AI Bankruptcy Petitions: Best Case vs NextChapter vs Custom

By Jude Lee · · Comparison

Paralegal and attorney reviewing client financial documents next to a laptop displaying bankruptcy schedules

Where the hours really go in a chapter 7 or 13 file

Ask a bankruptcy paralegal where the day goes and you rarely hear “filling in the forms.” You hear: chasing six months of bank statements, reconciling pay stubs against the means test period, reading a credit report line by line to populate Schedule E/F, hunting for transfers and payments to insiders, reconciling the client’s memory of what they own against what the documents show, and then re-doing chunks of it when the client sends three more PDFs the night before filing.

The forms themselves are largely solved. The Official Forms are published by the federal judiciary on uscourts.gov, and the means-test inputs — median family income figures and expense standards — are published by the U.S. Trustee Program. Petition software encodes all of that and keeps it current. That is not where AI earns its keep.

The unstructured pile of client documents is where it earns its keep.

The petition is a sworn document. That single fact decides which parts of the workflow an agent can own outright and which it can only ever prepare for a human.

The three layers of a bankruptcy AI stack

For a consumer bankruptcy shop, the stack is three layers, not one product:

  1. Practice/petition software — Best Case (marketed as Best Case by Stretto) and NextChapter are the two names most consumer firms weigh. Both have changed corporate hands over the years, so confirm current ownership, pricing, and AI roadmap in the vendor’s own documentation before you commit.
  2. A general AI assistant — Claude, ChatGPT, or Copilot, used for drafting correspondence, summarizing a long document, or explaining a transfer pattern. Often used informally and ungoverned, which is its own risk.
  3. Automation glue — Zapier/Make, or increasingly an agent connected to firm systems through MCP (the Model Context Protocol, an open standard for giving an assistant governed access to your data and tools).

Concretely, the work AI does well here: extracting figures from financial PDFs, drafting first-pass schedule entries, flagging gaps (“no statement for March”), preparing 341 meeting question lists, and summarizing a client’s document dump into a memo the attorney reads in five minutes instead of forty.

Petition platforms, extraction tools, and a custom agent

Best Case / NextChapter (off-the-shelf)
Owns the Official Forms, means-test math, plan calculations, creditor matrix formatting, and CM/ECF filing integration. Updated when forms and UST figures change — that maintenance burden is real and it’s theirs, not yours. Native AI features vary and evolve; check current product documentation rather than trusting a sales deck. Weakness: the messy inbound pile of client PDFs still lands on a human.
Custom document-intake agent
Reads statements, stubs, tax returns, and credit reports; normalizes them into structured fields; drafts schedule entries and flags inconsistencies; writes results into your petition platform or a review sheet. Strength: it attacks the part nobody has automated. Weakness: you own accuracy, prompts, evaluation, and upkeep — and it is worthless if your petition platform has no import path.

There is a third option that should be tested before either extreme: off-the-shelf document extraction. Generic bank-statement-to-CSV converters, financial-statement parsers built for lenders and accountants, OCR tools with table extraction, and intake or document-automation add-ons sold alongside practice-management systems all cover part of this job for a subscription fee and no build. Their limits are predictable: they produce generic transaction tables rather than bankruptcy schedule fields, they rarely parse credit reports well, and they don’t flag preference-sized payments or insider transfers because they know nothing about bankruptcy. But if 70% of your pain is retyping six months of deposits, a $50/month converter plus a good checklist may end the conversation. Buy the cheap tool first; build only if the gap that remains is genuinely firm-specific.

For most small firms, the sane architecture is: keep the petition platform, put extraction in front of it, and decide separately whether that extraction is a product or a build. That is the same conclusion we reached comparing AI for immigration forms across Docketwise, INSZoom, and a custom build — form engines are commodity; document ingestion is not.

A realistic agentic pipeline for petition prep

  1. Collect into one governed place

    Client uploads land in a single matter folder — your practice management system or document store — not scattered across email. The agent works only from that folder, so provenance is traceable.

  2. Classify and inventory

    The agent labels each PDF (pay stub, bank statement, 1040, credit report, title, lease) and produces a completeness checklist against your firm’s standard document list, naming exactly what’s missing and for which month.

