Skip to main content

AI at WorkFlex: EU AI Act risk assessments

How we assess and govern the AI features in WorkFlex

Written by David William

WorkFlex uses AI in several features to save you time, from filling in trip details to preparing assignment documents. Every AI feature is assessed against the EU AI Act (Regulation (EU) 2024/1689) before release. This article summarises how each feature is classified and how your data is protected.

Overview of our AI features

None of our AI features is prohibited or high-risk under the EU AI Act. Most are classified as Limited risk, which means you are clearly informed whenever you interact with AI.

  • Business Travel Smart Intake: Limited risk

  • Assignment Explore and AI Document Generation: Limited risk

  • Automated Business Trip Creation: Minimal risk

  • Dashboard Generation Assistant: Limited risk

  • WorkFlex SOS: Limited risk

Principles that apply to all AI features

  • No training on your data: Customer data is never used to train AI models.

  • Processing in Europe: AI processing takes place within the EU/EEA.

  • Humans stay in control: AI suggests and prepares; you review, correct and decide.

  • Transparency: You are told when you are interacting with AI and that its output should be checked.

  • No decisions about people: Our AI features do not evaluate employees, profile them, or make employment decisions.

  • GDPR still applies: Data protection obligations apply independently of the AI Act. Learn more in Data Protection Levels.

Feature by feature

Business Travel Smart Intake

Helps employees submit a business trip by turning their trip description into the answers the form needs: destination, dates, trip purpose, destination company details and, for Switzerland, industry and occupation codes from the official Swiss lists.

  • The employee can review and change every suggested answer before submitting.

  • The feature does not decide travel eligibility or compliance risk.

  • Model: Google Gemini on Google Cloud, processed in the EEA. Feature data is kept for 90 days.

Assignment Explore and AI Document Generation

Helps HR and mobility teams plan an assignment by collecting the details in a conversation, showing source-based salary and benefit options, and preparing the Balance Sheet and Cost Projection documents. Learn more in Set up an assignment with AI (Explore with AI).

  • You choose which salary and benefit options to use. The AI does not select, rank or recommend a compensation package.

  • All calculations are done in Excel, not by the AI model.

  • Names and email addresses are not sent to the model; they are added to the documents afterwards.

  • Models: Google Gemini and Anthropic Claude via Google Cloud Vertex AI, processed in the EEA. Feature data is kept for 90 days.

Automated Business Trip Creation

Creates business trip records from emails sent to a designated company mailbox, so trip details do not have to be typed in manually.

  • The AI only extracts trip information from the email; it does not write new content or make decisions.

  • Model: Google Gemini in WorkFlex's private Google Cloud environment, with caching disabled.

Dashboard Generation Assistant

Lets HR admins create charts and dashboards by asking questions in plain language. Learn more about analytics in Analytics tab in admin view.

  • The assistant only works with anonymised, aggregated travel data such as trip counts, destinations and approval rates. It does not access personal data of individual employees.

  • Model: provided by OpenAI via our analytics partner Luzmo, processed in the EU (Ireland).

WorkFlex SOS

Generates health and security briefings for upcoming trips and translates selected safety alerts. Learn more in Getting Started with the WorkFlex SOS Mobile App.

  • Briefings are based on professional travel risk intelligence, which reduces the risk of inaccurate content.

  • Names are removed from traveller data before it is processed.

  • Model: Google Gemini, privately hosted on Google Cloud.

How we keep these assessments up to date

Each assessment has a named owner in our technology team. We review it whenever a feature changes, for example a new model, new data sources or a different processing region, and reassess it if a feature's purpose expands.

Did this answer your question?