How the capability model works

An independent, evolving model

The AIUseCaseHub AI-Native Capability Model synthesizes practical AI transformation patterns into an editorial taxonomy. It is our own working model, not an industry standard or a reproduction of a proprietary framework. Version 1.0 is a curated starting point for discovery, assessment, and workshop prioritization.

A corpus-first comparison

3,846 public, unmerged records were analyzed from a read-only export of 4,050 corpus records on September 6, 2026. The existing publication and merge rules excluded 204 records.

The first model was curated from transformation practices. We subsequently ran a separate bottom-up pass: discover topics from case narratives, abstract organizational abilities, then compare them with the framework. The topic algorithm did not receive the framework, existing use-case categories, or capability flags. Editorial review produced 20 candidate capabilities in 5 groups.

What the corpus adds

The comparison suggests making the abilities below more explicit. Counts show records matching reviewed narrative phrases; they do not establish that a company implemented every control or that these abilities are necessary.

Candidate footprints overlap. 2,776 records match at least one candidate rule; 1,070 match none. An unmatched record may use other wording or describe a different practice.

Evidence limits and framework safeguards

The stored corpus quality tiers are 79 verified, 2,335 supported, and 1,432 collected. Those are record-quality classifications, not verification of our capability inferences. Two infrastructure-dominant topics contain 333 records; we retained them and flagged the reporting bias.

This analysis uses existing editorial and AI-assisted summaries, rather than a fresh audit of every original source. Public case studies often describe applications and outcomes more readily than investment governance, knowledge ownership, or accountability. A practice that does not emerge as a distinct topic can still be essential.

The 20 candidates are a research comparison, separate from the current assessment taxonomy. Existing capability IDs and saved scores remain stable. Topic discovery withheld the framework; the human reviewer knew it. The findings should guide editorial revisions and source review, not automatically replace the model.

Five distinct layers

Domain

A broad area of organizational ability, such as Knowledge & Context.

Capability

What an organization must be able to do, such as provide Enterprise Knowledge Access.

Building block

A business, technical, people, or governance practice that needs to exist, such as enforcing source permissions during retrieval.

Use case

A concrete application from the existing deployment corpus that these practices can enable.

Outcome

The intended business change. An outcome listed here is an objective, not a promised benefit.

A survey of observable practices

The guided assessment asks business and transformation leaders 32 questions across six sections. Answer for one organization or business unit, using actual practice in the last six months. A practice can inform more than one capability. The final evaluation covers all 28 capabilities, independently of catalogue filters and any scope saved with earlier self-ratings.

Choose the closest of five behavior descriptions, or explicitly answer “I don’t know” or “Not applicable”. These are missing evidence, not low scores. The evaluation is revealed only after all questions have a response and you choose to finish. Reviewing or changing answers hides it again until you finish the review.

Assessment scoring

Every capability has at least two mapped questions, including a direct practice. Direct practices have weight 2 and supporting practices weight 1. An estimate requires at least two known answers, including a direct practice, and at least 60% of its total mapped question weight.

The indicative practice score is the weighted mean of known answers, rounded to one decimal place and capped at the lowest answered direct-practice rating. Supporting practices cannot conceal a weak direct practice. Bands round to the nearest whole level: Not established, Early practice, Developing, Established, and Continuously improving. Each result explains its supporting answers.

Answer coverage describes completeness, not statistical confidence. Insufficient evidence produces no estimate. There is no overall enterprise readiness score. The questions, weights, and coverage rule are editorial choices, not a validated diagnostic instrument.

Transparent recommendations

• An estimate of 3.5 or higher indicates a strength to build on. Lower estimates identify practices to develop. Missing estimates identify topics to clarify.
• Development areas are ordered by selected business-priority matches, then lowest practice score, then stable capability ID. The first three are discussion starting points, not a ranking of risk or urgency.
• Business priorities do not change practice scores and are not treated as target maturity. No priority-minus-maturity gap is calculated for the survey.
• Suggested building blocks have a foundational maturity level above the estimated practice score. Validate them with the teams responsible before turning them into a plan.

No LLM calculates recommendations. Each result lists the answers, weights, and rationale used. Earlier direct capability self-ratings remain in saved data and exports; they are not combined with survey results.

Question design and validation

Question design follows the emphasis on concrete wording and a single concept in Pew Research Center’s questionnaire guidance. Governance coverage is cross-checked against Govern, Map, Measure, and Manage in the NIST AI RMF Playbook. Neither source validates this questionnaire or its mappings.

Next, test question interpretation with business leaders, compare answers with evidence from service owners, and examine agreement between independent respondents. Corpus evidence supporting the taxonomy does not by itself validate the survey.

Use-case mapping evidence

87 editorial mappings connect 20 corpus use cases to 28 of 28 capabilities. This curated set is separate from the full-corpus research analysis above; most corpus records have not yet received reviewed implementation mappings.

Mappings reference stable IDs in the existing corpus. Their rationale, relationship, confidence, and editorial source are shown alongside each case. Source-reviewed additions link directly to the supporting publication. An enabling relationship is an implementation judgment, not evidence that a named company has every building block. Live lookups suppress unpublished records; retired or unavailable mappings are not presented as working case links.

Local data and future benchmarks

Survey answers, progress, the optional business-unit label, priorities, and earlier self-ratings are saved on this device. You can explicitly save a copy to your signed-in account or export a JSON file. They are not included in interaction analytics. The current release does not collect benchmark contributions.

Peer benchmarks will appear once sufficient anonymized assessment data is available. A future opt-in aggregation process will use industry, company-size, and region cohorts, count distinct consenting organizations, and suppress every capability result below the configured minimum (at least 20 organizations). Individual organization data will never appear in benchmark responses. Small-cohort differencing and consent withdrawal need safeguards before collection launches.

Limitations and what comes next

Self-assessment is subjective, prerequisites vary by context, and this seed cannot cover every industry. Validate consequential decisions with appropriate domain experts. Process design is a future extension: an activity can connect performer, authority, context, output, review, iteration, and outcome. This release does not execute agents or automate workflows.