Four things that are true of every engagement
Four mechanisms. None of them is a certification.
EU-hosted analysis
By default, analysis runs on Google Vertex AI in the EU, under a Data Processing Agreement. It leaves the EU only if you choose that for a deal.
Zero retention at the model provider
The model provider does not keep your documents or train on them.
Pseudonymised before processing
Confidential data is pseudonymised before analysis. Reversed only in the final output.
Your reviewer signs off
No client export leaves the platform without your own reviewer’s sign-off.
The one mandatory human step
Your own reviewer signs off before any client export. These are the rules behind it.
- 01The pipeline checks claims against retrieved source passages and records evidence gaps for your reviewer.
- 02The gate runs every time, automatically.
- 038 hard-block conditions stop a draft and put it in front of your reviewer, unapproved.
- 04Your reviewer signs off each module’s summary, the top ten risks and every condition precedent. Without that, no client export leaves the platform. The rest goes along marked as not separately reviewed.
- 05Pseudonymised values are reversed only here, in the final output.
3 modules never run on a model that retains data
Zero data retention is a routing rule in the code, not a promise on a page. Tax, legal and financial are gated on it.
- ZDRThe gate sits at the one point every model call passes through.
- ZDRA provider that retains data is swapped for a zero-retention one, and the swap is logged.
- ZDRBy default that is Google Vertex AI in the EU, under a Data Processing Agreement.
- ZDRThe other 6 modules run on the same infrastructure. Today every provider in the platform is zero-retention; for these 3 it is enforced in code.
Two memories. One is shared, one never leaves your tenant.
What we learn across clients, and what stays with you.
Methodology lessons — how a check should be run, where a class of analysis tends to go wrong. Scrubbed of personal data before storage. The top lessons are re-injected into later runs.
- Method, not content
- Personal data scrubbed before storage
- No client names, documents or figures
Rules derived from your own engagements. Strictly scoped to your account; the service refuses to run without a client identifier.
- Your documents and findings
- Your derived rules and preferences
- Scoped by client ID, enforced in code
If cross-client methodology learning is a problem for your firm, say so on the call. It can be switched off for your account.
Why this is not a chat window
One from the small print. One from your own NDA. One from the statute book.
A public AI tool is not a private one
A public AI tool’s own terms of service usually let the provider retain what you type and disclose it under a legal order, wherever that provider is based. No confidentiality agreement or legal privilege reverses that once it has happened.
Your NDA may not cover the tool
Most M&A NDAs permit disclosure only to named Representatives: staff, advisers, financing sources. An AI platform is rarely one of them. Putting the data room into a consumer tool can be the breach itself.
The EU AI Act runs on a clock
General-purpose AI obligations have applied since 2 August 2025. The Article 50 transparency duties have applied since 2 August 2026. A documented human gate is easier to place inside that regime than a chat session nobody logged.
None of this is legal advice. All three are reasons the pipeline is shaped the way it is.
The terms
What is in the engagement agreement, before you have to ask for it.
- 01
An NDA first
Signed before a single document moves. Not after scoping, not alongside the engagement letter. First.
- 02
Who you contract with
Factum Capital is not yet a separate legal entity. Engagements are contracted under KvK 99495945, Netherlands. Incorporation is with the notary.
- 03
Liability, capped
Our liability is capped at the price of the engagement. Consequential loss, lost profit and missed IRR are excluded.
- 04
Professional indemnity
Cover runs through the current contracting entity, KvK 99495945. Factum’s own policy follows its incorporation.
- 05
What we keep, and why
Your documents and findings stay in your own tenant, so a later dossier can read across an earlier one. Deleted on request, at any point. The term is set in the engagement agreement.
- 06
Sub-processors
By default, analysis runs on Google Vertex AI in the EU, under a DPA. Scheduling runs on Cal.com. This website counts visits with Plausible Analytics, without cookies. The full list per module is on the table before you upload anything.
Ask for it in writing. Under DORA, a fund above the threshold has to name every ICT third party in a register filed with its supervisor — including us.
What the machine does, and what the person does
You should be able to tell, for any finding, what produced it and what checked it.
- Machine01
Each module retrieves evidence from your uploaded documents on its own.
- 02
A second critic agent critiques each draft.
- 03
A separate pass fact-checks claims against the source text.
- 04
Claims that can’t be tied to a passage are flagged as unconfirmed.
- Person05
Your reviewer checks the result and signs off.
What we don’t claim
Most of what a buyer wants on a page like this, we don’t have yet. Listing it beats implying it.
- No ISO 27001, SOC 2 or any other certification. No third-party audit of the pipeline.
- No penetration test, single sign-on or audit-log commitment.
- No single named data centre. By default, analysis runs on Vertex AI’s EU multi-region endpoint.
- No asserted GDPR-compliance status. Processing terms belong in an engagement agreement.
- No breach-notification or disaster-recovery commitment stated here.
- No audited accuracy rate. Internal benchmarks are single-dataset and unaudited.
If your process needs any of these in writing, raise it on the call. There is a real answer either way.