    76 โ€“ 100

    AI is woven into how the company works. The next moves are second-order โ€” compounding effects.

Scoring formulas

The math, laid out.

# Pillar score (per focus area, 0 - 100)
pillar_score = 100 ร— ฮฃ (answer_value ร— question_weight)
                   / ฮฃ (max_answer_value ร— question_weight)

# Overall score
overall = 0.20 ร— strategy
        + 0.25 ร— data
        + 0.15 ร— infra
        + 0.15 ร— talent
        + 0.25 ร— governance

# Maturity tier
overall in [0, 25]    -> Nascent
overall in [26, 50]   -> Emerging
overall in [51, 75]   -> Scaling
overall in [76, 100]  -> Integrated

Unanswered questions donโ€™t inflate the denominator โ€” the formula uses max_answer_value ร— question_weight only over answered questions.

Peer benchmarks

Seeded from research. Refreshed by you.

Segments are cut by industry ร— company size ร— pillar. A segment needs at least 30 completed assessments to render โ€” below that threshold we fall back to the industry-wide average rather than mislead with thin data.

  • Seed sources ยท McKinsey, Deloitte, Gartner, IDC

  • Continuous refresh from anonymised user data

  • Minimum 30 assessments per segment

  • Fallback to industry average when sparse

Framework mapping

Tagged to the frameworks auditors already reference.

  • NIST AI RMF

    Govern ยท Map ยท Measure ยท Manage

    Every question maps to a sub-function so findings align with frameworks your risk team already tracks.

  • EU AI Act

    Articles 4, 9, 10, 13, 14, 15, 17, 62

    High-risk categories and transparency obligations are flagged where the assessment touches them.

  • ISO/IEC 42001

    Clauses 5, 7, 8, 9, 10

    AIMS-aligned questions for teams on the ISO path โ€” the answers double as audit-ready evidence.

  • Sector overlays

    HIPAA ยท SR 11-7 ยท NYDFS 500 ยท PCI ยท ECOA

    Healthcare and financial-services banks add sector-specific questions on top of the generic set.

What we donโ€™t do

Three rules that keep the score honest.

  • No training on your answers.

    Answers are stored in our Supabase project and used only to score your own assessment. Aggregates used for benchmarks are anonymised and can never be reconstructed back to a company.

  • No vendor pay-to-play.

    Recommendations are editorial. Vendors do not influence which suggestions surface โ€” the impact ร— ease filter and trigger conditions are published.

  • No invented scores.

    Segments with fewer than 30 completed assessments fall back to the industry-wide average. We would rather be honest about sparse data than make one up.

FAQ

Methodology โ€” in more detail.

How is each pillar weighted?

Strategy & Leadership 20%, Data Readiness 25%, Infrastructure & Technology 15%, Talent & Skills 15%, Governance & Risk 25%. Governance and Data carry the most weight because they are the most common blockers to production AI.

How is each question scored?

Every question maps to a 0-4 integer. Likert 1-5 maps to 0-4. Yes/No maps to 0 or 4. Percent buckets (0-10%, 10-30%, 30-60%, 60-85%, 85%+) map to 0-4. A pillar score is the weighted sum of its questions divided by the max weighted sum, then normalised to 0-100.

Where do benchmark numbers come from?

Benchmarks are seeded from published research (McKinsey, Deloitte, Gartner, IDC) and continuously refreshed from anonymised user data. A segment (industry ร— size ร— pillar) needs at least 30 completed assessments to appear; otherwise we fall back to the industry-wide average.

How are recommendations chosen?

We score each recommendation by impact ร— ease (both on a 1-5 scale), filter by trigger conditions (for example, pillar score below 55), then surface the top 5. The full library is browsable on your results page.

What frameworks are referenced?

Recommendations and questions are tagged against NIST AI RMF, EU AI Act (Articles 4, 9, 10, 13, 14, 15, 17, 62), and ISO/IEC 42001 clauses. Vertical banks also reference HIPAA sections (healthcare), and SR 11-7 / NYDFS 500 / ECOA (financial services).

What is a maturity tier?

Tier is derived from the overall 0-100 score: Nascent (0-25), Emerging (26-50), Scaling (51-75), Integrated (76-100). Tiers are shorthand โ€” the real value is in the pillar breakdown and the three highest-leverage moves for this quarter.

Do you use my answers to train AI models?

No. Answers are stored in our Supabase project and used only to score your own assessment and โ€” in aggregate, anonymised form โ€” to refresh peer benchmarks and our annual State of AI Readiness report.

Spotted something?

Corrections welcome.

A weight that feels wrong, a framework we missed, a sector we should map? Email us โ€” we review within 48 hours.

hello@buzzi.ai

Run the check

Score yourself in 10 minutes.

The formulas plug directly into your answers โ€” one overall score, five pillar scores, and a peer benchmark.

Back to assessment