Departments · HR AI in HR: Onboarding, Documents and Employee Questions

HR tasks

Not every recurring task needs AI. If a form simply transfers data from one field to another, a fixed rule is often enough. AI is most useful when a process involves reading text, finding relevant documents or preparing drafts from different pieces of information. The examples below describe typical processes, not client projects. Each process needs defined inputs, a responsible person and a traceable outcome.

1. Assemble onboarding documents

Onboarding a new employee requires information from several teams and systems. The documents required depend on the role, location and internal policies.

• Input: a confirmed new starter notification containing the details needed for the process, along with approved templates and checklists.
• AI prepares: the relevant documents and steps, flags missing information and produces an overview for the teams involved.
• HR reviews: whether the documents, details and responsibilities are right for the individual, then approves their distribution.
• Outcome: a reviewed onboarding overview showing the required documents and outstanding points.

The process may also create tasks for IT or other departments. This does not mean that AI creates accounts or assigns access rights itself. Those steps remain with the people responsible for them. If templates change, it must be possible to identify which version was used for a particular case.

2. Answer questions about leave, expenses and policies

Many questions within a company relate to the same internal rules. Even so, a general answer may not be appropriate in an individual case.

• Input: a question submitted through a form, the intranet or email, together with the approved internal policies relevant to it.
• AI prepares: a draft answer based on the relevant passages and identifies its sources. If there is no clear basis for an answer, it flags the question as unresolved.
• HR reviews: whether the sources and answer are current, and whether individual circumstances require further clarification.
• Outcome: an approved answer or a question referred to the responsible person.

Such an assistant should not answer questions using arbitrary files. General questions can often be answered from policies without access to personnel records. Individual enquiries need a separate process with tighter permissions. How an AI agent fits into that process depends on the existing systems and approval arrangements.

3. Prepare references and certificates

Documents with a recurring structure can be prepared from approved templates. HR remains responsible for their content.

• Input: a documented request, the information permitted for that case and an approved template.
• AI prepares: a draft using the template and flags wording or required fields that need clarification.
• HR reviews and approves: the details, wording, recipient and final version. Relevant specialists are consulted on legal questions.
• Outcome: an approved document or a draft with specific questions to resolve.

A polished draft of an employment reference must not be mistaken for a factually verified statement. AI receives only the data needed for the particular draft. It does not independently assess the person concerned.

4. File contracts and documents correctly

A digital personnel file is useful only if documents can be found and are not placed in the wrong file. Classification can be prepared automatically, but it should be checked.

• Input: a received document and the defined filing structure and document categories.
• AI prepares: a proposed document category and filing location, flagging missing details or uncertain matches.
• HR reviews and approves: the person, document category, visibility and filing location before the document is stored.
• Outcome: a filed document with a traceable approval, or a case referred for manual clarification.

Fixed validation rules also matter here. For example, a document should not be assigned to someone solely because it contains a similar name. Read more about processing incoming documents under document processing.

5. Acknowledge receipt of an application

For job applications, we limit automation to administrative steps. Assessing applications is not part of this process.

• Input: a new application received through the designated channel.
• AI prepares: checks whether the contact details needed for an acknowledgement are present and drafts an acknowledgement using an approved template.
• HR reviews: the template, the intended sending method and cases with unclear or missing information.
• Outcome: an acknowledgement of receipt or an application flagged for clarification, without assessment or ranking.

Data access and retention must be addressed even for a limited process. An acknowledgement must not become a means of screening applicants indirectly. AI does not produce suitability assessments, scores or recommendations about individuals.

6. Prepare leave requests for review

A request may arrive through an app or an existing system. AI can organise the information, but it does not decide an employee's entitlement.

• Input: a leave request specifying the dates, the person concerned and the information required by the company.
• AI prepares: a check for missing fields, a summary of the request and its routing according to defined responsibilities.
• HR reviews: missing details and cases needing individual clarification. The responsible person decides on the request.
• Outcome: a complete request in the designated approval process, or a follow-up question.

