Insight Builder

At the core of a good online dashboard lies a solid structure. As a research consultancy firm, we excel in designing dashboards based on research questions and policy theories. Additionally, we conduct user research to determine the insights users of a dashboard require. However, not all needs can be predetermined – insights often generate new questions. With our Insight Builder, we provide a way to unlock more information and knowledge on a dashboard. Users are empowered to ask new questions based on the data, essentially enabling them to carry out their analyses. This allows a user to specify an analysis to a specific subset of a dataset. They can also modify the display format (e.g., switching from a bar chart to a line chart, changing the display period, etc.). A challenge is to prevent these new analyses from producing inaccurate or incomplete results. Furthermore, it is crucial not to overwhelm users with choices and information. An example of a catalogue-based Insight Builder can be found at pr-edict.nl. Users can explore a wide range of analyses in the fields of ICT and the labour market, with easy filtering options available. We offer two versions of the Insight Builder. In the first version, our researchers prepare numerous analyses in a catalogue. Users can easily browse this catalogue and apply filters to the analyses. Searching is based on natural language and a touch of AI. If a user searches for a term that does not exactly match a keyword on the dashboard, the system can still display the correct content through intelligent matching. Another version of the Insight Builder is the free analysis: users can ask a free-form analysis question in plain language. The dashboard will then generate an analysis (chart) where possible that answers the question. To achieve this, we feed an AI detailed information about the dataset, underlying methods, and assumptions. The AI translates the question into a query (database question), a proposed visualisation format, and an explanation. The AI can also check constraints and indicate limitations of the analysis. Additionally, strict conditions can be implemented, such as minimum cell fill requirements. [Images are not included in the translation to be displayed here.]