The honest answers before you license your data
Rights, privacy, liability and value. What to know before licensing company data to AI labs, and how we handle each part.
Genuine offers, named buyers
Data licensing is new, and not every offer in the market is what it claims to be. A careful look at who is buying, and on what terms, is the right starting point.
A real data deal looks like any other commercial contract. There's a named buyer, a written license, and payment from the lab to you. Nobody needs your data before that exists.
How we handle it
- No upfront fees, ever. If there's no deal, you pay nothing
- You see who every offer comes from before you accept it
- The license is signed between your company and the lab, and the lab pays you
- Nothing is exported until there's a signed license and you've approved samples
Anonymisation that covers the text itself
In email, chat and ticketing systems, personal and confidential details usually sit inside the message text, not in neatly labelled fields.
Removing names isn't the same as anonymising. Free text has to be scanned itself, and someone on your side should see the result before it leaves.
How we handle it
- We scan the message text, not just fields, for names, contact details, customer and company names
- Credentials like passwords, tokens and API keys are found and removed
- Preparation runs inside your environment wherever possible
- You review a sample of the cleaned data and can pull anything before delivery
- If a source can't be cleaned to a standard you're happy with, it stays out
Your rights, checked first
Company records are full of other people's information: customers, suppliers, partners, staff. Owning the systems doesn't automatically mean you can license everything in them.
Sometimes you can't, and it's far better to find that out before a buyer is involved. This is the first thing we check, not the last.
How we handle it
- Customer and client contracts, including confidentiality terms
- NDAs, partner and franchise agreements that limit sharing
- Open-source and third-party licenses in codebases
- Employee policies and the notice your team should get
- Anything without clear rights is left out, whatever it's worth
Liability, set out in writing
Both sides should know exactly what they are responsible for, during the deal and after the data is delivered.
Liability should be written down, not assumed. A good license limits what the lab can do, says what happens to the data afterwards, and keeps your promises to what you've actually checked.
How we handle it
- Use limited to agreed purposes, usually training and evaluation
- No resale or redistribution of your data
- Secure storage and deletion of raw files on an agreed schedule
- Your warranties limited to the rights we've reviewed together
- Every license reviewed with your own counsel before you sign
Your competitive edge stays yours
If your records show how you win work, price jobs or solve problems, handing them over could help someone compete with you.
Labs want patterns of how real work gets done, not your pricing sheet or client list. You decide what's in scope, and what you keep out stays out.
How we handle it
- You choose which systems, teams and time periods are included
- Pricing, strategy and anything you see as your edge can be excluded
- You can rule out any buyer, including anyone you see as a competitor
- Labs see an anonymised profile first. Your name is shared only once you choose to go ahead
Your team, treated fairly
Records are written by people. How they are told and what is left out matters as much as what the law allows.
It should be handled openly. The companies that do this well tell their team, leave out the most personal channels and make clear what's being shared.
How we handle it
- Direct messages, HR records and personal channels can be excluded by default
- We help you explain to your team what is shared and what isn't
- Personal data is removed before anything leaves
An honest view of the value
Most deals are a one-off license fee, and prices will move as the market matures. The decision should rest on real numbers.
For some companies it's clearly worth it, and for some it isn't. We'd rather tell you that early than push a deal that doesn't make sense.
How we handle it
- You get a realistic value range and the reasoning behind it, before you commit
- Licenses are non-exclusive by default, so the same data can earn more than once
- Recurring deals are possible for data you keep producing
- If the numbers don't justify the effort, we'll say so
How to judge any data offer
Use this on any offer you get, including ours.
Be careful if you see
- A large number with no named buyer
- Asking for data, repo access or exports before any contract
- Any fee you have to pay upfront
- Vague answers about how anonymisation works
- Pressure to decide quickly
- No liability or deletion terms in writing
What a real deal looks like
- You know who the buyer is before you commit
- Rights are checked before anything is exported
- You see and approve samples of the cleaned data
- Use, resale and deletion terms are written into the license
- The lab pays you directly, under a contract you signed
- You can walk away at any point before signing
Still unsure?
How do I know an offer is genuine?
Do we actually have the right to license this data?
How do you anonymise free text like Slack, email and tickets?
Who carries the liability if something goes wrong?
More on how we handle data on our Trust page, or ask us directly. A first call commits you to nothing.
Find out what your data is worth
Two minutes, no obligation, nothing leaves your systems.
Get an estimate