Use Case · Family Law

AI document classification for family law discovery.

BatesFlow's document classifier reads each PDF in a discovery production — bank statements, tax returns, deeds, emails, text exports, photos — and tags it by type, party, date range, and which numbered request from opposing counsel's Demand it answers. A typical 4,000-page matrimonial production is classified, organized, and Bates-ready in about ten minutes of paralegal review.

Updated May 8, 2026 · 2 min read · Use case

The problem

A divorce production is the messiest discovery in legal practice.

Other litigation arrives pre-organized — invoices in one folder, contracts in another. Family-law discovery does not. Clients hand you a ZIP of everything: years of bank PDFs, brokerage statements, joint tax returns, kids' tuition receipts, screenshots of texts, photos of jewelry, deeds. A junior paralegal then sits with all of it and asks one document at a time, which numbered request does this respond to?

8,400

Median documents in a contested NY divorce production

~7 hrs

Paralegal hours to manually identify and route them

12 min

With BatesFlow's AI classifier, including paralegal review

How it works

Three passes. One reviewer. Court-ready output.

BatesFlow's classifier is purpose-built on the document mix that shows up in matrimonial discovery, not the contracts-and-invoices mix that general legal AI tools were trained on.

Pass 01

OCR & document boundary detection

Every page — including scanned bank statements and photos of receipts — runs through Claude Vision OCR. The model detects where one document ends and the next begins, so a 432-page mixed PDF becomes 38 distinct documents.

Pass 02

Document type classification

Each document is tagged: bank_statement, brokerage_statement, tax_return, deed, credit_card, retirement_account, and 24 more. Categories are tuned for matrimonial practice, not generic eDiscovery.

Pass 03

Routing to discovery requests

The classifier reads opposing counsel's Demand for Discovery & Inspection, then maps each classified document to the numbered request it responds to. A 2024 brokerage statement routes to "Request 14: Investment Accounts." A matrimonial agreement surfaces for human review.

Inside BatesFlow

Review the classifier's work in one screen.

Every classified document shows you the AI's pick, its confidence score, and the reasons it routed the way it did. Override any decision with one click — the model learns within the case, not across cases. By the time you click Generate production, you've audited what the classifier got right and corrected what it didn't, and you have a deterministic record of every decision in the Bates Index.

Compared to general legal AI

Family law has different documents than BigLaw.

General legal AI tools — Harvey, CoCounsel, Spellbook — are extraordinary at contracts, briefs, and case law. They are not built around discovery production from a divorce practice. eDiscovery platforms — Everlaw, Relativity, Logikcull — handle the volume but cost like corporate-litigation tools and assume coding panels of attorneys. BatesFlow is the only tool whose document taxonomy and routing logic are defined inside the working matrimonial practice it was built in.

Run the classifier on a real Smith v. Smith case.

The sandbox is preloaded with a synthetic 3,247-document divorce case. No signup. Click in, watch it classify, browse the result.

Frequently asked

Questions lawyers ask.

What does AI document classification actually do?
It reads each document in a discovery production, decides what kind of record it is (bank statement, brokerage statement, deed, court order, retirement account, correspondence, etc.), and where appropriate maps it to a numbered request from opposing counsel's Demand for Discovery & Inspection. You get a fully sorted production with a path to a Bates-stamped output, ready for review.
Can it handle scanned bank statements and photos?
Yes. BatesFlow runs OCR (Claude Vision) on every page first, then classifies based on the extracted text. Scanned multi-page documents are detected as single records. Photo evidence (text exports, screenshots, photographs of correspondence) is OCR'd and classified just like native PDFs.
How accurate is the classifier?
On matrimonial productions of the type BatesFlow is built for, the classifier surfaces the right document type with high confidence on the majority of records, and surfaces a confidence score on every classification so a paralegal can spot-check the low-confidence ones in seconds. We do not claim perfect accuracy — every classification is reviewable in one screen, and a paralegal stays in the loop.
Does my client's data ever leave my firm's tenant?
Documents stay inside your firm's isolated tenant. The classifier runs on a per-firm boundary; operators cannot see case data. We do not train models on customer data.
How does this differ from Everlaw or Relativity?
Everlaw and Relativity are corporate-litigation eDiscovery platforms with coding panels — they are designed for BigLaw teams with dozens of reviewers and seven-figure matters. BatesFlow is purpose-built for family-law solo and small-firm practice: the document mix is matrimonial-specific (bank statements, brokerage, retirement, real estate, custody records), the workflow ends at a court-ready Bates production, and pricing is sized to a divorce practice rather than a class action.

See also