// Field notes / Data licensing
Getting your records ready for an AI data license
By Rob Eason, founderPublished
// The order
Legal review comes before the first file
Buyers often ask for a sample early, to price the deal. Both law firms writing on this market say no: Frankfurt Kurnit Klein + Selz and LumaLex Law each say not to share sample data until legal review is done, because a sample is itself a disclosure.
So the steps below assume your privacy lawyer has already looked at your privacy promises, contracts and the laws that cover your records, and told you what's off the table. Our note on whether it's legal lists what to bring to that meeting.
// Step one
List every system that holds records
Start with a plain list. For each system, write down what it holds, how many years it covers, whose name the account is in, and who on your team knows it best.
Buyers name the same families of tools. Scale AI's list is a useful map:
- Chat: Slack, Microsoft Teams, Google Chat.
- Email and calendars: Gmail, Outlook, Google Calendar.
- Documents and files: Word, Google Docs and Sheets, Excel, Drive, SharePoint.
- Work tools: Jira, Confluence, Asana, Notion, Linear.
- Business platforms: Salesforce, HubSpot, Workday, ServiceNow, SAP, Zendesk.
- Data warehouses: BigQuery, Snowflake, Databricks.
// Step two
Check what each tool lets you export
Your records sit inside other companies' software, and their terms come with them. Both law firms point out that Slack's API terms prohibit third-party apps from using API data to train large language models. Read the terms of every tool on your list for limits on exports and on use.
Note too where the records would be pulled from. Scale AI says it works from structured, de-identified records, never the raw systems themselves, and offers on-premises and private cloud options. Know whether any buyer's software would connect to your systems, and with what access.
// Step three
Carve out what never leaves
Frankfurt Kurnit's advice is to leave high-risk data out entirely rather than rely on de-identification. Decide the carve-outs before anyone touches the records, and write them down.
- Health, financial account and government ID details.
- Anything about children.
- HR files, payroll and personnel matters.
- Other companies' confidential material: messages with partners, vendors and customers under NDA or contract terms.
- Messages with your lawyers and accountants.
- Call recordings, until your lawyer has checked what callers agreed to.
- Any customer or account whose contract limits how you use their data.
// Step four
Decide where personal details come out
Names, emails, phone numbers and other details have to come out of whatever is left. The programs handle this differently, so ask each buyer where it happens and who checks it.
- Inside your systems first. Scale AI says it anonymizes everything inside your environment before any data is extracted.
- At the buyer, after transfer. micro1 says you don't need to anonymize first, because it has its own processes, and that personal and identifying information is removed or transformed before data goes to third parties. It adds that some datasets may get synthetic rewrites.
- Scoped by you. Handshake AI says you determine exactly what is shared, and that all data is scoped, de-identified and confidential.
// The bar
Removing names isn't the same as anonymous
The FTC's staff set the bar plainly in 2024: data is only anonymous when it can never be associated back to a person. Swapping an email address for a scrambled code, which is what hashing does, doesn't meet it, and the FTC says companies shouldn't treat hashed personal information as anonymized.
Work records make this harder than it looks. A message that mentions a job title, a client's city and a date can point to one person with no name in it. Ask the buyer how it tests for that, and look at a few finished records yourself before you approve a package.
// Steps five and six
Approve each package, and keep the record
Don't hand over everything at once. Scale AI describes per-package approval gates, with nothing shared without your explicit approval of each package. Ask for the same from any buyer, in the contract.
Then keep a folder from the first day: what was delivered, from which systems and years, what was carved out and why, how personal details were removed, who approved each package, and when raw data was deleted. micro1, for one, deletes raw data on an agreed schedule, so get the schedule in writing and keep each confirmation.
If you sell the business one day, that folder is what the buyer's lawyers will ask for. Our note on licensing before a sale lists what they'll want to see.
// Questions
Getting the records ready
How do I prepare company data for AI licensing?
Do I have to anonymize data before licensing it?
Can I send an AI buyer a sample of our data first?
Which systems do AI data buyers want records from?

Rob Eason, founder
// Book a call with Rob
Mapping which of your systems hold what?
A 30-minute call with Rob. He'll help you list your systems and what's in them, sort what to take to your lawyer, and look at what the same records could do inside your own company.
- Co-owned a 60-employee financial services firm that served more than 7,000 customers and has a 5-star rating across more than 600 customer reviews, and sold his share to his partner.
- Ran GOSO Technologies, a Boston solar company that Forbes Home named one of the top 10 solar companies in Boston, rated 5/5 across 24 Google reviews.
- Spoke on the construction panel at Build Connect 2026 in Charlotte, sponsored by Choate Construction.
- HighLevel Certified Admin with the second-most reviews in HighLevel's Certified Admin Directory worldwide: 5/5 across 83 reviews.
- Won HighLevel's Top New Subaccounts via AI Employee award in its 2025 Summer of AI Awards.
- 4.3/5 from 6 Google reviews of Massmarket AI.
- Every directory and Google review, critical ones included, is shown word for word at massmarket.ai/reviews.
- Takes every proof-of-concept call himself.
