# AI due diligence and data-room review

Deep Discovery is a sales-led, scoped Nexus and Enterprise workflow designed to organize a large private corpus, retrieve decision-relevant evidence, reason across the selected material, and subject findings to adversarial review before a human relies on them. Capacity, coverage, extraction quality and deliverables require a validated signed scope.

## Search is not the same as judgment

A data room may contain thousands of individually plausible documents and only a handful of facts that reverse the economics, expose a control failure or invalidate a thesis. Keyword search finds mentions. Due diligence must explain why a fact matters, connect it to the decision and show the path back to source evidence.

1. **Cull:** remove duplicates and obviously irrelevant material without pretending the discarded set was deeply reasoned over.
2. **Retrieve:** find evidence against explicit diligence questions, criteria and risk hypotheses.
3. **Challenge:** assign distinct analytic seats to attack findings, search for counterevidence and preserve unresolved disagreements. Cross-provider composition does not establish statistical independence.

## Early case assessment and e-discovery boundary

| Layer | 3Dogs public position | Required control |
|---|---|---|
| Collection and preservation | Not claimed as a public 3Dogs capability | Use approved collection, legal-hold and chain-of-custody systems |
| Processing, OCR and deduplication | May be part of a validated Enterprise scope; no universal coverage claim | Record source set, transformations, exclusions, errors and versioned manifest |
| Search and retrieval | Designed to retrieve evidence against explicit questions and hypotheses | Preserve queries, citations, missing evidence and retrieval limits |
| Judgment layer | Designed to connect retrieved facts to a business decision and challenge findings | Keep run record, alternatives, dissent and what would change the conclusion |
| Privilege determination | Not claimed | Qualified counsel controls privilege review and waiver decisions |
| Legal opinion | Not claimed; Deep Discovery is decision support | Qualified counsel remains responsible |
| Source traceability | Required for a finding to be reviewable | Retain document and passage references plus transformation provenance |
| Human review | Required before reliance | Named qualified reviewer accepts, rejects or remands findings |

No public page claims corpus completeness, recall, cost savings, privilege accuracy or legal-grade output. The [due-diligence citation analysis](/research/ai-due-diligence-document-review-sources/) reports recorded sources for these queries; it does not validate the underlying answers.

## Published evidence, correctly labeled

**Reported synthetic demonstration:** the Reynolds record reports that eight planted risks were found in an approximately 10,000-page M&A room. The corpus, answer key, manifest and scoring script are not public, so an independent rerun is not currently possible. This is not a production customer outcome.

https://3dogs.ai/case-studies/reynolds-due-diligence/

**Historical public corpus:** in an Enron public-corpus reconstruction, 3Dogs reports 45,320 deduplicated messages processed and 981 documents fully read. Transformation, coverage and blind-review protocol are not independently reproducible. The record does not establish predictive certainty or the same result on another corpus.

https://3dogs.ai/case-studies/enron-investigation/

## Enterprise delivery and data boundaries

Document-room work is negotiated because corpus size, source systems, decision questions, security review, retention, model routing and deployment constraints change the engagement.

- Managed, dedicated, customer-cloud and in-perimeter designs remain subject to validation.
- Retention and deletion follow signed engagement terms. A zero-retention option is available only when expressly negotiated.
- 3Dogs says it does not train models on customer case content.
- 3Dogs does not publicly claim current SOC 2, ISO 27001, FedRAMP, CMMC or HIPAA certification and does not borrow a provider's certification as its own. Absence of a public claim is not proof that no certification exists.

## Limits

Deep Discovery does not replace counsel, auditors, accountants, investigators, investment committees or statutory decision-makers. Model output can omit, misread or overstate evidence. Findings must retain citations and be reviewed by qualified humans in the transaction context.

- Enterprise: https://3dogs.ai/enterprise/
- Trust Center: https://3dogs.ai/trust/
- Privacy: https://app.3dogs.ai/privacy
- Security: https://app.3dogs.ai/security
- Book a briefing: https://3dogs.ai/enterprise/#briefing