For AI data teams
Whole case histories, from request to outcome
Emails, quotes, tickets, revisions and booked hours from real jobs, sourced from European businesses, including companies in insolvency proceedings. We're starting with German records.
- Outcomes in the record Where the records show it, each case ends with what actually happened: the quote won or lost, the fault fixed or reopened.
- A written agreement with every source You also get a written licence for your use.
- Names replaced by codes You never get the raw data or the code key.
- Your spec comes first Tell us what you need before we buy. Sending a spec doesn't commit you to anything.
What public data leaves out
Ground truth from the records
Where the records include it, each case has its actual outcome, such as hours booked against hours quoted. Cases without one are flagged. You can check a model's prediction against what actually happened.
The whole thread, linked
Request, follow-up questions, revisions and result, all in one case, and across systems where the records allow.
Only from the source
Records like these rarely turn up in public data. They sit in company systems, and when a company closes they're often archived or deleted.
What one case contains
We look for case histories from any sector, linked from request to outcome. In an archive, that means records like these.
A manufacturing job
- Request for quote
- Quote
- Engineering changes
- Booked hours
- Inspection results
- Delivery
- Complaint
A services project
- Brief
- Proposal
- Change requests
- Tickets
- Delivery
- Client sign-off
Example
Names become codes,
but content and connections stay
Exact dates are coarsened, and records stay in their original language. Your sample will show how this affects the fields you need.
Anfrage Tragwerksplanung
Sehr geehrte Frau Musterfrau,
für den Anbau an unsere Lagerhalle, 24 × 40 m, bitten wir um ein Angebot für die Tragwerksplanung. Die alte Statik liegt uns leider nicht vor.
Mit freundlichen Grüßen
Max Mustermann
Mustermann Logistik GmbH
Anfrage Tragwerksplanung
Sehr geehrte Frau P-07,
für den Anbau an unsere Lagerhalle, 24 × 40 m, bitten wir um ein Angebot für die Tragwerksplanung. Die alte Statik liegt uns leider nicht vor.
Mit freundlichen Grüßen
P-12
K-0412
Because K-0412 stands for the same company in every document, this request stays linked to what followed: the quote, two revisions, the order and 436 hours booked against 380 quoted.
Your spec first, and a sample before any licence
Your spec guides which archives we buy and how we prepare them. Once we've prepared a one-off archive, we delete the raw data, so anything we didn't prepare can't be added later.
You send a spec
What the data is for and what it needs to contain.
We look for a matching archive
We approach businesses with suitable records, and insolvency practitioners handling companies that have them.
We prepare the cases
We link documents into cases, mark how each case ended and replace names with codes.
You review a sample
The source approves it first. You then review it under a short evaluation agreement: you use it only to evaluate, don't try to re-identify people or businesses, and delete it afterwards.
What your legal team will ask
- One source, one written agreement Each archive comes from a single company. The company sells it – or, in an insolvency, its administrator or liquidator does, with any approvals needed. The agreement sets out what's shared and how it may be used.
- Insolvency sales are 'as is' Sellers in insolvency proceedings give us no warranties. Any warranty we give you is set out in your licence.
- A written licence Granted by Florian Findeis personally, for the use we agree: training, evaluation or environments. It's non-exclusive unless we agree otherwise.
- No other data buyers A company in insolvency proceedings sells a copy of its archive only to us – no other data buyer gets one. The originals stay with the company. This doesn't bind anyone who later buys the business.
- Who can license the data Companies that develop or test AI, and data providers working for them. If a source's contract limits who can receive the data or how it's used, we'll tell you before you see a sample.
- Outside the UK and EEA Recipients there only get data under a recognised transfer mechanism, such as standard contractual clauses.
- Pseudonymised Names, companies and contact details become codes, and identifying details are removed or coarsened. The datasets we deliver are treated as personal data under the GDPR and, for UK records, the UK GDPR.
- No raw data, no code key You never get either. Our contract with the source sets out how long we keep the key.
- The lawful basis comes first Before any data changes hands, each party's role and the lawful basis under the GDPR (and, for UK records, the UK GDPR) are set out in writing. Our agreement with the source isn't a lawful basis in itself.
- Sensitive records stay out HR files, job applications and payroll; health and other special category data; employees' private emails and chats; and records held by doctors, solicitors and others with a professional duty of confidentiality. Working hours appear only as totals per job.
- Other people's material stays out too Customers' plans, manufacturers' manuals, third-party code and documents under a confidentiality agreement. Each source can leave out more. If anything that should have been left out reaches us, we delete it rather than prepare it.
- What you agree to You won't try to re-identify people or businesses, and you'll use the data only for the agreed purpose. You'll pass it on only if the source's contract allows, and only to someone who takes on the same duties.
About Bayeset
We bring AI teams the business records that public data leaves out
Bayeset buys copies of business records from UK and German companies, including closed and insolvent ones, and prepares them for AI developers. We license them for training, evaluation and environments, and we're based in Munich.
Our background
Questions data teams ask first
Do you have data we can review today?
Not yet. Your spec can shape the first archive we buy.
Which records and languages?
Business email, documents, tickets, CRM, ERP and project files, often spanning several years. We document the source type, period and systems for every dataset.
We're starting with German-language records from German insolvency proceedings. We're also approaching UK companies for English-language records. Tell us which other European languages you need and we'll look for them. The records reflect European practice: VAT, fee scales, local forms.
How do we know an outcome is real?
It comes from the source's own records: the CRM status, the ticket log, the hours booked. Cases that end without an outcome – for example, because the company became insolvent – are marked. You can check this in the sample before taking out a licence.
Can a source take the data back?
A one-off archive is final once the source has approved the sample and been paid. A company that supplies data on an ongoing basis can stop future deliveries, but datasets already licensed to you stay licensed under your contract.
Individuals' rights under the GDPR still apply. Your contract says how we pass on their requests.
Who is the contract with?
Florian Findeis personally. If he later sets up a company, your contract will only pass to it with your agreement. This page describes how Bayeset intends to work; only a signed contract is binding.
Send us your spec
A few lines will do: what the data is for, record types, languages, fields and rough volume. We keep your spec confidential and never name your organisation without your permission. Sending one doesn't commit you to anything.
We never send data or access details by email. We only send a sample once the evaluation agreement is signed, and only in the way the agreement sets out. If a request claiming to be from Bayeset seems odd, call us to check.