OCR Employment Contracts and Personnel File Agreements

To OCR employment contracts held in scanned personnel files, first split each file into its separate documents, then convert only the agreements and signed acknowledgments, one file per document, named by employee and date. Check names, dates, pay figures and filled-in blanks against the page image before anyone relies on the converted text.

The splitting step is what makes the rest work. A personnel file is a bundle of unrelated papers about one person, and most of them are not contracts.

Which employment agreements sit in a personnel file as scans?

The agreements in a personnel file are the signed papers that set terms between employer and employee: the offer letter, the employment agreement, covenant and invention agreements, pay plans, later amendments and signed policy acknowledgments. They are scans because they were signed in ink, or signed years before the employer used electronic signing.

What a file may hold, sorted by whether the paper sets contractual terms:

Whether a given paper is a binding agreement is a legal question that depends on its wording and the governing law. For the conversion, the working rule is simpler: convert anything the employee signed that states an obligation, and let the lawyer decide what it amounts to.

For each employment agreement, copy the compensation figure, the start date, and any at-will employment sentence from the image. Look further in the PDF for a severance agreement and for a line that marks the person as an exempt employee or an independent contractor. If an employee benefits summary is attached, note its date. Record wage and hour labels as they are printed.

Stand-alone confidentiality agreements raise the same reading questions as any other NDA, and the post on how to review a scanned NDA covers them.

What makes a scanned personnel file awkward for OCR?

A scanned personnel file is awkward because it is one PDF holding many short documents of different kinds, scanned in whatever order they sat in the folder. Typed agreements, pre-printed forms, handwritten entries and photocopies of photocopies sit side by side, so recognition quality changes from page to page.

The specific problems:

• No document boundaries. A sixty-page PDF named with the employee's surname may hold fifteen documents. Nothing in the file says where one ends.
• Reverse or random order. Check the order before you read. A folder filed newest on top puts the 2024 pay letter before the 2016 agreement it amends.
• Filled-in blanks. Form agreements leave spaces for the name, start date, salary and territory. The entries are typed or written in after printing, often slightly off the line, and they are the terms you most need.
• Checkboxes. Elections such as exempt or non-exempt, or which bonus plan applies, are ticks in boxes.
• Acknowledgments that point elsewhere. A one-page receipt says "I have received the Employee Handbook dated January 2019". The handbook itself is not in the file.
• Generations of copying. The agreement was photocopied for the employee, the copy was faxed to head office, and the fax was scanned. Small print suffers each time.
• Stamps and notes. "Received HR", date stamps and sticky notes sit over text.
• Unrelated sensitive pages. Tax forms, bank details for payroll and medical notes are interleaved with the agreements.

Entries on pre-printed forms behave differently from running text, and the post on scanned forms explains why under handwritten and typed entries on a pre-printed form. The practical consequence for an employment file is that the printed clauses will usually convert far better than the salary, the name and the date written into them.

Converting offer letters, agreements and signed acknowledgments

Convert an employment file in three passes: inventory the scan, split out the documents that matter, then convert each one and check its filled-in terms. Keep the original personnel file scan untouched, and work only on the pieces you split from it.

Pass 1: inventory

Page through the PDF and list what it holds before converting anything.

This table is itself a useful work product. It tells the lawyer what exists for this employee and, as importantly, what does not.

Pass 2: split and name

Extract each "Yes" row into its own PDF. A naming pattern that sorts well:

Okafor-D_2016-04-11_employment-agreement.pdf
Okafor-D_2016-04-11_confidentiality-invention.pdf
Okafor-D_2016-04-12_handbook-ack.pdf
Okafor-D_2024-06-03_pay-change-letter.pdf

Surname first groups one person's documents. The date in year-month-day form puts them in the order they were signed, which is the order you need for working out which terms are current. If the employer uses employee numbers, use the number in place of the name. That keeps names out of file listings.

Pass 3: convert and check

1. Upload the split documents. In Schema Law that is a PDF, PNG or JPEG of up to 20 MB each, and the how it works section of the homepage shows the steps. You can choose several files at once on any plan, which suits a stack of files for many employees.
2. Export Word or plain text for agreements you will read and search. Export a searchable PDF as well if the documents will be filed or produced as images.
3. For each document, check these against the page image before anything else: - the employee's name as written, including middle initials and suffixes; - the employer entity, which in a group of companies may not be the one you expect; - the date of the agreement and the start date; - pay figures, bonus percentages and notice periods; - every filled-in blank and every ticked box; - any territory, customer list or time period in a covenant.
4. Confirm the signature lines. Is the employee's signature there? Is the employer's? Record what you see and leave the meaning to the lawyer. The post on extracting signatures and dates from scanned documents describes how to record who signed and when.
5. Note any pen edits. A form agreement may have been altered by hand at signing, and the post on handwritten changes to a contract explains why the converted text will not show them.

