A residential HVAC company knows more about its customers’ homes than most businesses ever will. The age of the furnace, which bedroom never gets warm, the dog in the back yard, the code for the side gate. That intimacy is part of why the records are useful to an AI company, and it is the whole reason privacy has to come first.
The good news is that buyers do not want any of it. They want the work: what was wrong, what was found, what was done, what it cost, whether it held.
What do the buyers say they do?
Every buyer with published terms on our list describes some form of de-identification.
| Buyer | Stated approach |
|---|---|
| Scale AI | Strips personal information on your systems ahead of extraction, needs your sign-off on each package, and never takes employee or customer identities |
| micro1 | Cleans and rewrites sensitive fields itself; the company can inspect sample records |
| Nyne | Releases de-identified data only; silent on who does the work |
| Labelbox Prism | Scrubs PII after the data arrives |
Scale AI’s approach, working inside your environment so raw records never leave, is the strictest on paper. The others process data after it reaches them or do not say where. Either way, the safer path is to hand over records that are already clean.
What should come out of a service record?
Identities
Customer names, phone numbers and email addresses come out entirely. If a buyer needs to see that the same customer called three times, replace the name with a random code that means nothing outside your office. Do the same for technicians: a code per person keeps the pattern without naming anyone.
Locations
Street addresses come out. If climate matters to the buyer, and for HVAC it often does, keep the state or a broad climate region. A five-digit ZIP code in a small town can be close to an address; ask before keeping it.
Free text
This is where names hide. CSR notes and technician notes are the most valuable part of a record and the most likely to say “spoke to Linda, gate code 4471, dog in yard”. Search for phone number and email patterns, for words like “gate”, “code”, “lockbox” and “spoke to”, and for the first names in your customer list. Then read a few hundred records by hand. Software catches most of it; people catch the rest.
Money
Card numbers, bank details and anything from customer financing paperwork stay out. Prices, line items and which option sold can stay, because they describe the job, not the person paying.
Equipment serial numbers
A serial number can be traced to one install at one house. Leave it out unless a buyer asks for it and your lawyer agrees.
When do contracts or the law say no?
You can only license records you own or have the right to license. Commercial accounts are the trap here: a property manager’s service agreement may say who owns maintenance records on its buildings, and drawings or submittals on a construction job may belong to the owner or general contractor under the contract. Personal information about customers can only go in where your contract and the law both say it may.
Call recordings are a separate question again. States set their own recording rules, and around twelve of them insist that every person on the call agrees; find out which kind yours is. The familiar “this call may be recorded” line is about making a recording, which is not obviously the same as licensing it for training, and Illinois BIPA and similar laws can treat a voiceprint as biometric. Most HVAC deals are better off leaving audio out.
Your privacy notice matters too. If it promises customers something specific about their information, the deal has to keep that promise. That alone is reason enough for an attorney of your own to go through the agreement first, however clean the data.
Who sees your data along the way?
The buyer you sign with, under the license you sign. Not us. Records never reach this site, and the quote form here takes no files of any kind. Our part ends at telling you which buyers fit your company and introducing you to the one you choose. The homepage sets out how to sell HVAC data on that basis, and our customer privacy page has the field-by-field table to work from.
A clean set protects your customers first and your deal second, and it costs less to do before signing than after.