This article provides you with an About AI-powered pay receipt scanning in your implementation of Personio Payroll. 

What is the AI scan function of salary receipts?

The AI-powered scan in Personio Payroll automatically extracts and validates employee and compensation data from the uploaded pay slip. This reduces the manual input effort in the implementation process. 

The feature is available actual Personio Payroll customers based in Germany during onboarding and is designed to make the implementation of Personio Payroll faster and more efficient, especially with a larger number of employees. 

Before you begin 

To use AI-powered scanning, the following prerequisites must be met: 

• You must be stored as an admin in Personio.
• Make sure that the AI features are enabled in Personio. If the AI is disabled, you won't see the option. 

To deactivate the AI features, do the following: 

1. Open the pane Settings > AI features
2. Check the box General Access under AI Features

What information is extracted?

• First name (civil)
• Last name (civil)
• Birthdate 
• Payroll ID
• Street and house number
• Social security-relevant data such as social insurance number, the statutory health insurance fund (collection office), person group key and contribution group key
• tax-relevant data such as tax identification number, tax class, child allowance and church affiliation

Further information on which data is contained in a remuneration certificate can be found in our Articles on the theme. 

Please note:
The AI scan can extract a variety of attributes required for the Persion Payroll implementation from the pay certificate. Depending on the document content and quality, however, it may happen that individual attributes are not recognized and need to be added later. 

AI-powered scan of pay slips

Uploading documents 

When uploading the Documents, you have the following options:

1. Individually:One document per employee
2. Mass import: A PDF document containing the pay slips for multiple employees

Personio automatically divides the Documents and assigns them to each employee as needed. Make sure that you only upload salary statements and no other billing documents. 

To upload Documents and import data, do the following:

1. Go to the onboarding portal in the Payroll accounting and select the appropriate legal entity.
2. Click Upload payslip of the people. 
3. Drag the Documents directly intoDrop file here or click Browse files and upload it from your device. 
4. Switch to the tab To be assigned. If necessary, assign the Documents to the employees and check the extracted data for accuracy. 
5. Switch to the tab To import and 
6. Click Import and confirm with Start import, to upload the data to the employee profiles. 

After all salary statements have been processed, you can also import the attributes contained in bulk. A manual check is recommended, but not mandatory. Attributes for which the system is not sufficiently secure are held for manual testing.

Please note:
The assignment only shows employees who are assigned to the selected legal entity. 

Review and edit data

Review and approve data 

To review and approve the extracted data from the pay statement, do the following:

1. Switch to the tab To be assignedto view the uploaded documents. 
2. Open the tab Excluded 
3. Check data. You can approve the attributes individually or clickApprove all to confirm all the information. 

Edit data

You have the option to edit the extracted data at any time. This involves the following steps:

1. Open under To be assignedthe uploaded document
2. Click on the pencil icon.
3. Enter the new value. 

Decline files 

In the case of documents uploaded incorrectly or accidentally, you have the option of rejecting them. This involves the following steps:

1. Switch to the tab To be assignedto view the uploaded Documents.
2. Open the relevant document.
3. Clickpayslip decline and confirm with Decline. 

FAQ

What document types are supported?

Salary receipts can be imported individually per employee or in a bulk import for several employees. This can be done in the form of a PDF file. Before doing so, make sure that there are no other pay documents in the file.

What happens if a salary certificate cannot be assigned to an employee?

First, check if an employee profile already exists. If not, create the employee again and then assign the salary certificate to the profile. Also, check that the employee is assigned to the correct legal entity. If there are still problems, contact our Payroll Specialists as the payroll peoples responsible. 

What if the extracted data is incorrect?

You can edit and customize all the extracted data at any time before Personio upload the information. If you have uploaded an incorrect document, you can decline it. 

Can I feature disable?

Yes, customers can Personio disable all features at any time. The feature is purely optional and is only available for use if the customer active chooses to use it. 

What data does the AI-supported scanning function in Personio access?

The AI-powered scanning feature only accesses data uploaded to Personio by the customer, as well as data that is already in the employee profile existing. 

Does the AI-powered scanning feature meet the access requirements in Personio?

Yes, the scan function implements the access requirements set up in Personio. The feature is only available to people who have editing rights in Personio Payroll. The data used is already accessible to these users. 

Who owns the personal data and the output produced by the scanning function?

All expenses generated by the AI scanning function are based on the inputs provided by the employer. These outputs are considered customer data, regardless of whether the inputs contain personal data or not.

How is the AI scanning function technically set up?

The AI scanning feature is based on Anthropic's LLM Claude. Claude LLM is hosted by Personio on its AWS servers in the EU using AWS BedRock.

Does AWS BedRock or Claude store personal data?

No, personal data is not stored by AWS BedRock or Claude.

Does the data leave the EU?

No, the LLM used for the AI scanning feature is hosted by Personio from its own AWS server using BedRock. The same offices and hosting scope apply as for AWS. 

Is personal data used for training purposes?

No, neither input nor output data provided by the employer will be used for training purposes by Personio or any third party. 

Is the use of the feature mandatory?

The use of the AI scanning function is optional. Customers also have other options for importing employee data as part of the Personio Payroll implementation. The AI scan of the pay slips is offered as an additional tool to facilitate the process. 

Does Personio store data related to the use of the AI scanning feature?

Yes, stores both input and output data from the AI scan for quality assurance and to improve the feature. This data is not intended for training purposes. 

Will the AI scanning feature make decisions for me?

No, there is no such thing as AI-based decision-making. The feature only supports customers in extracting employee data. The customer still retains full control over what is stored in Personio. The proposed attribute values still need to be reviewed and approved. Recommended attribute values can be edited and deleted by the customer at any time.

Who is responsible for the accuracy of the data?

The AI scan only accesses data provided by the customer and uploaded to Personio. Customers are therefore solely responsible for ensuring that the data that AI uses to create attribute values is correct. In any case, the extracted information should be checked.

Are there any risks of using the AI scanning feature?

Artificial intelligence and machine learning have both known and unknown risks and limitations. Customers are solely responsible for: 

1. Develop your own internal policies on the appropriate use of the AI scanning feature and train other users in their accounts on these guidelines. 
2. Provide transparency and explainability notices and obtain all necessary approvals required by applicable law and internal policies, and 
3. Implement sufficient human control for the use of the AI scanning feature.