Guides
What is process data? The traces of how work moves through your tools
By Francesco Wiedemann and Peter Meier · · 3 min read
In short
- Process data are the traces of how work moves through your tools: a pricing decision that goes from the CRM through a finance check in the ERP to a closed ticket.
- Companies produce process data anyway while they work. It shows real work across several tools, with corrections and a result.
- AI labs license anonymised process data to train and evaluate AI agents that work in real business software.
- Licensing process data is not selling customer or employee data: data processed for customers is excluded, and people are anonymised before anything is licensed.
Definition
Process data are records of how a piece of work moves through a company's tools from start to finish: who started it, which systems it passed through, what was corrected along the way and what the result was. Typical sources are CRM, ERP, ticket systems, code repositories and internal knowledge bases.
In German, the licensing of anonymised process data for AI training is called Prozessdaten-Recycling: the data already exists, and it is used a second time, as training data.
An example
- Sales proposes a price increase in the CRM.
- Finance checks the margin in the ERP and asks for a larger step.
- The new price is entered in the CRM.
- A ticket to update the price list is closed.
Each step on its own is an ordinary record. Together they show a complete decision across four systems, including a correction. That connected trace is what makes process data valuable.
Why AI labs want process data
AI models have learned a great deal from public text, but very little about how professional work actually gets done. AI agents are now expected to work in CRM, ERP and ticket systems. To train and test them, labs need realistic records of multi-step work with outcomes, which only companies produce.
What process data is not
- Not customer data you process for others. Data you handle on behalf of customers under a data processing agreement is excluded.
- Not personal profiles. Names are replaced, details generalised or removed, and the result is checked before anything is licensed.
- Not machine data (in this context). In manufacturing, "process data" can also mean sensor data from machines. Here it means records of business workflows.
How licensing process data works
Licensing follows four steps: check eligibility, clear the rights, anonymise and refine in the EU, then license and earn. Details are in our guide to licensing company data for AI training and our GDPR guide.
Does your company produce licensable process data? The eligibility check takes about a minute and needs no data upload.
Frequently asked questions
Is process data the same as operational data?
Largely, yes. "Operational data" is the broader term. Process data focuses on how work flows across tools, which is the part AI labs value most.
Which companies produce useful process data?
Any company that runs its work through software such as CRM, ERP, ticket systems or code repositories. IsarTrust currently works with companies of 25 or more employees with at least three consecutive years of data.