What’s in 30,000 documents shouldn’t stay buried under 30,000 documents.
Thousands of PDFs, presentations, emails or project files. One specific question. We build a sealed-off analysis environment for it, structure the material and search for the relevant connections using private AI models and human review. Every material finding leads back to its source.
You don’t buy infrastructure or another AI platform. You buy the analysis: a verified result, the evidence behind it, and your project state handed back at the end.
- 31,842documents supplied
- 8,116possible matches
- 684relevant passages
- 37cases for review
- 12confirmed findings
Finding 07→Contract_2022_04.pdf→Page 18→Section 4.2
The answer is in there somewhere. Nobody can read all of it.
Large, sensitive and finite: a project archive, a data room, material for an internal investigation. Too much for manual review, too confidential to hand to a public AI service.
- Which documents actually matter?
- Which contracts contain a given obligation?
- Where do statements contradict one another?
- What really happened in a past project?
- Are there unusual payments, side letters or exceptions?
This happens every week. Can it run automatically?Managed AI Automation
Somewhere in these 30,000 files is the answer. Find it and show us the evidence.
We want AI available inside our own environment, permanently.Sovereign AI
From a mountain of documents to a finding you can defend.
- 1 / 4Understand the corpus
We record what is actually there: file types, duplicates, scans, unreadable files, languages and the structure of the collection.
Completeness, not a black box - 2 / 4Analyse
Documents are extracted, classified and related to one another. AI searches the collection for the project’s questions; fixed rules take over where they are more reliable.
- 3 / 4Verify
Relevant results are checked against their sources. Critical statements stay traceable and can be confirmed by your team or our analysts.
- 4 / 4Hand over and delete
You receive the results, structured data and a resumable project state. Our working environment is then dismantled and deleted following the agreed procedure.
Not just an answer. The way back to the source.
An AI answer without evidence is not enough for decisions that matter. So we link relevant results to the underlying documents, pages or sections, and we document what could be processed and what could not.
- Files supplied
- 31,842
- Processed
- 31,613
- Duplicates
- 147
- Unreadable or encrypted
- 54
- Excluded by agreement
- 28
- Findings requiring review
- 37
- Confirmed findings
- 12
Material statements point back to the underlying documents.
We report which files were processed, excluded or could not be read.
Depending on the project, we measure samples, match quality and misclassifications.
Versions, configuration and structured results are documented for the handover.
Three kinds of material we are asked to take apart.
Technical and historical archives
- Evaluate old project records and technical documentation
- Reconstruct the cause and course of past problems
- Make knowledge from SharePoint, file shares and project archives accessible again
- Compare supplier, quality or change records across years
Deals and due diligence
- Pre-sort large data rooms
- Identify unusual clauses and obligations
- Flag missing or contradictory information
- Find relevant contracts, commitments and dependencies faster
Internal investigations
- Find indications and anomalies in large document and communication collections
- Reconstruct events and decisions over time
- Bring together relevant people, transactions and evidence
- Provide results with sources for further review
We prepare analysis and evidence. Legal assessment remains with your lawyers and in-house experts.
The analysis follows the confidentiality of the data. Not the other way round.
For projects where an isolated European infrastructure is acceptable. The environment is built for the project and dismantled once it is complete.
When data may not leave the company, we bring the analysis environment with us. Processing and working data stay on site for the duration of the project.
We bring the compute. Your data stays with you.
For capabilities you need permanently, we can build and operate the analysis on your own infrastructure.
In the end, the project state stays with you. Not with us.
When the project is complete, we can hand over the full, resumable project state on encrypted drives: source data, structured extracts, analysis results, source links and technical project documentation. Before our working copies are deleted, the restore is tested.
- Analyse
- Verify
- Hand over archive
- Test restore
- Delete working environment
- Two copiesOn request, we deliver two encrypted copies with the same project ID for separate storage.
- Your keysThe key and access concept is agreed with you. For especially sensitive projects, resuming can be designed so that it is not possible without your approval.
- ResumeFor a follow-up engagement, the handed-over project state can be loaded again and migrated to the new analysis.
- Deletion recordDismantling the project environment is documented following the agreed procedure.
What Confidential Analysis is not.
We don’t build yet another assistant your team then has to run. We work on a specific analysis brief and deliver the result.
Relevant statements have to lead back to their source. Where the technology is uncertain, that is made visible or checked by a person.
We don’t earn money by keeping your documents with us.
Common questions.
Does our data have to be uploaded to a cloud?
No. Depending on how confidential the material is, the analysis runs in an isolated European environment or on infrastructure at your premises.
Do you use ChatGPT or Claude for our documents?
Confidential Analysis is designed so that confidential document content does not have to be sent to public AI services. Which models and infrastructure are used is decided for each project. Technically, it builds on the private side of our Sovereign AI stack, without routing to external AI services.
Can the AI miss something?
Yes. AI-assisted analysis is no guarantee that everything will be found. That is why we define review criteria, measure quality where that is meaningfully possible, and make uncertainty visible.
What happens to our data after the project?
Handover, retention and deletion are agreed before the project starts. If you wish, you receive a resumable, encrypted project state; our working environment is then dismantled following the agreed procedure.
Can we continue later?
Yes. A handed-over project state can be loaded again for a follow-up engagement and migrated to the current analysis environment.
What volumes is this meant for?
Mainly for collections that are too large or too unwieldy for manual review. The right setup depends more on file types, share of scans and the analysis question than on a single gigabyte threshold.