# OCR for Legal Operations: Buying and Rolling It Out

- Author: Schema Law (https://www.schemalaw.com/)
- Published: 2025-11-22T18:57:43Z
- Updated: 2026-10-11T19:22:20Z
- Category: Choosing legal OCR
- Canonical: https://www.schemalaw.com/blog/ocr-for-legal-operations

> How legal operations teams evaluate, buy and roll out OCR: the business case, a vendor security review, a four-week pilot, usage reporting and adoption.

OCR for legal operations is a buying and rollout project in five parts: place the tool at the point where scans enter the department, build a case from counted pages and hours, run the vendor security review with the company's own security team, pilot with real documents across teams, then report usage and keep people using it.

This post is for the person who owns the legal department's tools and budget, whether or not "legal operations" is in the job title. It covers the process. What the lawyers do with the converted documents is in the companion post on [OCR for in-house counsel](https://www.schemalaw.com/blog/ocr-for-in-house-counsel).

Place the rollout where contracts enter contract lifecycle management and where invoices enter e-billing review. Ask each team which scans they retype into matter management, and whether outside counsel guidelines already name an approved converter. Add the tool to your conflict-of-interest database process only as a record of who may upload, and state the duty of confidentiality in the rollout note every user sees.

## Where does OCR sit among a legal department's other tools?

OCR sits at the front door of a legal department's other tools. It turns a scan into text before the document goes into whatever stores, tracks or analyzes it. A contract repository, a matter system, a spreadsheet tracker and an e-billing review all work better on text, and none of them is where the scan first lands.

Think of the department's systems in generic layers:

| Layer | What it does | What it needs from a scan |
| --- | --- | --- |
| Intake | Receives requests and documents from the business and from outside | Nothing yet; this is where scans arrive |
| Conversion | Turns page images into text and tables | The scan itself |
| Storage | Holds the department's documents under its permissions | A searchable file, so full-text search finds it |
| Tracking | Matters, contracts, deadlines, holds | Dates, names and numbers typed or pasted in |
| Analysis | Spend review, contract review, reporting | Structured data such as spreadsheet rows |

### Overlap to check before you buy

Some storage and contract systems recognize text on their own when a file is added. Some do it only for search and never give you the text back as a Word or Excel file. Before you write a business case, list what each existing system already does with a scanned PDF, and test it with three bad scans. You are buying for the gap, and you need to be able to describe the gap to finance.

### Decide what OCR is not

Write one sentence in the project charter saying what this purchase will not do. It converts documents. It does not review contracts, manage matters or decide privilege. A stakeholder who assumed otherwise should learn it from the charter, not from the pilot.

## Building the business case for OCR in a corporate legal department

Building the business case for OCR in a corporate legal department takes three numbers you measure yourself: pages of scans arriving per month, hours spent retyping or reading unsearchable files, and what outside counsel charges to make your scans usable. Compare those with the plan price. Do not borrow anyone else's statistics.

### Count before you argue

Run a two-week count. Ask each team to log every scanned document it receives: date, type, pages, and what had to be done with it. A shared spreadsheet is enough. At the end you will know volume by team and which document types cause the work.

### Three places the cost shows up

1. **Retyping.** Someone keys subpoena requests into a tracker, or a counterparty's paper terms into Word. Time it for five documents and multiply.
2. **Search by reading.** A lawyer opens forty signed agreements one by one to find a clause. Ask two lawyers how often that happened last quarter.
3. **Outside counsel preparation.** Firms that receive flat scans convert them and may bill for the handling. Ask your main firms what form they would prefer to receive and whether conversion appears on invoices.

### A worked example with invented numbers

Suppose the fictional Corvane Industries legal team logs 1,400 scanned pages in two weeks, so about 3,000 a month. Paralegals report roughly nine hours a month retyping from scans. At an internal cost the finance team sets at $55 an hour, that is $495 a month in retyping alone, before search time or firm invoices.

Against that, a plan sized for the volume has a known price. Using Schema Law's published plans as the example, 3,000 pages a month is above the Solo allowance of 1,000 and inside the Team plan's 10,000, listed at $199. Each account is for one person, so that price assumes one paralegal converts for the department. Three people who each need to convert would need three accounts: $147 on Solo if each stays within 1,000 pages, or $597 on Team. If the company's rules require a completed security questionnaire and contract review, the comparison is with the Enterprise plan at $999 instead. Use your own counts and your own vendor's prices; the method is the point. The post on [OCR cost per page](https://www.schemalaw.com/blog/ocr-cost-per-page#estimating-your-monthly-page-volume) shows how to forecast volume more carefully.

