What it costs you
Nothing, at any stage. We are paid only out of what a license earns. If it never sells, you are out the staff time and nothing else.
AI labs are paying for records of how real work gets done. We find the buyer, handle the contract and walk you through all of it. Personal details are stripped before anything leaves your systems.
Nobody has ever quoted your business a price for this, so there is no going rate to look up. This works from replacement cost instead: what it would cost to pay people to produce the same records from scratch.
Move the inputs to match your business.
This changes how much a record is worth, not whether it counts. Healthcare and consumer finance are out of scope for now.
Everyone, full and part time, across all locations.
History counted for licensing stops at seven years, because buyers rarely want more than that.
Not where it is supposed to go. Where it really ends up.
Whether it worked, what it cost in the end, whether the customer came back. The single biggest driver of value.
Illustrative estimate in USD. Not an offer, not a valuation, and not a guarantee of a sale. What a dataset actually clears for depends on buyer demand, the state of your systems, and the rights you are able to grant.
Illustrative profiles, not completed sales. Each runs the same arithmetic as the estimator against a shape of business we expect to see.
Notice how much the recording habits move the number compared to headcount.
Everything published openly has already been scraped and trained on. What has not been touched is the record of how actual work gets done inside actual companies.
You generate that every working day, and then you file it.
Every website, forum, manual and public document. All of it went into the last generation of models years ago.
The quote you nearly sent. Why you changed it. What the job actually cost when it was finished.
Which option your people compared, what the situation actually was, and why they rejected the obvious answer. The final number is the easy part. The path to it is the asset.
The job that came back. The quote that was wrong by thirty percent. The fix that held for three years. Almost nothing on the open internet has this attached, which is exactly why it is worth paying for.
The supplier they called, the spec they looked up, the thing they went and physically measured. Models are bad at precisely this, and the people building them know it.
Every business holds information about other people, under contracts, under privacy law, and under plain common sense. That constraint shaped the whole scope.
We built around it rather than asking you to find a way past it.
Your company's work, de-identified.
Removed before anything is transmitted.
Constructed example, a service call. De-identification runs on your machine, before transmission, and your staff approve the output before it moves.
Rarely in one place. Part is in a proper system, part is in a note somebody typed at four in the afternoon, and part is in an attachment nobody has opened since.
The threads where something got decided. Usually the only place the reasoning survives, and almost never where anyone thinks to look.
The structured spine of the work. What came in, who handled it, what they did, and when it closed.
The part nobody else can supply. Callbacks, warranty claims, change orders, final invoices against the estimate. This is what turns a record into something a model can be scored against.
Quotes, reports, specs, plans, photographs. Stripped of client, address and price before any of it moves.
Internal procedures, checklists, training notes, the document that explains how your company handles the awkward cases. Short, and worth more per page than anything else on this list.
No records move during onboarding. We work from counts and screenshots until a buyer agreement is in place, and you pay nothing at any stage.
Thirty minutes. Size, industry, and whether your records survived your last system change. Plenty of businesses are not a fit and we say so on the call.
Mutual NDA, then we map what you have and where. You set the exclusions: customers, departments, date ranges, anything under a confidentiality clause.
Our tooling runs inside your network, read only. De-identification happens before transmission. Your staff review a sample and sign off before anything moves.
We take it to buyers. You receive 30 percent of everything it earns, you keep ownership of the underlying records, and you can decline any individual buyer.
Roughly what this costs your staff, in total. Most of it is the review sample in step three. Businesses on older systems, or holding records across several platforms, should expect more. We do the rest.
The six things worth knowing before you spend half an hour on a call with us.
Nothing, at any stage. We are paid only out of what a license earns. If it never sells, you are out the staff time and nothing else.
Ownership of every underlying record. The license is term limited and you keep operating exactly as you do now.
Customer identities, personal data, pricing, and anything that could tie a record back to a named person. Stripped on your hardware.
Your counsel, before you sign. We expect it and we budget time for it, and we send the agreement early so they have room to work.
Any buyer, any department, any customer, any date range. Exclusions are set in the agreement and can be widened later, not narrowed.
A sale. Buyer demand is real but concentrated and slow. A signed license is permission to try, not a cheque, and anyone telling you otherwise is selling something.
Straight answers, including the ones that are not in our favour.
Neither. You are licensing a de-identified copy of how your company works, for a fixed term. You keep the original records, you keep the business, and nothing about your operation changes.
Your customer list, their identities and your commercial terms are out of scope and never leave your systems.
It is removed on your hardware, before anything is transmitted, and your staff approve a sample of the output before the first record moves. We would rather carry the cost of doing that locally than hold personal data at all.
Removing names is only the first step, and anyone who tells you it is the whole job is wrong. Context can identify people too, which is why the scope excludes addresses, pricing, and anything tied to a specific account.
Two groups. AI labs building models that have to handle real professional work, and the data vendors who supply them under existing contracts. Both are short on records of how expensive expert work actually gets done, because none of it is public.
You can decline any individual buyer. We name them before anything is transferred.
Normal, and not disqualifying. Email is often where the useful part lives, because it is where the reasoning happens. Pulling it into shape is our problem rather than yours.
What does hurt is not knowing how things turned out. If you cannot connect a job to whether it came back, the records are worth a fraction of what they otherwise would be.
Honestly, a bit, and we are not going to pretend otherwise. Models are going to get better at this kind of work whether or not your company takes part, and the training data will come from somewhere.
The choice in front of you is not whether it happens. It is whether you are paid when it does. If that trade does not sit right with you, do not do it, and we will not try to talk you round.
The buyer's right to use the data for new training stops at the term end. A model already trained cannot be untrained, which is true of every data license in this market, and the agreement says so plainly rather than implying otherwise.
You keep your records throughout. You are licensing a copy, not selling the original.
Thirty minutes, no files, no preparation. We go through your size, your systems and what you actually keep, and tell you on the call whether it is worth either of our time.
We are not a fit for everyone. We work with US businesses of 20 or more people with at least three years of records, outside healthcare and consumer finance. If that is not you, say so on the form and we will not waste the half hour.
If your size and record history clear the bar, you will get a time to pick from within two business days. If they do not, you will get a short note saying so rather than silence.