Method
How we analyse.
This page sets out how answers become reviewed findings: survey, analysis, review gate and data path. Every number in the report can be recalculated.
Five steps, in a fixed order
- 01about 10 min
Scoping
All respondents jointly fix the subject: unit, roles and dimensions, including agreement on the additional dimensions. The catalogue is then frozen under a version number. Every answer and every report carries that version.
- 0245 to 60 min per person
Survey
Each person answers separately, on the same anchored scale, and states how certain they are. Don’t know counts as a missing value, never as zero. The order of the dimensions rotates so that position in the questionnaire does not skew the result.
- 03under 5 min
Follow-up questions
At most three questions per person, triggered deterministically - for instance when an answer deviates by two or more steps from the median of the other roles. They sharpen the analysis but change no scores.
- 04automatic
Analysis
Median per question, mean per dimension, disclosed bands from not established to managed. No model can alter a score. The calculation rules are set out in the methodology appendix of every report.
- 05mandatory
Expert review
No report leaves the system without being checked and released by a person. That is enforced technically; there is no override switch. Every report states who released it and on which catalogue version.
What the catalogues build on
Our catalogues are models in their own right, built on established standards: built on them, not certifying against them.
- NIST AI RMF 1.0
- Structure of the maturity dimensions (Govern, Map, Measure, Manage)
- Assess
- ISO/IEC 42001:2023
- AI management system: governance, role and operating questions
- Assess
- ISO/IEC 23894:2023
- AI risk management: the risk and compliance dimension
- Assess
- EU AI Act
- Literacy obligation under Art. 4, prohibitions, transparency duties under Art. 50
- Assess and Learn
Validating the instrument
A maturity model is a measuring instrument. How dependable it is depends on how many perspectives are collected and how far apart they lie. That question can be simulated rather than asserted.
From how many perspectives onward does the agreement measure become stable - and how does it behave when answers are missing?
Figure and citation will follow after completion and peer review. Until then this space deliberately shows no result.
Questions about the method?
We will walk you through it, with your subject experts at the table if you like.
Write us an email