Data Management
This page describes the principles under which Vertical Horizontal handles the data collected, produced and processed in its study, advisory, accompaniment and training missions. It is intended for clients, partners and the people I meet in the course of my work who wish to have clear information about my practices. It also describes how deliverables are produced and the place that artificial intelligence tools occupy in that process.
Three principles guide these choices: strict separation by data sensitivity, proportionality to actual risk rather than uniform formalism, and sovereignty of the tools used for the most sensitive data. These principles go beyond the strict legal requirements of the EU General Data Protection Regulation (GDPR).
Three-tier architecture
Data are handled according to a strict separation between three levels of sensitivity, each calling for its own tools and storage locations.
Tier 1 — Strictly confidential. Personal, identifying data: interview recordings, raw non-pseudonymised transcripts, non-anonymised client documents, interviewees’ contact details, correspondence tables. Stored exclusively on a fully encrypted local workstation (FileVault, APFS backup disks). These data never pass through an online service, even one hosted in Europe.
Tier 2 — Professional, non-sensitive. Internal methodological documents, notes and intermediate deliverables, reports in progress, general correspondence, unpublished training materials. Synchronised with a sovereign storage service hosted in Switzerland (Infomaniak kSuite), outside the scope of US extraterritorial legislation.
Tier 3 — Pseudonymised or public. Aggregated summaries without identifying quotations, public corpora, finalised deliverables cleared for circulation, reworked intellectual output. These may be processed with remote AI tools, after systematic prior pseudonymisation.
Pseudonymisation of interviews
When interviews conducted for a mission are recorded, they are transcribed locally, without online transcription services. Transcripts are then pseudonymised: names of people, companies, organisations and places are replaced by coded identifiers, recorded in an encrypted correspondence table stored separately from the corpus.
Pseudonymisation is systematic before any indexing, any tool-assisted analysis and any quotation in deliverables. Under the GDPR, pseudonymised data remain personal data: they are treated as such for retention, security and data-subject rights.
Consent and data-subject rights
Every interview is preceded by informed consent, obtained before any recording or note-taking. Interviewees are informed of the purpose of the mission, the pseudonymised nature of the outputs, who will receive the deliverables, how long the data will be kept, and their rights under the GDPR: access, rectification, erasure and withdrawal of consent.
No data collected for one mission may be reused for another without the explicit agreement of the people concerned. No data from one client or mission may be passed on to another client.
Controlled use of artificial intelligence tools
Vertical Horizontal uses generative AI tools in its exploration, analysis and document production work. This use is governed by three non-negotiable rules.
1. No confidential identifying data is submitted to a remote AI tool. Local processing on a closed corpus, using a model that runs entirely on the encrypted workstation, is permitted on identifying data. Only data that have been pseudonymised beforehand may be submitted to a remote tool.
2. Accounts on remote tools are always configured to explicitly opt out of model training. Providers that do not offer this option are excluded.
3. Some AI tools are excluded on principle when their terms of use, hosting jurisdiction or known data practices are incompatible with the above requirements.
Producing deliverables and the editorial use of artificial intelligence
The previous section describes how data submitted to artificial intelligence tools are handled. This one describes how texts are produced. The two regimes are distinct and answer different requirements: the first concerns the protection of people, the second editorial responsibility.
The intellectual substance of Vertical Horizontal’s deliverables — problem framing, structure, thesis, arguments, key ideas, analytical choices, conclusions and recommendations — is produced by Olivier Thiery. It is neither generated nor decided by an artificial intelligence tool. Any suggestions from the tool are adopted, transformed or discarded by the author.
Where the tool sits in the production chain
A language model is used at identified moments, always in an executing role, working to specification.
1. Exploration, upstream of the analytical construction: documentary research, literature scoping, identification of actors and positions, discussion of working hypotheses.
2. Processing of materials selected or produced by Vertical Horizontal: ordering, cross-referencing and extraction on corpora assembled beforehand, such as public documents, bibliographies, analytical notes or interview analyses already reworked in analytical form. The scope and selection of materials are methodological decisions and are not delegated.
3. Contradiction, as a source of objections: looking for weaknesses in a line of reasoning, counter-arguments, minority positions or omitted sources.
4. Suggestion, on a non-binding basis: proposed angles or wordings, which the author adopts or discards.
5. Drafting to instructions, once the structure, arguments and key ideas have been written and passed on: producing paragraphs that follow detailed instructions on content, level of evidence, register and length.
6. Formatting: structuring, typographic harmonisation, document production.
Technical set-up
The use of these tools rests on two stable arrangements.
1. A permanent framework, applied to all exchanges, sets the epistemological requirements of the work: mandatory verifiable sourcing, explicit flagging of uncertainty and contradictory data, a ban on producing any unverified reference, a distinction between institutional positions and established facts, and a hierarchy of levels of evidence.
2. Workspaces separated by mission or by subject, each associated with a documentary corpus assembled by Vertical Horizontal. This corpus comprises public documents, critical bibliographies and materials produced by Vertical Horizontal: summary notes, documentary analyses, interview analyses reworked in analytical form. In the sense of the previous sections, the latter are intellectual products that no longer contain identifying quotations, or materials that have been pseudonymised beforehand. No confidential identifying data enters these spaces.
Editorial reworking
Every text produced to instructions is reworked in full. Paragraphs are recast, most often several times, rewritten, and their style adjusted. The complete deliverable is then reread many times in its entirety, until the argument holds together as a whole.
Every bibliographic reference, figure or quotation is checked at the source before being included. No unverified reference appears in a deliverable.
Editorial responsibility and regulatory framework
Vertical Horizontal / Olivier Thiery takes full editorial responsibility for the deliverables, their content, their sources and their conclusions. The identity and contact details of the person editorially responsible are given in the site’s legal notice.
This arrangement meets the conditions set by Article 50(4) of Regulation (EU) 2024/1689 on artificial intelligence, applicable since 2 August 2026. That article lifts the disclosure obligation where the content has undergone a process of human review or editorial control and a natural or legal person holds editorial responsibility for its publication. Here, the review is carried out by the author himself, within his field of professional competence, and covers the entire content.
Internal working notes produced with the assistance of an artificial intelligence tool and not intended for publication carry an explicit statement of their status. They are not publications of Vertical Horizontal and do not commit their author in the same way as a finalised deliverable.
What Vertical Horizontal does not do
1. No deliverable is produced by autonomous generation followed by a superficial review.
2. No structure, thesis or recommendation comes from an unreworked proposal by an artificial intelligence tool.
3. No reference produced by an artificial intelligence tool is used without being checked at the source.
4. The contractual provisions specific to each mission, in particular the clauses of the CCAP (special administrative terms) in public procurement, take precedence over these principles. They may impose additional restrictions or disclosure obligations.
Vertical Horizontal — Olivier Thiery · Saint-Léger (Alpes-Maritimes, France). Page updated October 2026. For any question about these principles, please contact Vertical Horizontal using the details given in the legal notice.