The team needed a scalable way to gather and enrich large volumes of web-based text data. Their goal was to train machine learning models that could accurately predict consumer trust in brands but success depended on building high-quality, labeled datasets that could evolve with new projects.
They adopted Databrewery to streamline text annotation. The platform allowed them to easily select and guide labelers, set clear annotation instructions, and evaluate both label quality and labeler performance all within a unified environment.
With Databrewery in place, the team now has a reliable system for collecting and iterating on training data at scale. It serves as a central solution for their data science and ML teams, allowing them to meet tight deadlines and launch trust prediction models faster and more effectively.
For over two decades, a leading global consultancy has been transforming complex public sentiment into actionable insights to detect early signals of cultural shifts. Their advanced analytics arm is now pioneering a new frontier: using machine learning to measure and predict consumer trust in brands, a concept long studied qualitatively, but difficult to model at scale using data.
With no off-the-shelf models to measure trust, their data science team set out to build proprietary algorithms trained on vast volumes of unstructured text including earned media, social posts, and research sources. A core challenge was creating a scalable, representative, and bias-resistant dataset that could feed their trust models. They needed a way to enrich large, diverse datasets covering various industries and issues without introducing overfitting or noise.
To build this training data, the team adopted Databrewery as their annotation platform. With its intuitive interface, programmatic flexibility, and robust QA features, they were able to quickly onboard labelers, guide annotations, and monitor quality all while integrating seamlessly with their existing AWS data lake using the Databrewery Python SDK. Metadata tagging and project creation were automated to keep workflows agile as new data came in.
Crucially, internal domain experts were embedded into the review loop to apply institutional knowledge to edge cases and ambiguous labels questions like “What indicates a breach of trust?” or “How can you detect a reputational crisis before it explodes?”
"Trust is subtle and contextual,” explained Maya Tenzing, who leads applied machine learning on the team. “You can’t rely on surface-level sentiment, you need expert insight baked into your labels. Databrewery helped us bridge the gap between raw data and human judgment, so we’re not just feeding text into models blindly. We’re injecting experience into every training cycle.”
By combining subject matter expertise with modern ML workflows including weak supervision and intelligent sampling the team is now producing high-quality labels at scale with less manual effort. They’ve successfully trained multiple production-grade models on tight deadlines and continue to evolve their trust prediction engine. Databrewery remains their single source of truth for annotation, empowering fast iterations as new projects emerge.