Case Study: UiPath boosts dataset growth 10x and reaches near-99% model accuracy with Encord

UiPath 10x'd its dataset and pushed model accuracy to near 99% with Cord

UiPath, a global software company specializing in AI and automation, faced significant challenges in scaling its annotation and data management processes across diverse data types like images, video, text, and voice. Their previous annotation vendor caused bottlenecks with slow turnaround times and inconsistent workflows, making it difficult to maintain high-quality training data for complex models like their table extraction system. They needed a more flexible and transparent platform from a vendor like Encord to handle multi-reviewer workflows and improve pipeline visibility.

The solution was implementing Encord's platform, which provided native support for multiple simultaneous annotators and a structured review flow. This enabled UiPath to more than 10x the size of their training dataset and iterate efficiently on their labeling process. As a result, Encord helped UiPath achieve a more than 4x reduction in the error rate for their table extraction model, pushing its accuracy from around 96-97% to near 99% mAP. This improvement, coupled with a dataset that better mirrored production data, was noticed by customers and allowed UiPath to scale models across their entire portfolio.

UiPath