We assess the specific process to decide whether a simple form rule or AI is appropriate. AI is often unnecessary for capturing and forwarding standardised fields.

Systems and integrations

HR information is usually held in several systems. A new process should not replace them without good reason. The first step is to establish which system is the authoritative source for each piece of information. We can then assess whether an existing interface, a controlled file exchange or another connection is appropriate.

The named systems are examples, not claims of a partnership or of specific product features. Whether and how they can be connected must be assessed for your systems. Exceptions matter too: what happens when information conflicts, a permission is missing or a system is unavailable? Automation should make these cases visible rather than proceed without warning.

Boundaries, approvals and co-determination

We do not build AI for selecting candidates or automatically assessing employees. Annex III of the EU AI Act identifies certain AI systems used in employment and workforce management as high-risk. Our services in this area are therefore limited to supporting processes such as assembling documents, finding approved information and sending administrative acknowledgements. The legal classification of a specific system must be assessed separately. Read more about the framework under the EU AI Act.

Approval is more than a button at the end of a process. It must be clear in advance who may see a draft, who corrects factual errors and who makes the final decision. HR reviews personal documents and individual answers before they are shared. Unclear information is not filled in by guesswork. Changes to the process should be documented and reviewed before use.

Introducing a system may engage works council co-determination rights. In Germany, Section 87 of the Works Constitution Act (BetrVG) needs to be considered; in Austria, Section 96a of the Labour Constitution Act (ArbVG). Whether consent is required and what agreement is needed depend on the organisation and the specific system. Employee data protection and existing internal policies also form part of the review. This is not legal advice. Legal decisions remain with the responsible legal professionals.

Data: personnel files and permissions

Employee data requires tighter access controls than general policies. We therefore separate questions about company-wide rules from matters concerning a particular person. An assistant answering questions about leave policies does not need to read personnel files. If it supports an individual case, it receives only the data and functions approved for that purpose.

Before implementation, roles and permissions are defined: who may ask a question? Who can see the draft? What data may a system read or write? Logs may also contain personal information, so they need their own access rules. Access that is technically possible is not necessarily needed for the task. See AI and data protection for planning considerations.

Data planning also covers retention and deletion. Incoming requests, drafts, completed documents and logs may need different rules. When an external service is connected, we assess which data is transferred and where it is processed. If your company requires servers in the EU, that requirement must be taken into account when selecting and configuring the technical environment, and verified for the specific solution. Server location alone does not replace a review of access and contracts.

There is no single answer to whether digital personnel files are mandatory; it depends on the country. The applicable jurisdiction, documents concerned and retention requirements all matter. We design the technical process. Your responsible specialists determine the legal position. This is not legal advice.

Bespoke apps for employees

Sometimes what is missing is not another HR system but a straightforward way to access an existing process. A bespoke employee app could, for example, capture a leave request, show its status or provide a digital document folder. A suitable interface could also make a work schedule accessible. Appropriate and permissible functions depend on the existing systems and internal rules.

Such an app should complement existing systems rather than create duplicate data stores. A request can start in the app and then be handled in the responsible system. The app shows only information the signed-in person is authorised to access. Decisions on requests and changes to employee data continue to follow the defined approval process.

Forms and fixed rules are often enough for clearly defined inputs. AI is added only when it supports a specific step, such as summarising a follow-up question clearly using several approved sources. We therefore assess separately what a bespoke app should do and where AI is actually needed.

Where to start

A useful first step is a short review: which questions does HR answer repeatedly each week? Which documents do people keep having to find? Where is the input clear, but assembling or checking the information takes work? Choose a process with known sources and a clearly responsible person.

We then review with you what data is needed, which systems are involved and who approves drafts. The opposite question matters just as much: what must the system explicitly not do? In HR, this includes assessing people, accessing files without permission and making decisions without a responsible person. The outcome of this review provides a basis for technical implementation, not a promise of a particular AI result.

We are based in Vienna and work for companies across Europe. If you would like to discuss a specific process, please use the contact form.