Open-Source Tool for AI-Assisted Survey Classification

Classifying text data data into standard codes like ISCO, ISIC, or COICOP is one of the most neglected yet error-prone tasks in survey and census operations. When rowsquared built a machine learning assisted classification system for the 2022 Maldives Census — processing 242,000 descriptions into ISCO and ISIC codes — the scale of the problem became clear. A decade of encountering the same challenges across countries and surveys motivated us to build classiflow: an open-source, AI-assisted tool that turns ad hoc classification into a structured, quality-assured workflow.

classiflow combines a lightweight AI model that adapts to any hierarchical taxonomy and runs entirely on the user’s own infrastructure with a workflow management interface where teams upload data, assign sentences to domain experts, set up double-coding, and bulk-label hundreds of similar descriptions at once. The AI improves incrementally as reviewers correct its suggestions, and its confidence-calibrated predictions ensure that uncertain cases are always flagged for human review.

The tool also connects to live fieldwork. In Grenada, classiflow’s API classifies incoming Survey Solutions interviews in real time; when the model cannot classify below a certain level, the entry is flagged for immediate follow-up with the field team — closing the feedback loop between data collection and classification quality.

classiflow was first deployed in Grenada and Saint LuciaSt. Vincent and the Grenadines as part of rowsquared’s survey support to the OECS Data for Decision Making project, and has since been adopted independently by other National Statistical Offices across the globe. It supports over 100 languages, handles taxonomies up to five levels deep, and is entirely free and open source.

More details and a demo booking option are available at rowsquared.com/classiflow.