Labs are no longer training models to describe work. They are
training them to actually do the work.
01
Text is spent
The open web has already been trained on. There is no second internet.
02
Agents need practice
A model cannot run a process it has never watched run, and that record is not
published anywhere.
03
Europe is missing
Today's models learn from American, English-first data. Europe needs agents built on how
it actually works.
The record of how a claim is assessed, how a close is run, how an exception is escalated
is not on the internet. It is in your systems, and nowhere else. That is what a lab
cannot buy anywhere but from a company like yours.
Who is buying, and why now
01
Frontier labs
Building agents meant to work inside European companies, in European languages, under
European rules.
02
European model builders
Sovereign programmes that cannot train on what American firms already hold.
03
The window
Operational data is scarce today. Once enough companies have licensed theirs, it will not
be.
We publish the benchmark that measures the gap:
EuroExec.
03OPERATIONS
We're looking for everyday operations data, not your trade secrets.
You set the scope, and anything you consider your edge stays out of it.
01
Judgement
Your specialists grading model output in their own field.
02
Routine
How a task is done here, step by step.
03
Exceptions
What your team does when the routine does not fit.
Sectors
Finance
Closes, reconciliations, controls.
Legal
Contract review, policy, governance.
Health
Protocols, case handling, submissions.
Industry
Procedures, fault histories, fulfilment.
Software
Design decisions, incidents, reviews.
Support
Tickets, escalations, quality reviews.
Where these records already live
You do not need a data project. In most companies this already sits in five
or six systems.
Process and policy
Handbooks, SOPs, work instructions, internal wikis.
Not this: personnel files, customer records, credentials, and anything that
identifies a person. Those are excluded before anything moves, and section 05 sets out how.
04HOW
Four steps. You reserve the right to stop after any of them.
Nothing commits you until the licence.
STEP 01
Assessment
A call and a look at what you hold, at no cost.
STEP 02
Legal and governance
Basis, scope and exclusions are agreed upon and written down.
STEP 03
Extraction and preparation
One export from the systems you named, prepared entirely on our side. No identifiable
data ever reaches a buyer.
Names, contact details, account references and identifiers buried in free text are
stripped out before anything else happens.
An automated pass runs first, then a named team in Europe reviews the result by hand.
Anything that cannot be safely anonymised is excluded rather than shipped.
Raw records stay with that team, inside Europe. A buyer receives the prepared dataset
and nothing else.
You see what the dataset contains and sign it off before it is licensed.
STEP 04
Licence and payment
The dataset is licensed and you are paid.
Step 1 of the data partnership process.
Step 2 of the data partnership process.
Step 3 of the data partnership process.
Step 4 of the data partnership process.
05HANDLING
What we take, and what we never touch.
Stays with us
Raw records
A named team in Europe, never a buyer.
Identifiers removed, then checked. Anything that cannot be anonymised is excluded.
Goes to the buyer
Anonymised dataset
No person, client or account is identifiable in it.
Raw records stay with a named team in Europe. Only the anonymised dataset is
licensed.
We take
One export
From the systems you name, once per delivery.
We never take
Anything personal
Customer data, personnel files, credentials.
Anonymising
Stripped, then checked
Identifiers go automatically, then a person reviews it.
Raw records
Our team only
A named team in Europe sees them, never a buyer.
Your time
A call and a sign off
The preparation work is ours.
Control
You approve scope
Nothing is included until you say so.
You keep ownership
You licence a use, you never sell anything.
You set exclusions
Name anyone your data may not reach.
You can stop
End future contribution on notice.
06ALSO YOURS
The money is not the only thing you get back.
01
An operations map
You see where the real process differs from the written one.
02
A provenance record
Your EU AI Act paperwork, done, and yours to keep.
03
Paid review work
Your experts are paid separately for grading output.
07EUROPE
Europe buys AI. Now it can build it.
Your records are how the next model learns the way Europe actually works.
Lawful basis
Documented before anything moves, in a form your DPO signs.
European processing
Handled by EU entities under EU law.
Your works council
Consulted before contribution starts.
On the record
Every scope decision written down as it is made.
08QUALIFY
Most operating companies do.
01
Five people up
Small specialist teams count.
02
Written records
Something a new joiner could learn from.
03
European
EU, EEA or UK.
Not sure? The six questions in section 01 will tell you.
Know a company that should do this?
Introduce them, and you earn a share of their first licence.
Because models are now being trained to do jobs rather than describe them, and the record
of how a job is actually done inside a European company is not on the internet.
Which systems does it come from?
Process and policy documents, case and ticket systems, document stores, approval and
control records, line-of-business exports, and issue trackers. Section 03 lists them, and
you name which of them are in scope.
How is the price set?
By how much you hold, how far back it goes, how unusual your field is, and whether the
licence is exclusive.
Does it cost us anything?
No, and nothing is deducted from what you are paid.
Who sees our raw records?
A named Sovrano team in Europe, never a buyer.
When are we paid?
By bank transfer against the licensed dataset, on the terms you signed.
Does customer personal data go anywhere?
No, it is removed first, and anything that cannot be anonymised is excluded.
What is the lawful basis?
We document it for each category and your DPO reviews it before anything moves.
Do we lose control of our data?
No, you keep ownership and licence one defined use.
Could a competitor get it?
Only if you let them, and exclusions go into the agreement.
How much of our time is it?
A call and a sign off, plus paid time if your experts grade output.
What about our works council?
It is consulted before any contribution starts.
Can we stop?
Yes, on notice, and delivered datasets stay under their agreed terms.
10GET IN TOUCH
Tell us what you hold.
A call, a written view of what is licensable, and a number. At no cost.
Received.
We will read what you sent and reply from marcus@sovrano.ai with a written view of what
is licensable, and a number.