DocuDigDocument Intelligence
Built for attorneys & investigators

Like having a team of analysts for every case.

DocuDig reads your discovery, surfaces the findings that matter, and traces every claim back to the exact passage it came from — so you can walk into any deposition knowing where the evidence sits.

Request a demo See a sample analysis
SOC 2 Type II No training on your data
app.docudig.ai / holcomb-v-vance / findings
Holcomb v. Vance · Findings

Analyze the financial records of Vance Landscaping to determine whether business funds were used for personal expenses.

Evidence of unreported cash income

Strong support

Cash deposits exceeded reported revenue by approximately $184,000 across 2022–2023.

47 citations · 3 source categories

Personal renovations paid from business accounts

Mixed

$72,000 in disbursements may fund family-residence work; one vendor disputed by review.

18 citations · 1 challenge

Asset transfer to third party before filing

Contradicted

Hypothesis refuted — wires predated the filing by four months and matched a supplier invoice.

Rejected by independent review
How it works

Three steps between a box of documents and a defensible finding.

No prompt engineering. No training runs. You upload the file, describe what you're looking for, and read the report.

Step 01

Upload your discovery

Bank statements, text messages, tax returns, contracts — drop them in or pull from your document management system. Scanned PDFs are fine.

148 documents~1,820 pages
PDFFirst Republic statements 2021–2024
PDFBookkeeper ledgers Q1 '22 – Q1 '26
XLSXCustomer invoice export
PDFForms 1120-S · 2021–2024
+ 144 more
Step 02

Describe what matters

State the research question in plain English, the way you'd brief a junior associate. DocuDig decomposes it into testable hypotheses.

"Determine whether business funds were used for personal expenses and whether income was accurately reported during the marriage."
11 sub-questionsentities: persons, accounts
Step 03

Read the findings

A narrative report with strongly-supported findings, contested findings, and rejected hypotheses — each one traceable to the exact passage of the exact document.

9 findings generated
Strongly supported4
Mixed evidence3
Insufficient1
Contradicted1
Findings, not haystacks

A report, not a search bar.

DocuDig returns conclusions with evidence, not a ranked list of documents for you to read. Each finding is graded by the strength of the evidence behind it, in plain language.

  • Narrative executive summaryOpening paragraph you can paste into a brief, generated from the findings themselves — not from a template.
  • Evidence strength in plain language"Strong support", "Mixed", "Contradicted" — not a confidence percentage you have to explain to a client.
  • Exportable as a briefFormatted Word, PDF, or Markdown with numbered findings and footnote-style source citations, ready to attach to a motion.
Executive summaryThe analysis produced strong evidence that Vance Landscaping was used to route personal expenses through the business and that a meaningful portion of cash revenue was not reflected on the 2022 and 2023 returns.
Finding 01Evidence of unreported cash income
Strong support
Finding 02Personal renovations from business accounts
Mixed
Finding 03Recurring household expenses
Strong support
Finding 05 · rejectedAsset transfer before filing
Contradicted
Trace: Evidence of unreported cash income
①
Finding

Cash receipts exceeded reported revenue by $184,000

②
Supporting facts

412 cash deposits · 23 ledger entries labeled "R.V. personal" · reconciled against Form 1120-S Schedule K.

③
Source documents

First Republic statements 2022–23 · Bookkeeper Ledger Q3 '22 – Q2 '23 · Forms 1120-S (2022, 2023).

④
Exact passage · Bookkeeper Ledger p.14
"Aug 14 — Deposit $4,800 cash (R.V. personal) · reconciled to Acct. 4102. Aug 21 — Deposit $3,200 cash (R.V. personal) · no invoice on file."
Provenance you can hand to opposing counsel

Every claim traces to the exact passage.

DocuDig doesn't just cite the document — it highlights the line, on the page, in the file you uploaded. If a finding can't be traced to source, it doesn't make it into the report.

  • Four-step trailFinding → supporting facts → source documents → exact highlighted passages. Readable top-to-bottom.
  • No black boxIf we cannot show you where a claim came from, we don't make the claim. Unsourced conclusions are filtered before they reach you.
  • Point-in-time replayRegenerate the exact state of a finding as of any prior date — including which documents were in evidence at the time.
Independent adversarial review

A built-in opposition.