### What to leave out

Leave out soft claims about risk reduction unless you can tie them to an event that happened, such as a response date nearly missed because a faxed notice sat unread. One real incident persuades a budget owner more than a paragraph of generalities.

Put the checking time in. Converted text has to be verified, and a case that pretends otherwise will be found out in the pilot.

## What should legal operations ask in an OCR vendor security review?

In an OCR vendor security review, legal operations should ask what happens to an uploaded document at each stage: who can see it, whether it trains a model, where copies persist, how deletion works, and how users are admitted and removed. Ask in writing, with the company's security team as the reviewer of record.

When a corporate legal department buys, the review differs from a law firm's in one way that matters. A firm protects many clients' documents. You protect one client's, and that client may already have an information security function, a vendor onboarding process and a standard questionnaire. Where it does, your task is to get legal's requirements into that process, not to run a parallel one.

### Questions that belong to legal

The company questionnaire covers general security. These are the ones a legal department must add, because a general questionnaire may not ask them:

| Question | Why legal asks it |
| --- | --- |
| Are uploaded documents or their text ever used to train or improve a model? | Privileged and confidential material must not become training data |
| Can any other customer, or any vendor staff, see our documents? Under what circumstances? | Disclosure outside the privilege |
| When a user deletes a document, what is removed and when? | Retention schedules and legal holds |
| Can retention be set by policy instead of by each user? | Consistency across the department |
| How are users added and removed? Is there single sign-on tied to our directory? | A leaver must lose access the day they go |
| Can documents be separated by matter or team inside our account? | Investigations and HR matters kept from the rest of legal |
| Will you complete our security questionnaire and negotiate our data terms? | Some vendors offer this only on a top tier |
| What happens to our documents if we cancel? | Exit |

### Read the published answers first

A vendor may answer some of this on its site. Schema Law, for instance, lists the same [security commitments](https://www.schemalaw.com/#security) for every plan, among them "Your documents are never used to train an AI model", "Documents stay inside your own account", "Encrypted in transit and at rest" and "Delete any document, at any time". Its plan list places a security questionnaire and contract review on the Enterprise plan. So if you need OCR with a security questionnaire completed, you are pricing that tier, and the business case should say so.

Published pages can also answer a question with a no. If you need OCR with single sign-on, Schema Law does not offer it on any plan: sign-in is with a Google account, one person per account. It has no retention setting, since a document stays until its owner deletes it, and no division by matter or team inside an account. A department that requires any of those must meet the requirement by its own procedure or choose another tool.

Do not infer anything a vendor has not written down. If the site is silent on where data is stored or which subprocessors are involved, the question goes in the questionnaire and the answer goes in the contract.

### Keep the roles straight

Security reviews the controls. Privacy reviews the data terms. Legal decides whether the answers are good enough for privileged material. Procurement runs the paper. Legal operations keeps the list and the calendar. How questionnaires and contract review work on an enterprise purchase is covered in the post on enterprise OCR, in its section on [security questionnaires and contract review](https://www.schemalaw.com/blog/enterprise-ocr-for-law-firms#security-questionnaires-and-contract-review).

For renewal mechanics, notice periods and price changes, read the subscription post's section on [renewal, cancellation and price-change clauses](https://www.schemalaw.com/blog/legal-ocr-subscription#renewal-cancellation-and-price-change-clauses-in-an-ocr-subscription) before you sign an order form.

## Running an OCR pilot across legal department teams

Running an OCR pilot across legal department teams means four weeks, three or four teams, a fixed set of real documents, and pass criteria written before anyone uploads. The pilot answers two questions: does the output hold up on your documents, and will your people use it without being chased?

### Before the pilot starts

Get the security team's approval for the pilot's scope. A pilot with real documents is real use. If full approval is pending, restrict the pilot to documents that are already public or that carry no privilege, and say so in the plan.

### A four-week plan

1. **Week 0: assemble the test set.** Ask each team for ten documents that caused trouble: a faxed notice, a sideways signature page, a price schedule, a contract with handwritten changes. Thirty to forty documents in all.
2. **Week 1: controlled test.** One person per team converts the test set and scores each document against the scan. Use a simple scale: usable as is, usable with fixes, not usable.
3. **Weeks 2 and 3: live use.** Each team uses the tool on documents as they arrive and logs each one: type, pages, export used, minutes spent checking.
4. **Week 4: review.** Collect the logs, interview two users per team, and decide.

