Jotting
The informal record of how work actually gets done. A Slack thread. A Loom. A voice note after a job. The napkin. The "ask Maria" message. For years that was exhaust. Now it can be the raw material of a licensable package.
Definitions
Words we use
If a word on this site sounds like it belongs in a lab, it probably just means something you already do.
We have to use some of this language because that is how buyers talk. You do not have to. Here is the translation.
Operational data
The history created by running the business: jobs, matters, tickets, estimates, reservations, decisions, outcomes. Not your marketing site. Not your customer list. The work itself, sitting in Jobber, Clio, Toast, a CRM, a drive.
Corpus
Buyer-speak for the package we actually license. How the company works, cleaned up enough that a buyer can review it. Not a hard drive of leftovers. A defined set: what is in, what is out, what it represents.
License
Permission to use an agreed package for an agreed purpose - usually training or evaluating AI. You keep the company. You keep the underlying records. You are not selling the business or giving up equity.
AI training data
Examples a model learns from. Labs already used the public internet. What they lack is how a real business actually operates: the steps, the exceptions, the human calls. That is why ordinary businesses are suddenly interesting.
PII
Personally identifiable information. Names, emails, phones, addresses, payment cards, medical or legal client files. This does not go in the package. Buyers want how the work is done, not who your customers are.
SOP
Standard operating procedure. A written "this is how we do the job." Most businesses do not have clean ones. We write them where they are missing. That is part of making a corpus buyer-ready, not a separate homework assignment for you.
Workflow
A repeatable job with a start and an outcome. Estimate to install. Intake to filing. Reservation to close. Buyers care about workflows that happened hundreds or thousands of times, with decisions and results attached.
Manifest
The inventory sheet for a corpus. What sources, what size, what was excluded, what was scrubbed. The document that lets a buyer - and you - see exactly what is being licensed.
Anonymize / scrub
Strip or transform anything that identifies a person or should never leave the building. Credentials, secrets, and payment data are out. Customer identity is out. The pattern of the work stays.
Rights
Whether you are allowed to license the records. If your terms never mention this, that is normal. Counsel can read them and tell you if you have a clean right - or what language to add.
Buyer / AI lab
A company paying for operational history to train or evaluate models. You do not need to know them or take the meetings. That is our side of the table.
Qualifier
The two-minute form. Ordinary questions: what you do, how long, how many people, where the work lives, whether you own the records. It tells you if this is even a licensing conversation.
Success fee
How we get paid on a qualified licensing deal: a percentage of proceeds after it sells. If nothing sells, you owe nothing on that deal. Typical range is 15% to 40%, agreed in writing before work. The comparison is $0 vs a check, not the gross vs your net.
Transaction-ready
Far enough along that a buyer can actually purchase it. Inventory done. PII out. Rights reviewed. SOPs written if they were missing. Packaged. Most companies that get a nibble from a buyer page are not here yet. That is the job.
You do not need the vocabulary. You need to know if the history is worth preparing.
Two minutes to see if it is even a conversation. For qualifying companies, we get paid when you do.