Article
How de-identification works on your records
Updated
Every owner who hears about licensing company records asks the same thing. What happens to the customer in the file? The short answer is that the records are de-identified before use. This page shows what that means on the files you actually keep: an estimate, a work order, a change-order email thread, a Slack channel and a CRM note, one at a time.
Our page on whether it’s safe to license company records covers the controls: who sees the records, access, sample review, retention and deletion. This page doesn’t repeat them. It goes one step closer to the paper. What comes out, what stays, and what the record looks like after.
What de-identification means, in one sentence
There is a standard definition, and it is short. The US National Institute of Standards and Technology, in its glossary and in its guide to de-identifying government datasets, calls it “a general term for any process of removing the association between a set of identifying data and the data subject.”
In plain words: the record stops pointing at a person. The work in it can still be read. The homeowner, the tenant, the site contact and the technician can’t be picked out of it.
NIST’s guide is written for government agencies, and it describes de-identification “with the goal of preventing or limiting disclosure risks to individuals”. That is the useful frame for a trade company too. It isn’t legal advice, and we don’t give any. It is a way to read your own files.
micro1 uses the same word. In its 9 October 2026 announcement it says it “uses de-identified company data.” Its data partnerships page names two tools for getting there. We’ll take them one at a time.
Apply & License Your DataThe two tools micro1 names
Removal. micro1 says “Personally identifiable information (PII) is removed when relevant.” On what happens after records are shared, it says: “Depending on the nature of the data, personally identifiable information may be removed.”
Synthetic rewrites. In the same answer, micro1 says “certain datasets may undergo synthetic rewrites that preserve workflow structure while reducing direct ties to original records.”
Removal takes something out. A rewrite changes the details and keeps the shape. The first is a black bar over a name. The second is the same job told again with a different date, a different street and a different figure, in the same order of steps.
Neither tool is fixed in advance. micro1’s words are “when relevant,” “may be removed” and “certain datasets.” What decides it is a list agreed before anything moves. micro1 says that “Before any participation, together we define” four things, and one of them is “Anonymization and redaction requirements.” The others are the “Approved data scope and clear boundaries,” the “Security and confidentiality standards” and the “Internal approval processes and permissions.”
So the rules for your records are not a black box. They are a list you help write. The rest of this page is about what goes on that list.
Why the steps survive and the names don’t
Here is the core of it. Customer names and payments are never what is wanted. What is wanted is how the work was done. How an estimate became a job. How a change order was settled. How a repair was diagnosed.
micro1 says it is looking for “data that reflects how work is actually performed within an organization.” It says “The most valuable datasets capture real operational expertise, decision-making processes, and repeatable workflows” and goes on from there. One of its cards is titled “Decision-making patterns,” with one line under it: “How teams evaluate information and make operational decisions.”
A name doesn’t evaluate anything. A phone number doesn’t make a decision. Take them out and the decision is still there. That is why de-identification works on trade records better than owners expect. The value and the identity sit in different parts of the file.
Our cornerstone page on what data buyers want from a trade business goes through that split by trade. Below, we go through it by record.
A note on the examples. Every before-and-after below is our own illustration, written for this page. None of it is a real record, and none of it is micro1’s. The bracketed parts stand for what a real file would hold.
Record one: an estimate
An estimate is the record most companies have the most of. Our page on old estimates and quotes explains why the versions matter more than the totals.
What usually comes out. On a commercial estimate, the identifiers are different from a homeowner’s. The building owner’s name. The property manager’s name, email and phone. The building address. The tenants named in the schedule notes. The purchase order number. The signature block.
What stays. The equipment sizes. The site findings. The options priced. The schedule constraint. Why the second version costs more than the first.
Illustration, written by us. Before:
Estimate v2, RTU replacement, [building owner’s company], [building address]. Contact [property manager’s name], [mobile number]. Roof walk found the existing curb rusted through at two corners. Version 1 assumed a curb adapter on the old curb. Version 2 adds a new curb and roof patch, and moves the crane lift to a Saturday so [ground-floor tenant] stays open. Options: like-for-like 10-ton unit, or a 12.5-ton unit for the planned kitchen. Owner chose like-for-like. Total [price].
