Scanned mail, filed to the right matter.

Mailroom reads your firm's incoming physical mail and files each document to the correct client matter.

Summonses · Complaints · Discovery · Medical bills

How Mailroom works, in four steps

1

Mail arrives

Your own mailroom or a virtual one — Mailroom doesn't care which.

2

Scanned or photographed

Your existing scanner, a phone photo, or the PDF your virtual mailroom already sends you.

3

AI matches the case

It reads the case number, parties and sender — including a number stamped on by the clerk — and weighs them against every open matter.

4

Filed into your system

Straight into the matter in Clio, MyCase or whatever your firm runs on.

Secondsto read a document and file it
Within the hourevery deadline on the page becomes a dated task
No third-party AIyour documents are not sent to an outside model

Measured on a local machine over a test set of seventeen litigation documents — summonses, complaints, discovery, medical bills and collection notices.

Accuracy

Three outcomes, not two

The failure mode that matters isn't a document nobody filed. It's a document filed confidently into the wrong client's matter.

How each confidence tier is handled
ConfidenceWhat happens to the document
High Filed automatically.

A case number read off the page is decisive on its own. Most of the daily mail lands here and nobody touches it.

Moderate Filed, badged, and queued.

Matched on party names rather than a docket number. It's in the matter so it can be found, but marked as awaiting confirmation so nobody mistakes a suggestion for a filing.

Low Filed nowhere.

It sits in the queue with suggestions and never appears in any client's file until a person chooses. A weak guess is not a filing.

How it works

Scan it. Everything else is automatic.

The mail goes through your existing scanner. Mailroom does the rest, and asks a human only when it genuinely cannot tell.

Read the page

Text is lifted from the document where it exists, and read optically where it doesn't — including the case number a court clerk applies as a stamp, which plain text extraction cannot see.

Weigh the signals

Case number, client name, opposing party, recipient address and sender are each scored against every open matter. No single weak signal is enough on its own.

File it, or ask

A clear match files itself. Anything less goes to the review queue with the document, its ranked suggestions and the reasoning behind each one.

The hard cases

What plain text extraction gets wrong

These aren't edge cases. Every one of them came out of real case files.

Case numbers applied as a stamp

An e-filed complaint has a perfect text layer for everything typed into the form — and the case number added afterwards as a stamp image, which text extraction reads as an empty field. Mailroom reads the stamp.

Form codes that look like case numbers

PLD-C-001 and CM-200 identify the blank Judicial Council form, not the case. They are never mistaken for a docket number, and the real one is recovered from the page.

Mail carrying no case number at all

Pre-litigation letters, collection notices and medical bills rarely cite a docket. Those are matched on your client together with the opposing party — the pair that separates two matters for the same client.

Two clients with the same name

When more than one client could be the person named and the document cites no case number, Mailroom refuses to choose. Filing one client's summons into another client's file is the one error it will not make.

Learns from each filing

File it once, and it files itself after that

Most mail without a case number still carries something durable — an adjuster's claim number, a records request number, an account reference.

First letter

An adjuster writes about a claim. There is no docket on the page, so it goes to the review queue. A paralegal files it to the matter — one click.

Every letter after

The claim number and the adjuster are now handles for that matter. The next letter in that thread files itself, with no one touching it.

Nothing is configured and no model is retrained — the system records what a person decided and applies it next time. Month one is mostly review. Later, only genuinely new correspondents reach the queue.

Confidentiality

Your documents never go to an AI company

Not OpenAI, not Anthropic, not Google. Mailroom runs its own models, so client files stay confidential and never become someone else's data.

Inside your environment

Scanyour scanner
Readour own models
Fileyour matter system

Questions

What firms ask first

What happens when it gets one wrong?

That's the whole design. A match made on anything weaker than a case number is flagged rather than filed silently, and a match too weak to defend isn't filed at all. Correcting one is a single click from the queue, the client file, or the document itself.

Does it work with our document or practice management system?

Yes — Mailroom files into your DMS or practice management software through their APIs: Clio, MyCase, and any of the DMS platforms out there. Tell me what you run.

Do we need new scanners or new hardware?

None. Your existing scanner is fine, a phone photo works, and if a virtual mailroom already scans your post, Mailroom takes those PDFs directly. Nothing to install and nothing to buy.

Is a paralegal still reviewing the mail?

Yes, but only the ambiguous ones. A confident match files itself.

See it on your own mail

The demonstration that's worth your time is on documents you recognize. If filing the mail is a real cost at your firm, get in touch.

Request a demo