Civinc

Municipality of Amsterdam

The use of the Civinc application, which enables staff to engage in anonymous discussions and carries out sentiment analysis using generative AI.

Last change on 2nd of July 2026, at 15:17 (CET) | Publication Standard 1.0

• Publication category: Other algorithms
• Impact assessment: The Ethical Guide
• Status: In use

General information

Theme

Organisation and business operations

Begin date

2022-07

Contact information

[e-mail]

Link to publication website

https://civinc.co/nl

Responsible use

Goal and impact

Civinc is a new application that enables staff to exchange views on key organisational issues in a safe manner with colleagues who have a different perspective. This takes place in anonymous one-to-one chat conversations. In structured sessions, employees with differing perspectives can be brought together on a large scale to engage in dialogue, free from social desirability or peer pressure.

On the one hand, the app gives large groups of staff a voice and enables the organisation to discuss important (and sometimes sensitive) topics in a safe and inclusive manner.

On the other hand, the application provides insight into staff sentiment, enabling the organisation to stay in touch with what is on people’s minds and thus prevent ‘blind spots’. An AI analysis of the anonymous chats maps out how employees view certain topics, what ideas emerge in this context, and where the tension lies between different perspectives.

Considerations

This algorithm provides the organisation with insight into staff sentiment regarding key organisational issues. Generative AI makes it possible to analyse textual data on a large scale and process it quickly, in a way that would not be feasible or cost-effective using any other type of (manual) analysis. Furthermore, the use of generative AI ensures that participants’ chats do not need to be viewed, which is desirable from a privacy perspective.

The risk of bias or inaccuracies in the analysis is mitigated by a quality assurance process that periodically verifies whether the analyses carried out are sound and reliable. 

Human intervention

The algorithm’s output is always checked by a human. Furthermore, the output of an analysis in this application must never be taken in isolation, but must be viewed and assessed within a broader context of data points and experience. 

Risk management

To ensure that the algorithm meets the high standards for information security, comprehensive compliance processes are carried out.

Elaboration on impact assessments

A DPIA is not required because users only share personal information on a voluntary basis in the anonymous chats or open text fields. Users are always advised against doing so.

A human rights assessment has been carried out in the form of an Ethical Information Sheet, based on the Amsterdam model.

Impact assessment

The Ethical Guide

Operations

Data

The (generic) model was trained using data from OpenAI. This does not involve access to the dataset used.

 https://platform.openai.com/docs/models

Technical design

GPT-4(o) is a language model developed by OpenAI. It uses a transformer architecture, which is based on deep learning. The model is trained using large amounts of text data and, in doing so, learns to understand the structure of language. It can recognise context, make predictions about which words are likely to follow, and generate meaningful responses based on the text entered.

In the context of the Civinc application, specific instructions (‘prompts’) are given to the language model to provide relevant analyses of the aggregated, anonymised chats within the application.

External provider

Civinc is the provider of the application, which incorporates OpenAI’s algorithm (GPT-4-o). For more information about Civinc, visit www.civinc.co/nl/; for more information about OpenAI, visit www.openai.com.