Case Study: Deque boosts accessibility model performance with Labelbox

Deque cuts training data spend by over 50% with Labelbox

Deque, a company specializing in digital accessibility, faced the challenge of efficiently automating its manual accessibility testing process across thousands of web and mobile datasets. Their team relied on disparate tools with no easy way to find model errors or prioritize data evaluation. They turned to Labelbox and its Model Diagnostics and Catalog services to address this.

Using Labelbox's Model Diagnostics, Deque targeted its model's weaknesses and detected noise in its datasets, rapidly filtering out one-third of less trustworthy data points. This improved model performance by over 5% and reduced its annual training data spend and needs by over 50%. Specific accuracy rates for elements like checkboxes and radio buttons saw dramatic improvements.

Deque