OCR Software for Legal Documents: A Requirements List

OCR software for legal documents has to accept the files your office receives at their real size, read page furniture such as line numbers, footnotes and stamps without mixing it into the text, and export the format each task needs. Write those needs down from your own files before you open a single vendor page.

A requirements list is one page. It records what goes in, what must come out, and which items would rule a product out. With it, a sales demonstration becomes a series of yes-or-no questions. Without it, you end up judging products on whatever each vendor chose to show you.

Write the list from a real week of intake. Count deposition transcripts, medical records, contracts, and discovery responses, and note which ones a paralegal must search the same day. Ask the attorney which export the case management system can store. A requirement that names those files will knock out a tool that only handles short letters.

What makes legal documents harder on OCR software than office paperwork?

Legal documents are harder on OCR software because the page carries more than its sentences. Line numbers, footnotes, stamps, captions, exhibit stickers and Bates numbers sit beside or on top of the text, files run to hundreds of pages, and a single wrong digit in a date or an amount can change what the document means.

An invoice or a staff memo is a single block of print on a clean sheet. Legal paper differs in five ways that belong in your thinking before you list anything.

Furniture around the text

A deposition transcript numbers every line down the left margin. A pleading may do the same, depending on the format rules of the court it is filed in. A brief carries footnotes in smaller type at the foot of the page. A produced document has a Bates number in one corner and perhaps a confidentiality legend in another. OCR software reads all of it. The question for your list is where each piece ends up in the output: in the running text, in its own place, or lost. The post on which structure survives OCR explains what happens to headings and numbering.

Marks on top of the text

A received stamp, a filing stamp, a clerk's initials, a highlighted passage and an exhibit sticker can each cover words. Recognition under a mark is a general weak point of OCR, so a firm whose intake is heavily stamped should say so on its list and test for it.

Copies of copies

A lease signed in 2009, faxed to a lender, photocopied for a file and scanned last week has been through four machines. Each pass costs sharpness, and the page you receive is the last copy in the chain, not the first.

Tables that carry the evidence

Billing ledgers, payment histories, rent schedules and account statements are tables, and the figures in them are the point of the document. Text that reads correctly but loses its columns is of little use for these pages.

Length

A closing binder, a medical chart or a production volume can run to hundreds or thousands of pages in a single file. Software sized for a ten-page letter may refuse the upload or make you split it by hand.

Turning a firm's document mix into OCR software requirements

Turn a document mix into requirements by counting what arrived over a recent period, sorting it into a handful of types, and writing one line per type that states the file, the task and the result you need. The list comes from your intake folder, and no vendor's feature page should shape it.

Step by step

1. Pick a window. Take the last 30 days of scanned material that reached the office, or the last three matters you opened if intake is uneven.
2. Sort into types. Use names your staff already use: transcripts, pleadings, signed agreements, medical records, bank statements, correspondence, handwritten notes, productions from the other side.
3. Count pages per type. A rough count is enough. File properties give you the page count of each PDF.
4. Record the largest file of each type. Write down its page count and its size in megabytes.
5. Record how each type arrives. Scanned at your own copier, received as a fax, emailed as a phone photo, downloaded from a portal.
6. Name the task. For each type, write what a person does with it next: search it, quote from it, edit it, total its figures, load it into another system.
7. Write the requirement. One sentence per type that joins the three: the input, the task and the output.

A worked example

Take a fictional four-person firm, Okafor Reyes LLP, that handles landlord and tenant disputes. Its last 30 days look like this.

Each row becomes a requirement:

• Accept a scanned PDF of at least 64 pages and 9 MB, and return editable text with section numbers attached to their clauses.
• Accept a faxed ledger and return its rows and columns as a spreadsheet.
• Return a searchable PDF that leaves the page image of a court order unchanged.
• Accept a JPEG photo of a single page.
• Read handwritten amounts well enough to be worth checking, or accept that these pages are typed by hand.

The figures are invented to show the method. The fifth line matters as much as the rest, because it records a limit the firm accepts instead of a feature it hopes for.

Which OCR requirements are must-haves and which are preferences for a law office?

A must-have is a requirement whose absence stops a task you do every week or breaks a duty you owe a client. A preference saves time or money without stopping the work. Mark each line of your list as one or the other, and let any missed must-have remove a product from consideration.

Three tests for a must-have

Ask these of every line:

• Does the work stop without it? If the tenant ledgers cannot become rows, someone retypes 140 pages a month. That is a must-have for the example firm.
• Does a duty depend on it? Terms that keep client files out of model training, let you delete them, and say how long they are kept bear on confidentiality. Treat them as must-haves and read the post on whether it is safe to upload client documents to an OCR tool for the questions to ask.
• Is there no acceptable workaround? If a 64-page lease can be split into two files without harm, a 50-page limit is an inconvenience. If your productions run to 900 pages each, the same limit is disqualifying.

How the same requirement lands in either column

The right-hand column is not a list of things to ignore. Preferences break ties between products that pass every must-have. Rank them by how many pages or hours each one touches in your count.

Keep the must-have list short

A list with fifteen must-haves eliminates everything or nothing. If a line is on the list because it sounded sensible and not because a counted document type needs it, move it to preferences.

