AI recommends. People decide.
At AddUp, AI and agents work at every stage of the AddUp Loop: they analyze, connect information and propose the next step. Decisions that affect people are always made by a person.
What the AI does, and what it doesn't.
Each agent has a defined level of autonomy, based on the impact of what it does.
Only tasks that don't affect people, such as organizing and analyzing comments. The support assistant hands off to a person whenever in doubt.
Diagnoses, reports, action plans, pay calculations, data loading and quality, and survey design.
Follow-ups and reminders, always within the rules your organization sets.
Not at AddUp.
Decisions about people.
In pay and performance, AI never decides
It recommends, a person decides, and the decision is logged with its reasoning.
The reasoning stays visible
Every recommendation shows the factors and data behind it.
A person is always accountable
If the system gets it wrong, the recommendation never reached anyone without first going through approval.
Your data.
- Neither AddUp nor the AI provider we use trains models on your data.
- Analysis happens inside the platform: you don't need to move your data to an outside chat.
- Cloud infrastructure, encrypted in transit and at rest.
- Role-based permissions and audit trail.
- SOC 2 in progress.
Regulatory framework.
AddUp is designed with the European Artificial Intelligence Act (Regulation (EU) 2024/1689) in mind, which treats AI systems used in employment decisions as high-risk, as well as the personal data protection laws of the countries where we operate.
Does the AI make decisions about people?
No. In pay and performance, the AI recommends and a person decides. Every decision is logged with the reasoning behind it.
Who is accountable if the system gets it wrong?
Always a person. No agent makes decisions that affect people on its own: its recommendations go through approval and are logged, with the factors behind them in plain view.
Can I query AddUp from ChatGPT, Claude or Copilot?
Yes. Through MCP, the open standard for connecting AI assistants, each person can query AddUp results, reports and plans from the assistant their organization uses, such as Claude, ChatGPT or Copilot. They see exactly what they see in the platform, never more: the same permissions apply, the assistant can only read and never changes data, and every query is logged.
How do you prevent bias in recommendations?
We audit for bias regularly, especially in performance and pay, to detect whether a recommendation treats people differently based on gender or age. And in those decisions the AI recommends, but a person always decides.
What happens to sensitive data in comments?
Before any information is sent to the AI and before any report is published, AddUp masks sensitive data such as names or health information.
What happens if the AI provider changes its model?
Every model change first goes through an automated evaluation set that checks answer quality. The architecture also allows switching providers if needed.
Can a comment manipulate what the AI says?
No. At AddUp, survey comments are always data, never instructions, and we test manipulation attempts before every release.
Do you use our data to train models?
No. Neither AddUp nor the AI provider we use trains models on your data.
Tell us how you do it today. We'll show you what it looks like in AddUp.
30 minutes, no strings attached. You'll talk to the people building the product, not a sales rep.
Book a demo