After removal:
Estimate v2, rooftop unit replacement, commercial client, [building address removed]. Roof walk found the existing curb rusted through at two corners. Version 1 assumed a curb adapter on the old curb. Version 2 adds a new curb and roof patch, and moves the crane lift to a Saturday so a ground-floor tenant stays open. Options: like-for-like 10-ton unit, or a 12.5-ton unit for the planned kitchen. Owner chose like-for-like. Total [price].
Read the second version. The reasoning is all there. A roof walk found a hidden problem the first price didn’t cover. The estimator repriced it, planned the lift around a tenant and offered a bigger unit for a future use. That is an estimate becoming a job.
Prices, if agreed. Prices are a decision for you. They aren’t personal information, so nothing on micro1’s pages says they come out by default. As our own reading, a price is often part of the reasoning: the new curb is why version two went up. As our own advice, ask on the call whether a rewrite would change the figures or whether prices come out by rule. Also as our own advice, ask what either choice does to the offer. micro1 says the final payout “is determined based on data sources, volume, quality, access terms, and due diligence.”
Record two: a work order
A work order is shorter than an estimate and denser with identifiers. That is because it was written for a technician to find a house.
What usually comes out. The service address. The gate code and the lockbox number. The dog’s name, which is more identifying than it sounds. The callback number. The technician’s full name, if your rules treat staff names as personal information. As our own advice, they should.
What stays. The complaint. The checks, in order. What was found. What was replaced. Why the technician chose that fix over another one.
Illustration, written by us. Before:
WO [number]. [Customer’s full name], [street address], gate code [four digits]. Complaint: no hot water. Tech [technician’s full name]. Checked pilot, OK. Checked thermocouple, OK. Found gas valve not opening. Replaced valve, not tank: tank 6 years old, no leak, anode fine. Customer OK’d by phone at [mobile number].
After a synthetic rewrite:
Work order. Residential customer. Complaint: no hot water. Technician A. Checked pilot, OK. Checked thermocouple, OK. Found gas valve not opening. Replaced valve, not tank: tank 7 years old, no leak, anode fine. Customer approved by phone.
Look at what changed. The tank’s age moved by a year. The number, the name and the code are gone. Look at what didn’t change. The order of the checks. The reason for repairing rather than replacing. That reasoning is the record. It is how a repair was diagnosed.
The rewrite is what the phrase “preserve workflow structure while reducing direct ties to original records” means on a ticket. The steps match the job. The ticket no longer matches one job.
Record three: a change-order email thread
Email is harder than a form. A form has fields. An email has people, and people put names everywhere: the greeting, the sign-off, the forwarded chain, the attachment’s file name.
What usually comes out. Every email address. Every name in a greeting or a signature. Phone numbers in the footers. The project address in the subject line. The names of the client’s own staff. Attachments that carry a name or a stamp.
What stays. Who asked for what, by role. The cause of the change. The price discussion. The settlement and the reason given for it.
Illustration, written by us. Before:
From: [project manager’s name] <[work email]> To: [owner’s rep name] <[client email]> Subject: CO #4 – [project address] Opened the north wall, found the header undersized for the new opening. Need to replace with an LVL. Added 2 days and [amount]. Photos attached.
Reply: The drawings showed this, didn’t they? Can you split the cost?
Reply: The drawings showed the opening, not the existing header. We’ll take half the labour. Material at cost.
After removal:
From: Project manager. To: Owner’s representative. Subject: Change order 4. Opened the north wall, found the header undersized for the new opening. Need to replace with an LVL. Added 2 days and [amount]. Photos attached.
Reply: The drawings showed this, didn’t they? Can you split the cost?
Reply: The drawings showed the opening, not the existing header. We’ll take half the labour. Material at cost.
That is a change order being settled. Three messages, one argument, one compromise and the reason for it. None of it needs a name.
As our own advice, flag one thing for your list: the photos. A site photo can carry a street sign, a house number or a face. Decide whether photos are in scope at all before anyone discusses how to clean them. Our page on the company inbox covers what else an email archive tends to hold.
Record four: a Slack or Teams channel
Chat is the loosest record a company keeps, and often the most useful one. Our page on Slack and Teams chats explains why. It is where the dispatcher, the estimator and the foreman actually decide things.
