What I look for here is provenance. Not just what data went into the model, but where it came from, who decided it was fit to use and what was done to it before it was used. If that is not written down anywhere, it is very hard to answer a question about bias later.
This form template is supplied in fully editable MS Word format and can be easily customised and edited. The document uses pre-defined style formatting throughout, with all text in Calibri font for improved readability.
The following document labelling is included in the footer:
- Form number and title
- Information-classification
- Version
- Page number / total pages
alphaZ documents - beautifully designed, tried and tested form templates. No junk, jargon or unnecessary fields. Simple and usable form templates
developed over 25 years through practical use in the real world. No AI generated nonsense here!
Further guidance on the data behind AI systems:
Download this form template with your company name and logo already added!
Document Preparation available with all document toolkits.
To get access to this file please click on Add to Cart to purchase for immediate download. If you have an alphaZ Subscription just click on Download Files to view all the files available to download. This form is also included in the ISO 42001 Toolkit.
If you'd like hands-on help documenting the data behind your AI systems, or setting up your wider AI management system, our remote support services provide direct expert input from our team, by the day, by project, or on an ongoing basis.
AI Data Matrix Form Template
A record of the data used in the development of AI systems - what is needed, where it is obtained, the quality it has to meet, how its provenance is tracked and how it is prepared before use.
The Annex A controls of ISO 42001:2023 deal directly with data for AI systems, covering acquisition, quality, provenance and preparation. This form is the documented information behind those controls, and it is the record that makes a later question about bias or data quality answerable.
Built for ISO 42001
The form follows the data controls of the standard rather than general data-protection paperwork:
- Data requirements determined before development, not reconstructed afterwards
- Acquisition and selection recorded as a process, so the same decisions are made consistently
- Provenance tracked across development life cycles, which is what makes an output defensible
- Preparation criteria and methods captured, so a later result can be reproduced
What's Included in this AI Data Matrix
A short header for who completed it and when, then five recorded areas:
- Description of the data required - what data the development of the AI system needs
- Acquisition and selection - the process by which data is obtained and chosen for use
- Data quality requirements - the standard the data has to meet to be used in development
- Provenance - the process for recording where data came from, tracked over development life cycles
- Preparation criteria and methods - what is done to the data before it is used
Who Needs an AI Data Matrix?
Any organisation that trains, tunes or configures an AI system on its own data, and anyone who has to answer a customer or regulator asking what the model was built from. It is a short form, and it is usually the first thing missing when an organisation has otherwise documented its AI use well.
Included in the ISO 42001 Toolkit
This form is part of the ISO 42001 Toolkit - the policies, procedures, registers, audit checklists and guidance for running an AI management system in one download. The other three AI forms are listed on the F-AI forms page.
Pairs with the AI System Life Cycle Form
The data matrix records what went into the system; F-AI1 AI System Life Cycle records how it was specified, validated and deployed. Both feed the ER42 AI Risks and Controls Register.