FACTUM.
Share a data room

How the platform reads a data room

What happens to your documents, step by step. 9 modules read the room in a fixed order, every claim is checked against its source, and your own reviewer signs off.

Three material findings your team did not already have, or the engagement is free, on pilot mandates.

Output contract — every finding6/6
ModulePre-scoped, one of the library
FindingOne claim, in one sentence
Verbatim excerptQuoted word for word, per citation
SeverityOne of five levels, Info to Critical
Source and locationData room or knowledge base, per citation
Human reviewReview status; export requires your sign-off

+3 fields, on a wider screen

The schema every finding has to satisfy. Source material behind each one, from the data room or the knowledge base (statute, IFRS, case law, market research), with its exact location. That, and your reviewer’s sign-off, can still be opened months later.

Inside the review workspace

A constructed dossier on a fictional German company: 8 findings across 4 disciplines. Open one, check the source, decide.

Halvoria Antriebstechnik GmbH
  1. 1. Set up dossier
  2. 2. Review
  3. 3. Report to client
Constructed example

Fictional company · 101 documents in the data room8 findings await your decision

FIN-01High1 of 2

EBITDA 2025 includes a one-off €640,000

Module
Financial · Quality of earnings
Finding
Reported EBITDA 2025 of €3.22m includes a €640,000 release of a warranty provision. Adjusted EBITDA is €2.58m. The management accounts show no normalisations.Halvoria Antriebstechnik GmbH
Why a finding

Observed: The notes call the €640,000 release non-period income 1. It sits in other operating income of €720,000 2. The management accounts list zero normalisations 3. The IM states there were no one-offs 4.

Standard: Quality of earnings removes one-off and non-period items from EBITDA. The draft SPA prices the business at 7.76 × EBITDA 2025 from the management accounts 5.

Application: Adjusted EBITDA is €2.58m, a 10.5% margin against 13.1% in the IM. At the SPA multiple, the release carries about €4.97m of enterprise value.

Verbatim excerpt
In den sonstigen betrieblichen Erträgen ist die Auflösung der Rückstellung für Gewährleistungen aus dem Serienanlauf Baureihe HX-4 in Höhe von EUR 640.000 enthalten. Die Gewährleistungsfrist ist im Geschäftsjahr abgelaufen; Inanspruchnahmen erfolgten nicht. Es handelt sich um einen periodenfremden Ertrag.Marked in the source, p. 3 · ✓ Word for word from the document
Severity
High
Calculation
EBITDA 2025, reported 3€3,220,000
−Release of warranty provision 1€640,000
=Adjusted EBITDA €2,580,000
Adjusted EBITDA €2,580,000
÷Revenue 2025 2€24,600,000
=Adjusted EBITDA margin 10.5%
Release of warranty provision 1€640,000
×SPA multiple 57.76×
=Enterprise value carried €4,966,400
Every operand cites its source.
Source and location
  1. 1Jahresabschluss 2025 (HGB)p. 3
  2. 2Jahresabschluss 2025 (HGB)p. 2
  3. 3Management Accounts 2023–2027sheet GuV Management!D3:D4
  4. 4Information Memorandum (Auszug)p. 1
  5. 5Geschäftsanteilskaufvertrag (Entwurf)§ 3.1
Human review
Awaiting your reviewer. Nothing reaches the client before sign-off.

Demo only. Nothing is saved.

Constructed example. Halvoria Antriebstechnik GmbH and every name, document and number here are fictional. Quotes stay in the document’s language. Decisions are not saved.

From upload to a finished finding

Six stages. The same six, every time, whether or not anyone is watching.

  1. 01

    Grounded retrieval

    The module goes into the data room and pulls the passages that answer its question.

    Confidential financial and personal data is pseudonymised before it reaches this layer.

  2. 02

    Drafting against the schema

    A claim without a document and a passage behind it is not a finding.

  3. 03

    Generate-critique-retry

    A critic agent reviews the draft and can send it back once, with its feedback.

  4. 04

    Four-layer grounding check

    Index, quote, entailment, second opinion. Each layer catches what the one before it missed.

  5. 05

    Cross-discipline reconciliation

    Findings are read against each other and resolved into one narrative.

  6. 06

    Your review and sign-off

    Your reviewer checks the findings and signs off before export.

    Pseudonymisation is reversed only here, in the final delivered output.

Four checks, each narrower than the last

These checks help expose evidence gaps. Your reviewer decides what can be used.

  1. 01

    Citation calibration

    Every citation index is checked against the set of passages that were actually retrieved.

    CatchesCitations pointing at a document the run never opened.
  2. 02

    Lexical quote check

    The quoted text is checked against the source passage, allowing for formatting and PDF-extraction differences.

