Skip to content

Speed, cost and accuracy.

For most teams the alternative is Adobe, a spreadsheet and a long afternoon. It is slow, expensive and easy to get wrong. This page sets out what changes.

Try RedacticsRun a pilot

Redacting by hand.

Nine questions anyone redacting documents has to answer. What the answer looks like today, and with Redactics.

Have we covered every page?
Adobe, a spreadsheet and peopleFiles opened one at a time by different people. No single record of what was checked.
RedacticsEvery page in the set is processed, with a log of what was uploaded and what was redacted.
Is this the same person?
Adobe, a spreadsheet and peopleA reviewer notices, or does not, that Dan, D. Okafor and NG-40217 are one person.
RedacticsMentions linked into person records with evidence, so the same person is redacted the same way everywhere.
Where did this information come from?
Adobe, a spreadsheet and peopleA page number in a PDF, if someone wrote it down.
RedacticsSource file, message, attachment, page, paragraph and span on every redaction, with the original retained.
Does this contain another person's information?
Adobe, a spreadsheet and peopleRead every page and hope the second reviewer reads it the same way.
RedacticsThird-party findings redacted automatically, grouped by person and role, with mixed sentences highlighted.
Why was this redacted?
Adobe, a spreadsheet and peopleA black box in a PDF. The reason lives in someone's memory.
RedacticsA reason code on every redaction and a schedule listing all of them.
Who approved the export?
Adobe, a spreadsheet and peopleAn email thread, a spreadsheet column, or nothing.
RedacticsReviewer, timestamp and second-reviewer sign-off recorded before anything leaves.
Can we prove what happened later?
Adobe, a spreadsheet and peopleReconstruct from mailboxes and memory.
RedacticsExport the append-only audit trail for the set, any time.
Where does skilled time go?
Adobe, a spreadsheet and peopleSearching, opening, reading and drawing boxes.
RedacticsChecking the redactions that need a human call.
What does it cost?
Adobe, a spreadsheet and peopleHours of skilled time per set, hard to predict in advance.
RedacticsPriced on pages and shown before processing starts: from £249 for 1,000 pages, then 20p a page.

Basic tools.

Basic tools start with one document and a search box. Redactics starts with the whole set and redacts it automatically. That difference runs through everything else.

Starting point
Basic redaction toolsA document. You open it, search for terms and mark up what you find.
RedacticsA whole set. Every file is extracted, checked and redacted automatically before anyone opens a page.
Unit of review
Basic redaction toolsPages and search terms.
RedacticsPeople and findings. Everything about the requester in one view; everything about each other person in theirs.
Person resolution
Basic redaction toolsNone. A name is a string to search for.
RedacticsAliases, identifiers and email addresses linked into person records with evidence.
Third-party information
Basic redaction toolsFound by reading. Marked up page by page.
RedacticsIdentified, redacted automatically, grouped by person and role, confirmed with recorded reasons.
Scans and handwriting
Basic redaction toolsOCR if you are lucky. Handwriting is read by a person.
RedacticsPrinted and handwritten pages transcribed with confidence scores, redacted, and checked side by side.
Provenance
Basic redaction toolsA box on a page.
RedacticsA finding with a source span, a redaction, a reviewer and a reason.
Audit
Basic redaction toolsA redaction log, sometimes.
RedacticsAn append-only trail of upload, extraction, detection, views, changes and exports.
Output
Basic redaction toolsA redacted PDF.
RedacticsRedacted documents with an index, a redaction schedule and the evidence behind every mark.

One record per person.

A redaction is about a person, not a string of characters. The question on every page is whether a piece of information is the requester's, someone else's, or both. Tools that work one file at a time cannot help with that question because they do not know who anyone is.

Person resolution links every mention across the set, whether a full name, an initial, an email address, an employee number or a phone number, into one record per person. Confident links are automatic and evidenced. Ambiguous ones are queued for a reviewer. The result is a review organised by person: everything about the requester, and everything about each other person, in one place.

It also catches the mistakes that cause complaints: two people with the same surname treated as one, or one person under four names treated as four.

A source for every mark.

Sooner or later, someone asks about a redacted document: where did this come from, and why was it redacted? The requester may ask. A regulator may ask. A court may ask. Answering from memory is not an answer.

Provenance means every redaction points to its source: the file, the message, the attachment, the page, the paragraph and the exact span. It means every change carries the reviewer, the time and the reason. It means the original is kept beside the extracted text so the chain can be checked.

With provenance, answering takes seconds. Without it, answering means doing the work again, and hoping the answer is the same.

People approve the export.

Redactics applies redactions automatically, but it does not decide what your organisation releases. It extracts, detects, links and redacts, and it records what it did. A named person reviews the result, adjusts it where needed and approves the export.

This is a design principle, not a limitation. Whether a document can be released, and under which rules, is a judgement that belongs with your organisation, and your organisation has to be able to explain it. A system that made that call for you would not remove the accountability; it would make it harder to discharge.

The practical effect is that the searching, reading and marking up, which is most of the time, is done by Redactics, and the final check, which is most of the risk, is done by people with the evidence in front of them.

Whole sets, not single files.

Many tools can highlight names, dates and email addresses in a document. That is useful for a single file and unhelpful for a set of thousands of pages. It tells you that a string appears on page 12. It does not tell you who it refers to, whether the same person appears on page 40 under a different name, or whether the sentence is about the requester, a colleague or both.

Redactics treats the whole set as one body of evidence. People are resolved across documents. Findings are categorised with context. Third-party information is grouped. A redaction is applied once per finding and consistently wherever it appears.

That is the difference between redacting a file and redacting a document set, and it is why Redactics is built around the set.

People/Daniel Okafor

Person record

Daniel Okafor

Data subject. Confidence 98%.

1,963 mentions across 611 sources, from 5 identifiers.

Identifiers linked

5

  • Daniel Okafor

    HR case file

  • d.okafor@northgate-ht.example

    Mailbox export

  • D. Okafor

    OH-assessment-DO-Mar24.pdf

  • Employee NG-40217

    Payroll adjustments Q3.xlsx

  • Dan

    Handwritten margin note

Kept separate D Okafor Deliveries, 54% match. Appears only in supplier invoices. Different address and phone number from the requester.

Person resolution: the requester under five identifiers, and a similar name kept separate for a reviewer.

Redact your first document set.

Try Redactics on a synthetic sample set, or talk to us about your documents.

Try RedacticsTalk to us