Every finding is challenged by a second analysis pass that actively looks for alternative explanations, confirmation bias, and evidentiary holes. Challenges are shown alongside the finding — not hidden in a log.

  • Structured objectionsEach objection cites the evidence it draws on and explicitly states how the finding is weakened, narrowed, or rejected.
  • Bias flagsWhen the review detects possible confirmation bias across multiple findings, the project is flagged on the home page — not silently passed.
  • Rejected hypotheses stay visibleFindings that didn't survive review are preserved as "rejected" with the refuting evidence — so you know what was considered and ruled out.
Personal renovations finding
Reviewed Apr 21, 2026 · 1 challenge
Weakened
Review objection

Meridian Hardscapes invoiced the business for legitimate commercial landscaping at two managed properties during the same months. Without separating per-invoice work orders, the $72,000 figure may include legitimate business expenses.

Effect: reduces attributable amount to an estimated $48,000–$62,000

Refuting evidence (rejected hypothesis)

Wire transfer record shows the $220,000 transfers occurred four months before the filing, addressed to a legitimate equipment supplier with matching invoice 41-2202.

Confirmed by review — 7 other findings survived adversarial review with no objections.
Why not just a chatbot?

A legal-grade system — not an LLM with a file upload.

The difference between a tool you can cite in a motion and a tool you have to double-check by hand.

Dimension
DocuDig
A typical LLM tool
Dimension
Hallucinations

What happens when the model makes something up that sounds plausible.

DocuDig Algorithmic weighting Claims are scored against independent extractions across the document set and down-weighted when they disagree. Unsupported statements never reach a finding — they are filtered by the weighting pass.
A typical LLM toolRisk A confident paragraph, a plausible-sounding citation, no way to tell which sentence came from which document until you read them all yourself.
Dimension
Self-criticism

Whether the system argues against its own conclusions.

DocuDig Automatic red-teaming at every layer Every stage of the pipeline — extraction, synthesis, and final findings — runs an adversarial pass whose only job is to poke holes in the output. Findings that survive are labeled "confirmed"; findings that don't are weakened or rejected in the open.
A typical LLM toolRisk The model argues for the answer it already generated. Confirmation bias is the default mode; disagreement has to be prompted by the user.
Dimension
Volume

How many documents the system can genuinely reason across.

DocuDig Unbounded document sets The pipeline decomposes, extracts, and reconciles across entire discovery productions. There is no single context window to overflow — 50 documents and 50,000 documents use the same machinery.
A typical LLM toolRisk Hard context limits. Long document sets are truncated, summarized, or silently ignored. The model never "sees" the whole file.
Dimension
Verifiability

Whether you can check the system's work step by step.

DocuDig Every step is inspectable Each finding is traceable to the facts that support it, to the documents those facts came from, and to the exact passages in those documents. You can audit any intermediate stage, not just the final answer.
A typical LLM toolRisk A paragraph of output with no way to see what the model considered, what it rejected, or what it never read.
Dimension
Transparency

Whether the process is visible or hidden behind a prompt.

DocuDig No black box Every hypothesis tested, every challenge raised, every source cited and every rejection is preserved in the project record. You can replay the reasoning that produced any finding, at any time.
A typical LLM toolRisk One opaque call in, one opaque paragraph out. Explaining the reasoning to a partner, a judge, or a client means retracing the work yourself.
In practice: DocuDig is a pipeline of specialized steps, not a single prompt to a general-purpose chatbot. See how the pipeline works

I used to lose a junior associate for three weeks every time a complex financial discovery landed. Now I get a defensible first pass before lunch — and I can put my finger on the line in the statement before opposing counsel can.

Dana MerrimanPartner · Merriman Reyes LLP · Matrimonial & Family Law
Where attorneys use DocuDig

Built for matters where the documents are the case.

Matrimonial & family law

Financial discovery

Reconcile bank records, returns, and ledgers to identify undisclosed assets, unreported income, and commingling — without losing an associate for three weeks.

  • Undisclosed asset identification
  • Commingling & dissipation analysis
  • Lifestyle analysis from transaction records
See example →
Corporate investigations

Related-party & compliance

Trace intercompany transfers and surface undisclosed relationships across years of general ledger entries, contracts, and board minutes.