    CatchesParaphrase presented as a verbatim excerpt.
  3. 03

    Entailment judge

    A separate model reads the claim and the passage and rules on whether one follows from the other.

    CatchesReal quotes that do not support the claim built on them.
  4. 04

    Second-opinion verify

    Claims still disputed after repair can receive a second model judgement and an optional local NLI check. Without supporting evidence from either, the warning remains.

    CatchesPotential false negatives from the first judgement.

Failed claims can be re-quoted and re-checked. Remaining evidence gaps are marked for review.

The output contract, field by field

Six fields. A finding that cannot fill all six is not delivered as a finding.

  1. Module

    Which of the 9 produced it. Pre-scoped, so the question was written before the data room was opened.

    When it can’t be filledThere is no free-text lane. No module, no finding.

  2. Finding

    One claim, in one sentence. Compound claims are split, because a sentence with two claims in it can only ever be half-checked.

    When it can’t be filledToo vague to check is treated the same as wrong.

  3. Verbatim excerpt

    The supporting passage, quoted word for word. Not summarised, not tidied, not shortened to fit.

    When it can’t be filledA paraphrase presented as a quote fails the lexical check and goes back.

  4. Severity

    The consequence, measured against the deal value, on a five-level scale. The module that wrote the finding assigns it.

    When it can’t be filledAn unranked finding reads like every other line on the page.

  5. Source document and passage

    File name and page, so you can open it yourself. This is the field that makes a finding arguable instead of authoritative.

    When it can’t be filledA citation that does not resolve to a real passage is a hard block, not a warning.

  6. Human review

    Your reviewer approves or rejects each finding; a rejection needs a reason. Before export, the module summaries, the top ten risks and every condition precedent must be signed off.

    When it can’t be filledNo sign-off, no client export. What was not reviewed goes along marked as such.

A failed check can trigger repair and another judgement. Remaining evidence gaps are marked for your reviewer; a claim’s review status stays visible.

8 hard-block conditions

A rejected draft gets one retry. These conditions end the loop and put the draft in front of your reviewer, unapproved.

01

Privacy leak

Data crossed a boundary it should not have.

02

Fabricated source

A citation that does not resolve to a real passage.

03

Unsupported after repair

The fabrication check still finds an unsupported claim after the retry.

04

Scope refusal

The module cannot answer inside its brief, and says so.

05

Reviewer unavailable

The critic agent could not review the draft, so it does not pass as reviewed.

06

Placeholder text

Filler left where evidence belongs.

07

Near-empty draft

Too little substance to review.

08

Low reviewer score

The critic agent scored the draft below threshold.

What your reviewer does

In the review workspace, before anything is exported.

  1. Opens the cited passage and reads it against the claim built on it.
  2. Decides materiality: which findings change the price, or whether to sign at all.
  3. Weighs conflicts between modules. Two disciplines reading one fact differently is itself a finding.
  4. Sees what the pipeline held back, and decides on it.
  5. Signs off the module summaries, the top ten risks and the conditions precedent. Only then can the export leave.

8 hard-block conditions end the automated loop early and put the draft in front of your reviewer, unapproved.

What the engine is actually for

A module is a piece of the machine. A dimension is a question you already have, and today you get a separate invoice for each one. All 8 run over the same data room in one pass, so what one finds is on the table for the next.

  1. Financial
  2. Commercial
  3. Legal
  4. Tax
  5. HR
  6. AI
  7. Operational
  8. Valuation & deal economics

One dimension without the others is not diligence. The list is derived from the product source, not from a brochure.

The module library, in dispatch order

A module that waits on another module’s output cannot start any earlier. That dependency sets the running order. Every module hands back findings.

  1. 01

    Independent reads

    4 modules

    These modules need nothing upstream of them. They all start at once.

    • Commercial DDFinding
    • HR DDFinding
    • Operational DDFinding
    • AI DDFinding
  2. 02

    Legal

    1 module

    Reads HR and commercial before it reasons: an employment exposure or a contractual claim usually turns on facts those two reads have already established.

    • Legal DDZDRFinding
  3. 03

    Financial

    1 module

    Reads the independent reads and legal before it reasons: an earnings figure moves the moment a contractual claim or an employment exposure is on the table.

    • Financial DDZDRFinding
  4. 04

    Tax

    1 module

    Reads financial and legal before it reasons: a tax position is a consequence of the structure the financial read has just described. It often overlaps with what legal found in it.

    • Tax DDZDRFinding
  5. 05

    Deal economics

    1 module

    Reads across everything above it and asks what the deal actually returns.

    • Deal economicsFinding
  6. 06

    Valuation

    1 module

    Reads deal economics on top of everything before it, and recalculates the price through the findings rather than around them.