Finding restrictive covenants across a workforce's contracts

To find restrictive covenants across many employees, convert every agreement that could contain one, search the text for the language those clauses use, and record the result in one row per employee. The chart, not the search, is the deliverable: it shows who is bound by what, under which document.

The general method for searching a contract set, including how to log searches and show that a clause is absent, is in the post on how to search scanned contracts for a clause. What follows is specific to employment files.

Where the covenants hide

Do not search only the documents titled "Non-Competition Agreement". The same promises turn up in:

• a clause near the end of the main employment agreement;
• the confidentiality and invention assignment form;
• an equity or bonus plan acknowledgment, as a condition of the award;
• a promotion letter that adds obligations with the new role;
• a separation agreement that restates or extends them.

Language to search for

Search for word stems where your tool allows it, and remember that a poorly scanned page can hide a clause from every search. A document with no hits should be opened and paged through before you record "none".

One row per employee

A layout for the chart, with a fictional example:

Quote the scope. Do not paraphrase it, because the exact words are what the lawyer will analyze. Add a column for governing law if the agreements name one, because agreements signed in different places and years may not name the same one.

Whether any of these covenants can be enforced is outside the chart. The law on restrictive covenants differs widely between jurisdictions and changes, so the chart should report what each document says and stop.

How do you tell which version of an employment agreement an employee signed?

Look for a form identifier on the signed pages, and when there is none, compare the converted text with each known template. An employer may have revised its standard agreement between one hire and the next, so two employees with the "same" agreement may have signed different covenants, and the signed scan shows which.

Identifiers to look for on the scan, most reliable first:

1. A form number or revision date in the footer, such as "Form EA-3 (rev. 09/2019)". Check that it appears on every page, including the signature page.
2. A document number or file path printed in the footer.
3. The date of signing, set against the dates each template was in use, if the employer can supply them.
4. Distinctive wording. One version says twelve months and the next says six. One has an arbitration clause and the earlier one does not.
5. Layout. Page count, clause numbering and typeface.

When identifiers are missing, use the text. Convert the signed agreement, obtain each template version from the employer, and compare. The method is in the post on how to compare two scanned contracts. Differences confined to the filled-in blanks mean you have the right template. Differences in the clauses mean you do not, or that this employee negotiated changes.

Watch for three traps:

• A signature page from one version attached to the body of another. The footers will disagree. Report it. The post on counterpart signature pages lists the marks to compare.
• A later version sent to everyone but never signed by this employee. An unsigned template in the file is not the agreement the employee signed.
• An acknowledgment that refers to a handbook or policy by date. Record the date quoted, and ask the employer for that edition. The current handbook is not a substitute.

Add the result to the chart as its own column: the form version, and how you identified it.

Handling pay and personal data in converted employment contracts

Converted employment contracts put salary, home addresses and sometimes identification numbers into searchable, copyable text. Limit what you convert, limit who can open the output, and delete uploads when the task is finished. Rules on personnel records and employee privacy vary by jurisdiction, so check those that apply to the employer.

Searchable text spreads more easily than a scan. A salary figure in an image has to be read by a person. The same figure in a Word file is found by any search and travels with any paste. Plan for that:

• Convert less. Splitting the personnel file first means tax forms, bank details and medical notes are never uploaded at all.
• Keep identifiers out of file names where an employee number will do.
• Restrict the folders. The covenant chart may be suitable for the whole team. The converted agreements with pay terms may not be.
• Leave pay out of the chart unless the matter needs it. A covenant chart does its job without salary figures.
• Watch the exports. Each Word, text or Excel export is another copy to track and eventually remove.
• Redact before sharing outside the team, and test that the redaction removed the text and not only the picture of it. The post on OCR before or after redaction gives the sequence.

Agreements and onboarding forms sometimes carry a national identification number. The post on OCR for documents with Social Security numbers deals with that case.

Before uploading employee documents to any converter, read what the vendor says it does with them. Schema Law's pricing page lists six lines that apply to 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". The security section of the homepage explains each one. Those are the vendor's statements, and your own review decides whether they are enough for this data.

Schema Law deletes nothing automatically, so decide at the start how long the uploads and exports will be kept, and write the date down. The post on whether OCR software stores your documents explains retention periods and how to delete a document for good.