A limited request
The model receives a request and the context needed for the task. It does not have unrestricted access to client company databases.
Privacy and AI
Data is processed for a defined task.
Payroll needs personal data. Not everything in the service needs to be shared with AI. The application limits the language model’s task and the information it receives.
Rule-based automation handles much of the work.
Payroll calculation, checks and data transfers largely use conventional software automation. A language model helps interpret a request; the application calculates results and controls access.
A checked action
The application checks company scope, access rights and permitted actions. A model response does not itself grant permission to read or change data.
Calculated results
The application calculates pay amounts and report figures. Workflow approvals and exception handling remain with the people responsible.
A local model in our own server environment.
The employee assistant and analytics query routing use the Ministral model from European company Mistral AI, running in a server environment under our control. Model requests for these functions are not sent to an external language model service.
Local model inputs are limited too
An employee’s name may be needed to interpret a request. Technical checks restrict personal identity codes and IBAN account numbers from model inputs. These details are viewed and edited in personnel screens protected by access controls.
Analytics figures come from the application
The local model helps choose a report and its filters. Financial data rows and calculated report results are not sent to the model for explanation.
Local means a server environment controlled by AI Payroller, not the user’s device. The model’s European origin and the location of data processing are different matters: how we run the model determines processing.
Privacy covers the whole processing chain.
Primary servers in Finland are one part of the picture. Service providers, integrations, access rights and the data lifecycle also matter.
Who is responsible for employee information?
An employer generally acts as the controller of its employees’ data. AI Payroller processes payroll data on the customer’s behalf. Roles and the processing chain involving an accounting firm are specified in customer agreements.
What is agreed before onboarding?
A data processing agreement defines responsibilities and processing instructions. We review providers, processing locations and the principles for retaining, returning and deleting data.
How can employees request a review of their data?
Employment and payroll data requests should go to the employer first. For questions about privacy in the service, contact hello@aipayroller.com .
This page describes the payroll service’s operating principles. It does not replace a data processing agreement. AI Payroller Oy’s own registers are covered in the general privacy notice (in Finnish) .
Security and continuity in practice.
Explore the server environment, access controls and how we prepare for disruptions.