### Who to include

| Team | Documents it brings | What it tests |
| --- | --- | --- |
| Commercial contracts | Counterparty paper, executed agreements | Word output, searchable archive |
| Litigation and disputes | Subpoenas, demand letters, old case files | Fast intake, text for trackers |
| Employment | Personnel files, agency charges | Mixed typed and handwritten pages, privacy handling |
| Corporate and compliance | Minutes, filings, regulator letters | Tables, older documents |

Include one skeptic. The objections of someone who expected to dislike the tool are the ones the rollout will have to answer.

### Pass criteria, written first

Write three or four statements and get the general counsel to agree to them before week 1. For example: at least eight in ten test documents rated usable as is or with fixes; logged checking time lower than logged retyping time from the two-week count; no team reports a document type it cannot process at all; security approval granted for full use.

A fair trial depends on your own documents, and the post on choosing OCR for lawyers has a section on [running your own test on real documents](https://www.schemalaw.com/blog/what-is-the-best-ocr-for-lawyers#running-your-own-test-on-real-documents) that applies here too.

## Which OCR usage numbers should legal operations report?

Legal operations should report five OCR usage numbers: pages converted against the plan allowance, active users against licensed users, documents by team, exports by format, and time saved as estimated from samples. Report monthly for the first quarter, then quarterly. Keep the estimate labeled as an estimate.

| Number | Where it comes from | What it tells you |
| --- | --- | --- |
| Pages converted vs. allowance | The vendor's usage view or invoice, or your own log | Whether the plan is the right size |
| Active users vs. people with access | Sign-in records or your log | Adoption |
| Documents by team | Your log, or workspaces if the tool divides by team | Who is getting value and who is not |
| Exports by format | Your log | Whether people use it for search only, or for Word and spreadsheet work too |
| Hours saved (estimate) | Timed samples from the pilot multiplied by volume | The return you promised in the business case |

### Where the data comes from

Ask in the vendor review what usage reporting the tool gives an administrator. If the answer is thin, keep a light log of your own: a form with four fields that users fill in when they convert something for a matter. A log people fill in beats a dashboard nobody opens.

### Numbers to avoid

Do not report an accuracy percentage. You have no sound way to measure one across mixed documents, and a single figure invites people to stop checking. Report instead how many documents needed fixes, by type, which points at where scans should be improved at the source.

Do not report raw page counts alone. Thirty thousand pages converted and never searched is storage, not value.

### One number for the general counsel

Pick a single line for the quarterly update, such as "contracts in the executed archive that are full-text searchable: 2,140 of 2,600." It is concrete, it moves, and it connects the tool to something lawyers feel.

## Rolling OCR out to a legal department and keeping people using it

Rolling OCR out to a legal department works when conversion becomes a step in routines people already follow, not a new destination to remember. Name an owner, write a one-page guide, train in twenty minutes with the team's own documents, and check usage at thirty, sixty and ninety days.

### A rollout checklist

- **Name an owner and a backup.** Someone answers questions and removes leavers.
- **Set up access.** Accounts for named people, tied to the company directory if the tool offers single sign-on.
- **Agree the folder and naming rules.** Where converted files are saved, and how the original and the converted copy are paired.
- **Write the one-page guide.** When to convert, which export for which task, what to verify, what never to upload without asking.
- **Train by team.** Twenty minutes, live, on a document that team received last week.
- **Put conversion into intake.** The person who logs incoming paper converts it at the same time.
- **Publish the rules on privileged and restricted matters.** Who may upload investigation or personnel files, and where the output lives.
- **Schedule the three check-ins.**

### Why people stop

They stop when the first result disappoints and nobody told them what to expect. Say plainly in training that recognition is weaker on faxes, stamps over text and handwriting, and that names, dates and amounts get checked against the scan every time.

They stop when the tool sits outside their path. If a contracts manager must leave the repository, sign in elsewhere, convert, download and re-upload, the habit will not last. Make the route as short as the tools allow, and give intake staff the conversion step so lawyers receive documents already searchable.

They stop when the champion leaves. That is why the owner has a backup and the guide is written down.

### After ninety days

Compare the usage numbers with the business case. If one team is not using the tool, ask what its documents are; the answer is often that its scans are a type the tool handles poorly, or that it already had another way. Resize the plan to the measured volume at renewal, up or down, and report what changed.

## Try it on a real document

Upload a scanned PDF and export it as Word, a searchable PDF or plain text on the free plan. Your documents are never used to train an AI model. [Start free](https://www.schemalaw.com/api/auth/google)