Use caseHuman resources and legal
Opposing counsel needs an 87-page disciplinary file full of colleagues' names, pay and health details. Today: a black marker and a scanner.
The AI suggests what to hideA person approves every markThe originals stay whole in the case
Colleagues' names, phone numbers, emails, pay, 4 health mentions and 3 bank account numbers, each on its page.
source · 5 documents on the HR caseAccepted by a person · hidden in the copy to share
A colleague's cell number in an email footer. A medical leave on page 63. A home address in a margin. A discovery production goes wrong on what it shows by accident, and a marker on paper misses some of it, while a black box drawn in a PDF editor can leave the text underneath.
The AI takes the first pass and people keep the decision. It suggests what to hide, page by page, with the kind of data it found. HR and counsel approve every mark, and the copy that leaves is produced in the viewer while the originals stay whole in the case.
Each screen on its own is a familiar tool. The safety comes from the order: the AI suggests, HR reviews, counsel approves, and only then is the copy to share produced. Kevin's Friday deadline sets the pace for the steps below.
Kevin Lambert has five documents to produce by Friday, 87 pages in all.
Kevin, the HR officer at Bertram Logistics in Columbus, opens the HR case of Derek Perkins, the former employee. He selects the five documents requested in discovery and starts the production step.
The AI reads the text of the documents and suggests 143 marks. It works from the context of each sentence, beyond a list of words: a phone number in a signature, a salary in a comparison, a medical leave mention.
Kevin opens page 12 of the investigation report. Each suggestion is outlined on the text with its kind next to it: colleague name, cell number, pay. He can accept it, or reject it with a reason.
The AI found most of it. Kevin and counsel decide what is hidden, mark by mark.
Kevin rejects the mark on Derek Perkins's name: he is the plaintiff. He rejects the HR director's name, since she will testify. For phone numbers he applies one recurring pattern across the 87 pages, and keeps the public main line visible.
On page 7 of the scanned witness statements, a handwritten note in the margin gives a colleague's home address. No suggestion covers it. Kevin draws an area over it. On the photo of the doctor's note, he covers the clinic stamp.
The production moves to Amira Haddad, the employment attorney. She approves the four health mentions. She rejects one pay mark: Derek Perkins's own salary is at the heart of his claim and must stay readable. Then she approves the production.
The approved marks become a copy to share. The originals stay as they were.
Kevin produces the copy to share: one PDF of 87 pages. The approved text and areas are removed from this version, and the masks travel with the export. The five originals stay whole in the case.
The rights on the case decide who may share what. Kevin can prepare and redact. Only Amira can share the copy, and nobody can share the originals out of this case. She sends the copy to opposing counsel.
In the case, the originals stay as they were and the copy to share is filed as a document of its own. The history lists every step: the suggestions, each decision, the marks drawn by hand, the approval and the sharing.
Every mark is approved, rejected, or held for counsel. That state decides what the copy that leaves will show.
A person accepted the suggestion, or drew the mark by hand.
After that: the mask travels with the export; the original keeps the text.The plaintiff's name, a witness for the company, a public number.
After that: the reason is written next to the decision, for the record.Health and pay are not decided by HR alone.
After that: the copy cannot be produced until counsel approves.It hides nothing on its own and sends nothing. Across the nine steps on this page, every mark is accepted or rejected by a person, and counsel approves before the copy is produced.
The AI reads the text and suggests, with the kind of data each time. The control is in the process: the approval step and the rights on the case, so it holds on every production.
From the Perkins production, 87 pages. Your own files and kinds of data will set different counts.
Where these numbers come from. They are counted on the scenario described on this page, with example data. They are not a customer result.
HR, counsel and IT each put their name to a production. This is what each of them can hold onto.
A black marker, a scanner and a Friday deadline.
Signing a production you could not check line by line.
Redaction happens in desktop tools, outside any record.
With one kind of production. Take the files you send most often to courts, auditors or partners, and run the suggestions on the next one.
It works on the HR cases and documents you already store.
Your move
Bring an anonymized file you had to produce recently. We will run the suggestions, walk through the approvals and produce the copy to share.