  3. Extract into structured fields

    Gross pay by period, deposits and unusual transfers, recurring debits, account balances, listed creditors and account numbers. Every extracted value carries a citation back to the source page — non-negotiable.

  4. Draft schedule entries and flag anomalies

    Draft entries for the relevant schedules, plus a flag list: payments over the insider/preference thresholds, large pre-filing transfers, income that doesn’t reconcile with the stubs, undisclosed accounts appearing in statements. These are flags for a lawyer, not conclusions.

  5. Hand off to the petition platform

    Push structured data into Best Case or NextChapter via its import path, or produce a review sheet a paralegal keys. The forms engine still does the math and the filing.

  6. Human verification and signature

    Attorney and client review against source documents. Nothing files itself.

A repeatable, packaged instruction set for one job, run the same way every time, is what people mean by a “skill.” We walk through building one in AI skills for repeatable document review.

What has to stay human, and why the stakes are unusual here

Bankruptcy is stricter than most practice areas about attorney responsibility. Under 11 U.S.C. § 707(b)(4), a debtor’s attorney’s signature constitutes a certification tied to a reasonable investigation into the circumstances giving rise to the petition — and the schedules are verified by the debtor under penalty of perjury. Fed. R. Bankr. P. 9011 applies on top. Confirm the current text and your district’s local rules with the primary sources and, where a specific practice is in doubt, with a qualified professional in your jurisdiction.

Practically, that means:

The extraction failure modes on this work are specific and worth naming. Scanned or phone-photographed statements degrade OCR accuracy, and a misread digit in a balance column is invisible downstream. Credit reports routinely list the same underlying debt under an original creditor and one or more collection agencies, and a model summarizing tradelines will happily produce duplicate Schedule E/F entries — or drop the original and keep only the collector. Joint or shared accounts are a second trap: deposits from a non-filing spouse, a roommate’s rent transfer, or a Venmo reimbursement get read as debtor income and inflate the means test. Pay stubs invite a third, where year-to-date columns are picked up instead of the pay-period amount. None of these are exotic; all of them look plausible on a review sheet, which is exactly why the citation-to-source-page rule matters.

Confidentiality deserves its own check, because this workflow puts tax returns, bank statements, and credit reports — some of the most sensitive documents a client owns — through a third-party model. Before any tool touches real client files, read the vendor’s data-processing terms and confirm four things in writing: whether your inputs are used to train models (and how to opt out), how long prompts and uploads are retained and where, who at the vendor can access them, and whether a business associate–style or confidentiality addendum is available. Then confirm the tool sits inside your existing client confidentiality and engagement-letter posture, and check whether your state bar expects client notice or consent for this kind of disclosure. ABA Formal Opinion 512 (2024) addresses generative AI use by lawyers, including competence, confidentiality, and supervision; state guidance varies and some of it is newer. Verify with your own bar and, on close calls, with qualified counsel. Our oversight models for supervising AI agents map the review tiers.

Automate document chasing, not judgment

In a bankruptcy file, the slice of activity that generates most of the drudgery is document chasing and transcription from client PDFs. Automate that and leave the strategy, counseling, and courtroom work alone. That is our opinion rather than a measured finding, but it’s the pattern we’d test first — and it’s the one with the least downside if the tooling disappoints.

Costing it out without inventing numbers

There is no such thing as an “AI lawyer” with a price tag. What you can price is the stack. Petition software is typically a per-user or per-case subscription; a general AI assistant is a per-seat monthly fee; an extraction tool is a subscription; a custom agent is a one-time build plus usage and maintenance. Get current figures from each vendor — they move.

For the build decision, use a formula rather than a headline:

hours saved per case × cases per month × loaded staff cost
Monthly recovered capacity (plug in your own figures)
Illustrative formula
build cost ÷ monthly benefit
Months to payback
Illustrative formula
new cases capacity absorbs × average fee
Upside if recovered hours go to intake, not idle time
Illustrative formula

The third line matters most and is the one firms skip. Recovered paralegal hours only become money if they get reallocated to more files or better client service. If you can’t name where the hours go, the honest expected return is quality and turnaround, not revenue — still worth something, but price it accordingly. Our automation ROI walkthrough does this in more detail.

Where is your firm losing billable hours?

Get a free automation audit: we map your intake-to-invoice workflow and show you exactly what's worth automating — before you spend a dollar.

Get a free automation audit

Related guides