What usually comes out. Every display name and handle. Customer names typed in passing. Addresses pasted for a tech. Phone numbers. Personal chatter: the sick child, the doctor’s appointment, the leave request. As our own advice, put that last group out of scope rather than redacted. It isn’t how the work was done.
What stays. The problem raised. Who was moved where, by role. The trade-off made. What happened next.
Illustration, written by us. Before:
[dispatcher’s name]: [customer’s name] on [street] says no heat again, third call this month [service manager’s name]: send [senior tech’s name], not the new guy. Pull the history first [senior tech’s name]: history shows two igniters in 30 days. It isn’t the igniter. Bringing a flame sensor and a manometer
After a synthetic rewrite:
Dispatcher: repeat no-heat call, third this month Service manager: send the senior tech, not the newest one. Pull the history first Senior tech: history shows two igniters in 30 days. It isn’t the igniter. Bringing a flame sensor and a manometer
Here the rewrite replaced people with roles. Nothing else needed to move. The judgment is intact: a repeat call goes to experience, and the history rules out the part that was already replaced twice.
Record five: a CRM note
A CRM is built around a person, so it carries more identity per line than anything else. micro1 lists “CRM data” among the records it looks for, and has a “Sales & CRM” card with “CRM data, sales processes, playbooks, customer lifecycle documentation, and revenue operations.” Our page on CRM records explains what those notes show.
What usually comes out. The contact’s name, title and company. Emails and phone numbers. The referral source, if it’s a person. The commercial client’s name, if your contract says it is theirs.
What stays. The stage history. The objection. The follow-up. Why the deal was won or lost.
Illustration, written by us. Before:
[Contact name], facilities manager at [client company]. Wants 3 bids. Ours was second-lowest. Lost on schedule, not price: they needed the roof done before [tenant name] moves in on the 1st. Next time lead with crew availability.
After removal:
Facilities manager at a commercial client. Wants 3 bids. Ours was second-lowest. Lost on schedule, not price: they needed the roof done before a tenant move-in. Next time lead with crew availability.
The last line of that note is the lesson. It is the kind of line micro1 means by decision-making.
Our page on whether the buyer sees your customers’ names follows the customer through each of these systems, one place at a time.
The identifiers owners miss
Names, addresses and phone numbers are easy. They are what every owner thinks of. The harder ones are the details that point to one job without naming anyone. This section is our own advice, not micro1’s.
- Free-text fields. The technician’s notes, the job comments, the “misc” box. That is where a name lives when no form asked for one.
- Dates on rare jobs. A crane lift on a named street on a named day is one job. Ask whether dates are shifted in a rewrite.
- Small places. In a town of 900, “the church on the corner” is an address.
- Unusual work. The only geothermal install your company ever did is identifiable from its scope alone.
- File names. “[surname]_roof_final_v3.pdf” undoes a clean redaction inside the file.
- Attachments and scans. Signatures, licence plates, faces and letterheads on a scanned page.
- Your own staff. Names, phone numbers and anything from the HR side that wandered into a job file.
None of these is a reason not to license. Each is a line for the list you bring.
The redaction list: what we suggest you bring to discovery
Our page on whether it’s safe to license company records has the basic list. Here, as our own advice, is a fuller one-page version for the discovery call, in five parts.
- The systems. Which ones are in scope. The estimating tool, the job software, the shared inbox, Slack, the CRM. Name each.
- The fields. For each system, the fields that hold names, addresses, phones, emails, account numbers and signatures.
- The free text. For each system, the comment and note fields that might hold any of the above.
- The decisions. Prices in or out. Staff names removed or kept as roles. Photos in or out. Commercial client names, which your contracts may decide for you.
- The questions. Which records get removal and which get a rewrite. Who does it, and at what step. How many samples you’ll review, from which systems and years.
Who inside the company should own it
This is our own advice. One person owns the list. Two others check it.
The owner of the list is whoever knows the files best. In a company of 30 or more, that is usually the office manager or the operations manager. They know which field the dispatchers misuse and which tech writes customer names in the notes.
The first check is someone who knows the contracts. Your counsel or your controller. They answer the question the list can’t: which records a client owns, and which client agreements say anything about sharing.
The second check is the person who signs. micro1 lists “Internal approval processes and permissions” among the four things defined before any participation. That is where your sign-off order goes.
