05 · Thomson ReutersAI operations · Internal tools
Label Insights
Designing an internal tool that helps teams label training data and improve AI products.
Problem
Labeling work was slow and opaque, which blocked model quality and day-to-day product decisions.
Constraints
Internal tooling limits, varied document types, and the need for speed without sacrificing accuracy.
Process
Observe labeling sessions, strip redundant steps, and design a focused status-and-throughput experience.
The turning-point decision
Optimize for labeler throughput and clarity—not for feature density.
Final solution
A streamlined Label Insights workflow with clearer status, review, and completion paths.
Measured outcome
A streamlined labeling experience that reduced repetitive effort and supported higher-quality training data.
- 20%
- less labeling time per document
- AI
- model training workflow
- 1
- focused labeling experience
Interested in the decisions behind this work?
Let's talk