What de-identified company data looks like

The question we hear more than any other is some version of: "What will they actually see?" It's a fair question. Abstract promises about privacy are hard to evaluate. Examples are easy.

This guide shows what typical operational records look like before and after de-identification, system by system. Every example is fictional, but they're modeled on the kinds of records owners of 20- to 200-person companies actually have. For the methods behind the examples, read how to de-identify business data.

The principle: remove identity, keep the work

De-identification has one job: make it impractical to identify the people and organizations in the data, while keeping the work itself readable. Buyers don't want to know that Dana talked to Mike. They want to see that an operations lead escalated a recurring shortage to a sales rep, who negotiated a credit with the customer's accounts payable contact.

That means three kinds of treatment:

Consistency is the key detail. If [PERSON_1] is the same person across every channel, thread and system, the work still makes sense, and nobody learns who [PERSON_1] is.

Slack: before and after

Before

#dispatch — Marcus Webb: truck 14 broke down outside Roanoke, Tyler's on the shoulder. Anyone have a relay driver near Christiansburg? Need to make the 6am at Kessler Foods DC

Priya Shah: Calling Ridgeline Transport now, they covered us in March

After

#dispatch — [PERSON_1]: truck [UNIT_ID] broke down outside [CITY_1], [PERSON_2]'s on the shoulder. Anyone have a relay driver near [CITY_2]? Need to make the 6am at [CUSTOMER_1] DC

[PERSON_3]: Calling [VENDOR_1] now, they covered us in March

What survived: the breakdown, the time pressure, the idea of a relay, the fallback to a vendor who helped before. That's a complete, realistic example of operational problem-solving.

Email: before and after

Before

From: Laura Kim laura@acmesupply.com To: Ben Ortiz ben.ortiz@harlowfoods.com Subject: RE: PO 44821 backorder

Ben, the stainless fittings on PO 44821 are backordered until the 18th. I can ship the 200 we have now and the rest when they land, or substitute the Halvorsen equivalent today. Let me know by 3.

Laura Kim | Inside Sales | 304-555-0188

After

From: [PERSON_1] <[EMAIL]> To: [PERSON_2] <[EMAIL]> Subject: RE: PO [PO_ID] backorder

[PERSON_2], the stainless fittings on PO [PO_ID] are backordered until the 18th. I can ship the 200 we have now and the rest when they land, or substitute the [BRAND] equivalent today. Let me know by 3.

[SIGNATURE REMOVED]

Signatures are removed entirely because they're dense with identifiers and add nothing. Product brands may be kept or generalized depending on the agreement.

Support tickets: before and after

Before

Ticket #20931 — Customer: Greenway Dental (acct 7781). "Since the update our front desk can't print appointment cards." Agent note (Sam): Reproduced on Win10. Driver conflict. Sent rollback steps, customer confirmed fixed. Tagged: printing, regression.

After

Ticket #[TICKET_ID] — Customer: [CUSTOMER_1] (acct [ACCOUNT_ID]). "Since the update our front desk can't print appointment cards." Agent note ([PERSON_1]): Reproduced on Win10. Driver conflict. Sent rollback steps, customer confirmed fixed. Tagged: printing, regression.

The diagnosis and fix are the valuable part, and they're untouched.

CRM and ERP records: before and after

Documents and SOPs: before and after

Documents need the lightest touch. A procedure titled "How Jen handles month-end AP" becomes "How [ROLE: AP specialist] handles month-end AP." Internal URLs, passwords and account numbers are removed. Vendor-copyrighted manuals are excluded entirely. See monetizing SOPs and internal documentation.

What's always removed or excluded

• Names of people, replaced consistently
• Emails, phone numbers and street addresses
• Government ID numbers, bank and card numbers
• Passwords, API keys and other credentials
• Health information, unless handled under a compliant HIPAA process (we exclude it by default)
• HR, payroll and legal channels or folders
• Anything you mark as off limits

What we check before delivery

Automated tools catch most identifiers, but not all. Before anything is delivered:

1. Pattern scans look for names, emails, phones, addresses and numbers in every field, including free text.
2. Rare-combination review looks for details that identify by combination, like a unique job title in a small town.
3. Human spot checks read random samples across systems.
4. Your review: you see a redacted preview and can narrow the scope or walk away.

No method is perfect, and we won't claim otherwise. That's why contracts also prohibit re-identification and limit how the data can be used. Our security page explains the full process.

What isn't changed

Owners sometimes expect de-identification to strip data down to something unrecognizable. In practice, most content stays exactly as written:

• The words people used to describe problems, decisions and outcomes
• Timing: when things happened and how long they took
• Quantities and measurements: units, counts, durations
• Structure: who replied to whom, which ticket linked to which order, what status changed when
• Tone: urgency, frustration, relief, all of which help show how work really happens

That's why de-identified data stays useful. The parts that identify people are a small fraction of the text; the parts that show how work gets done are most of it.

How the redacted preview works

Before any delivery, you get a preview: a representative sample from each system in scope, de-identified exactly as the full dataset will be. You can read it, search it and flag anything that bothers you. Common requests at this stage are excluding an extra channel, removing a customer entirely or adding a term (a product code name, a nickname) to the redaction list. Changes are applied to the whole dataset, and you can ask for another preview.

Edge cases we watch for

Some identifiers hide in places automated tools miss. Nicknames ("Big Mike"), initials in sign-offs, email addresses embedded in forwarded text, customer names inside file names, and job titles that only one person holds all need attention. So do references to local landmarks, unusual events and anything that would let a reader who knows your town recognize your company. These are exactly the cases human spot checks are designed to catch, and the cases where your own review of the preview is most useful, because you know your business better than any tool.

Related reading

• How to de-identify business data: the methods behind these examples
• What AI labs look for in enterprise data: why the workflow matters more than identity
• Selling Slack data and selling email archives: system-specific scoping

Frequently asked questions

Is pseudonymized data still personal data?

Under some laws, consistently pseudonymized data can still be treated as personal data, because the key that maps placeholders to people exists. That's one reason the mapping key stays with you or is destroyed, never delivered to the buyer. Ask counsel how the laws that apply to you treat it.

Do buyers ever get the mapping between placeholders and real names?

No. The mapping is never delivered, and agreements prohibit attempts to re-identify anyone.

How long does de-identification take?

It depends on volume and the systems involved, but for most companies with 20 to 200 employees it's a matter of days to a few weeks, including your review.

Can someone figure out who my company is from de-identified data?

The goal is to make that impractical. Company names, addresses and distinctive details are removed or generalized, and buyers are contractually prohibited from trying to re-identify anyone. No method can promise zero risk.

Do placeholders like [PERSON_1] stay the same across systems?

Yes, that's the point. The same person or customer gets the same placeholder everywhere, so threads and records stay linked.

Are dates and numbers changed?

Usually not, because timing and quantities carry the operational signal. Dates can be shifted consistently if timing itself is sensitive.

Will I see my data before it's delivered?

Yes. You review a redacted preview and approve it before any delivery.

What happens to attachments and images?

They're often excluded in a first delivery, because images and files can contain identifiers that are harder to catch. Reviewed attachments can be added later.

Can I ask for extra redaction?

Yes. You can exclude channels, mailboxes, customers, date ranges or specific terms at any point before delivery.

Getting started

The quickest way to see what your own data would look like is to ask. DataOffer can run a small redacted sample with your approval, at no cost, before you commit to anything. Get an offer to start.

This guide is general information, not legal, tax or financial advice. Figures and ranges are illustrative; talk to qualified advisors about your situation.