The same office manager should sit in the sample review. micro1 says companies “are given the opportunity to review representative samples before prepared data is used.” The person who wrote the list is the one who will see a name that slipped through. Who in your office would spot a customer’s name in a tech’s note on the first read?
What this means for a general contractor with 30 or more staff
Here is one company, as our own description, not a list from micro1. Say a residential and light-commercial general contractor of 40 people, as a hypothetical, with ten years in its estimating and project software. micro1’s referral page sets the line at “30+ employees,” so this company is in range. Our page on companies with 30 or more staff covers that lane.
The records it keeps. Estimates with every version. Daily logs. Change-order threads in the shared inbox. A Slack workspace with a channel per job. A CRM of commercial leads. Subcontractor bids.
What its list would say. Out: client names, site addresses, phones, emails, signatures, lockbox codes, subcontractor contact names, staff names. Out of scope entirely: the HR channel, payroll, bank details, the insurance claim files. A decision to make: prices, and whether photos are in at all.
What stays. How a scope was revised after the first walk. How a hidden-condition change order was priced and settled. How the superintendent re-sequenced a job when the windows came late. How a commercial bid was lost and what the team changed next time.
That is the record a buyer licenses to train AI models on real work. It reads the same with every name taken out.
What is never included
micro1’s own words are short. On its payout page: “We understand business data can contain sensitive information.” It says every partnership is designed around the company’s requirements, “with clear controls over what is shared, how it is handled, and what is never included.” micro1 settles the exclusions with each company, before any participation. It doesn’t publish a fixed list.
De-identification is for records that belong in the scope. Some records shouldn’t be in the scope at all, however well they could be cleaned. What follows is our own advice, not micro1’s. Keep these out unless micro1 agrees otherwise in writing: attorney-client privileged records, health information, payment-card and financial-account details, Social Security numbers, minors’ data, biometric data, and classified or export-controlled material.
Two more points, also ours. The payments ledger and the cards on file were never how the work was done. And anything a client owns under its contract belongs to the client. Have counsel read those contracts before the discovery call. Our How it’s safe page gives the short version of all of this.
Where we fit
We, TrustPast, explain how this works and give you a range with our estimate. The range is ours, not an offer. You apply to micro1 yourself, through the button below. micro1 pays us a referral fee. You pay us nothing.
We never see your files, so we never touch the de-identification. Your records go from your systems to micro1 on the redaction rules you agree with micro1. The list on this page is only a way to arrive at that conversation prepared.
Worth two minutes to see your range first?
TrustPast
Questions and answers
Does micro1 take the customer names out of our records?
micro1 says “Personally identifiable information (PII) is removed when relevant.” On what happens after data is shared, it says that depending on the nature of the data, personally identifiable information may be removed. The anonymization and redaction requirements are defined with the company before any participation.
What is a synthetic rewrite?
micro1 says certain datasets “may undergo synthetic rewrites that preserve workflow structure while reducing direct ties to original records.” In our own plain words: the order of the steps is kept, and the details that point back to one job are changed.
Who decides what gets redacted?
Both sides. micro1 says that before any participation, together with the company, it defines the approved data scope and clear boundaries, the security and confidentiality standards, the anonymization and redaction requirements, and the internal approval processes and permissions.
Do we see the de-identified records before they are used?
Yes. micro1 says companies “are given the opportunity to review representative samples before prepared data is used.”
Should we scrub our files before we apply?
Our own advice is no. Leave your systems as they are. micro1 says the anonymization and redaction requirements are defined together before any participation, so bring a list of where the names live to the discovery call instead.
*Purchases are made by micro1 under its data partnership programme. TrustPast is an independent referral partner of micro1.
Sources
- micro1, data partnerships: PII removed when relevant, synthetic rewrites that preserve workflow structure, what is defined before participation, sample review, retention and deletion, the data it looks for (micro1.ai)
- micro1, how much is your data worth: what is defined before participation, clear controls over what is never included, and access terms among the final payout factors (micro1.ai)
- micro1, refer a company: the 30+ employee requirement (micro1.ai)
- micro1, 9 October 2026: micro1 uses de-identified company data (micro1.ai)
- NIST Computer Security Resource Center glossary: de-identification
- NIST SP 800-188, De-Identifying Government Datasets: Techniques and Governance (September 2023)