  • Related-party transaction detection
  • Policy exception investigation
  • Whistleblower claim substantiation
See example →
Commercial litigation

Contract & performance disputes

Map obligation-to-performance across thousands of pages of emails, invoices, and service reports. Establish timelines with passage-level citations.

  • Breach-of-contract timelines
  • Most-favored-customer analysis
  • Warranty & service-level disputes
See example →
Insurance & fraud

Claim substantiation

Cross-reference claim narratives against medical records, police reports, and financial statements to identify inconsistencies and supporting evidence.

  • Claim-narrative consistency checks
  • SIU investigation support
  • Subrogation evidence packaging
See example →
Regulatory & compliance

Internal reviews

Test specific policy questions against the communications and records you already have on hand. Produce an auditable findings package in hours, not weeks.

  • Policy-violation screening
  • Retention & records audit
  • Sanctions & AML pattern review
See example →
Estate & trust

Probate disputes

Reconstruct distributions, trust transactions, and arm's-length determinations across decades of correspondence and account records.

  • Arm's-length transfer analysis
  • Fiduciary duty review
  • Decedent financial reconstruction
See example →
Security & privilege

Designed for privileged material from day one.

Your documents are yours. No training, no retention past your project, and tenant isolation at the storage and compute layers so each matter stays walled off.

SOC 2 Type II

Annual audit covering security, availability, and confidentiality controls.

No training on your data

Your documents are never used to train models — yours, ours, or any third party's.

Ethical-wall controls

Per-matter access controls with full audit log. Tenant isolation at the storage and compute layers.

Pricing

Priced by matter, not by seat.

Every person at your firm who touches the case can use DocuDig — including co-counsel and reviewing partners.

Solo & small firm

For firms of 1–10 attorneys handling a handful of complex matters at a time.
$1,200 /matter
Up to 500 documents per matter
  • Unlimited users per matter
  • Exportable reports (PDF, Word, MD)
  • Full provenance trails
  • Standard support
  • SSO & SCIM
Start a matter
Most firms

Firm

Annual subscription for firms running multiple complex matters concurrently.
$36k /year
Up to 40 matters · 2,000 documents each
  • Everything in Solo & small firm
  • SSO, SCIM, ethical-wall controls
  • Branded report templates
  • iManage & NetDocuments connectors
  • Priority support · dedicated CSM
Request a demo

Enterprise

For AmLaw 200 firms, in-house investigations teams, and regulators with strict data-residency requirements.
Let's talk
Custom matter volume & deployment
  • Everything in Firm
  • Dedicated VPC deployment
  • Custom data-residency region
  • Bring-your-own model
  • Named solutions engineer
Contact sales
Bring a matter

Walk us through a live matter. We'll show you the findings before the call ends.

45-minute walkthrough with a solutions expert. Bring a sanitized document set or use ours — we'll analyze it live and hand you the report.

Request a demoTour the product

What you'll see

Your own documents analyzed Bring a sanitized discovery set, or use the Holcomb v. Vance sample matter.
The findings walkthrough How attorneys read the report and hand it to co-counsel or the client.
A straight answer on fit If DocuDig isn't the right tool for your practice, we'll tell you in the call.
Common questions

Questions attorneys ask first.

DocuDig produces findings, not testimony. Every finding traces to the specific passage in your source documents, and those passages are what you put into evidence. The report itself is attorney work-product; the admissible artifact is the highlighted source document.

Findings that cannot be traced to at least one exact passage in a source document are filtered before they reach the report. An independent adversarial pass additionally challenges each surviving finding and can reject, narrow, or weaken it. No unsourced conclusions ever appear in an attorney's view.

Native PDFs, scanned PDFs (with OCR), Word, plain text, CSV and XLSX spreadsheets, and photographs of documents. Audio and video transcripts can be imported as text; we don't transcribe media directly yet.

No. Your documents are not used to train any model — ours, our providers', or anyone else's. This is contractual and enforced technically through the model integrations we use. Self-hosted customers can additionally route all inference through their own infrastructure.

A typical 150-document matrimonial matter completes in 10–20 minutes. A 2,000-document corporate investigation completes in 60–90 minutes. You can close the tab and come back — findings populate in real time as they're generated.

You do. The report, the trail, and any exports are your firm's work-product. DocuDig retains no rights to your findings or your source documents, and deletes both on request.