Intake requirements: file types, file sizes and page counts legal OCR software must accept

Intake requirements state the largest and most awkward file the software must take without help: the file types you receive, the size in megabytes of your biggest regular file, its page count, and the number of pages you send through in a month. Read them off your own count, then compare them with each product's published limits.

The four intake numbers

Add a fifth line if a single day's intake can exceed what a daily cap allows. A free tier may cap documents per day in place of pages per month, so check which kind of cap applies.

File types

PDF, JPEG and PNG cover scans from a copier and photos from a phone. TIFF turns up in productions and older archives, and HEIC comes off some phones. If a type you receive is not accepted, someone converts it first, and that step belongs in your cost. The post on image file types as OCR input covers the less common ones.

Reading a vendor's limits against your numbers

Published limits are specific, so the comparison is mechanical. As an illustration, Schema Law accepts a PDF, PNG or JPEG of up to 20 MB. Its pricing page lists a Free plan at 1 document a day and up to 3 pages per document, a Solo plan at 1,000 pages a month and up to 250 pages per document, a Team plan at 10,000 pages a month and up to 1,000 pages per document, and an Enterprise plan at 100,000 pages a month with no fixed page limit per document.

Against those lines, the example firm's 64-page, 9 MB lease and its roughly 610 pages a month fit the Solo limits, while a firm whose regular production volume is a 400-page file would look at the Team plan's per-document limit. Run the same comparison for every product you consider. A file that exceeds a size limit can sometimes be reduced or divided, and the post on converting a large scanned PDF online covers how.

Output requirements: the formats each legal task needs from OCR software

Output requirements pair each task with the file format that serves it: a searchable PDF for finding pages, Word for quoting and redrafting, a spreadsheet for figures, plain text for comparison and pasting, and a structured format when another system will read the result. List the formats your counted tasks call for and no others.

One upload, several formats

A single document can need two outputs. The tenant ledger attached to a lease is a spreadsheet job inside a Word job. Write on your list whether the software must export one upload in more than one format, because otherwise the same file is processed twice.

Formats can depend on the plan

Check which formats each price tier includes. Schema Law, for instance, exports Word, Excel, CSV, JSON, searchable PDF, plain text, Markdown and HTML, as its export formats section shows, and every plan includes all eight. A requirement for spreadsheet output therefore does not decide the plan there, and page volume does. Another vendor may hold a format back for a higher tier, so read each vendor's plan table for it.

Requirements about the recipient

If a court, an opposing party or a client dictates the format, the requirement is theirs and not yours. Copy the wording from the rule, the order or the engagement terms onto your list, and check it against your own court or agreement. The guide to choosing an export format by who receives the file goes through the common recipients.

A one-page OCR requirements list to hand to whoever does the buying

A one-page requirements list has five blocks: the document mix, intake limits, outputs, terms for client files, and practical constraints such as users and budget. Each line is marked M for must-have or P for preference, and each must-have is written so that a product either meets it or does not.

Copy the template and replace the bracketed parts.

OCR REQUIREMENTS: [firm or team name]            Date: [date]
Prepared by: [name]         Based on intake from: [period]

1. DOCUMENT MIX (pages per month)
   [type 1] ........ [pages]    arrives as [scan / fax / photo / download]
   [type 2] ........ [pages]    arrives as [...]
   [type 3] ........ [pages]    arrives as [...]
   Total ........... [pages]    Busy month: [pages]

2. INTAKE
   [M] Accepts: [PDF, JPEG, PNG, other]
   [M] Largest regular file: [pages] pages, [MB] MB, without splitting
   [M] Monthly volume: [pages] pages
   [P] Several files uploaded at once

3. OUTPUT
   [M] [format] for [task]
   [M] [format] for [task]
   [P] [format] for [task]
   [P] One upload exported in more than one format

4. CLIENT FILES
   [M] Written statement that uploads are not used to train a model
   [M] We can delete a document and what was read from it
   [M] Stated retention period, or one we set
   [P] [anything your firm policy adds]

5. PRACTICAL
   [M] Users: [number], on [per-seat / shared] terms
   [M] Runs on: [browser / Windows / Mac], installation [allowed / not allowed]
   [P] Budget: up to [amount] a month
   [P] Support: [email / phone / named contact]

KNOWN LIMITS WE ACCEPT
   [for example: handwritten receipts are typed by hand]

RULED OUT IF
   Any line marked M is not met.

How to use the page

Give the page to whoever will contact vendors, whether that is an office manager, a paralegal or an outside IT provider. Ask for each must-have to be answered with a link to the vendor's own page or a sentence in writing. The post on building a side-by-side OCR software comparison shows how to check vendors against the list on paper before anyone starts a trial.

Once two or three products pass every must-have, the preferences decide. The scorecard for comparing legal OCR software turns them into weighted scores. If your document mix leans strongly toward one kind of page, the guide to the best OCR for scanned legal documents explains which tool traits suit which kind of scan.

When to rewrite it

Redo the count when your intake changes: a new practice area, a client that sends everything as phone photos, a court that changes its filing format. A requirements list written for last year's files buys last year's tool.