    • ValuationFinding

Derived from the dispatch graphs in the product source. ZDR marks the modules whose provider routing is hard-gated to zero retention. EU residency is the default.

The platform in numbers

Counted from the product source, not estimated.

9

Modules per engagement

Each one pre-scoped to a discipline. The full pass runs all of them.

6

Dispatch waves

Ordered by dependency. Wave 1 opens with 4 modules at once.

3

Zero-retention gated

Tax, legal and financial never run on a model that retains data.

8

Hard-block conditions

Conditions that stop a draft from being auto-approved, however many retries it has had.

Three other ways to do this

Compared as categories of tool, not as named products. Marks answer one question each: does this category of tool do this thing at all.

Virtual data roomAI diligence toolsGeneral-purpose AIFactum
Pre-scoped module per disciplineVirtual data roomNoAI diligence toolsPartlyGeneral-purpose AINoFactumYes
Runs without being promptedVirtual data roomNoAI diligence toolsPartlyGeneral-purpose AINoFactumYes
Citation required by the schemaVirtual data roomNot applicableAI diligence toolsPartlyGeneral-purpose AINoFactumYes
Cross-discipline reconciliationVirtual data roomNoAI diligence toolsNoGeneral-purpose AINoFactumYes
Produces the deliverable, not an answerVirtual data roomNoAI diligence toolsPartlyGeneral-purpose AINoFactumYes
Zero retention at the model providerVirtual data roomNot applicableAI diligence toolsPartlyGeneral-purpose AINoFactumYes
Human sign-off before exportVirtual data roomNot applicableAI diligence toolsNoGeneral-purpose AINoFactumYes
Published accuracy auditNobody in this table has one, including us.Virtual data roomNoAI diligence toolsNoGeneral-purpose AINoFactumNo
  • Yes
  • Partly
  • No
  • Not applicable

Categories, not vendors. Named products change quarterly and most public claims about them are vendor-published.

What we don’t claim about the platform

Everything above describes a mechanism. None of it is a certified metric.

  • —No audited accuracy rate. Internal benchmarks are single-dataset and unaudited.
  • —No claim that the human gate is optional. It is mandatory.
  • —No completed public case studies yet.
  • —Factum does not perform product-level technical due diligence, source-code review, architecture assessment or scalability testing.
  • —Factum does not perform integrity or reputational screening, or background checks on people and counterparties.
  • —No jurisdiction-free legal and tax read. Both are tuned to Dutch law; for a target elsewhere, have local counsel check them.
  • —No general availability. Factum is pre-launch.
What AI cannot do

Questions we get

What do you mean by AI due diligence?

Due diligence on the company you are buying or selling, carried out with AI. Factum’s AI reads that company’s data room, every document in it. Each finding cites its source document and page. The company’s own use of AI is one of the disciplines Factum covers. Your own reviewer checks each finding and signs off.

Why use AI at all?

It reads every document in the room instead of a sample, and finds what links one document to another. The arithmetic is done by code: recalculations across the statements, the LBO model and the sell-side checks run without the language model. A number the model writes that is in none of the quoted passages is held back. People decide: your reviewer checks and signs off.

Isn’t this just an AI chatbot with extra steps?

A general-purpose AI model is a Swiss Army knife: it does a bit of everything, including reading a data room. Factum is a surgical scalpel: built for exactly this one cut. This is 9 pre-scoped modules in a fixed dependency order, with four grounding checks and a mandatory human gate. The difference is what happens when nobody is watching.

Why waves instead of running everything at once?

Tax cannot reason without the financial and legal reads. Valuation cannot price without tax. The dependency graph sets the order; wave 1 still runs 4 modules in parallel.

What happens when the model doesn’t know?

A finding carries its evidence and review status. Evidence gaps remain visible. Fabricated sources are one of the 8 hard-block conditions.

How do I know the findings hold?

You check them. Every finding carries the passage it rests on, and your own reviewer signs off before anything is exported. There is no audited accuracy rate yet, and we do not imply one.

How fast is it?

The pipeline pass runs in hours. Your team’s review comes after it, at your pace. We don’t quote a date before seeing the data room.

How often does AI just cite something wrong?

AI can misattribute sources. Factum combines citation, quotation and entailment checks with targeted second opinions. Your reviewer checks the result.

What deal size is this built for?

For the deals and add-ons where a full DD process normally doesn’t pencil out. Once it holds there, scaling to any deal size is a matter of more documents, not a different system.

See it run on your own room

Share a room under NDA. We run the full pass and follow each finding back to its source passage with you, on a call.

NDA first, sent the same day. A deal you closed last year works too.

The Findings Guarantee

Three material findings your team did not already have, or the engagement is free.

For pilot mandates only. The guarantee does not apply to demos.