Customer Impact

Between May 14, 2026 at 4:17 pm UTC and May 15, 2026 at 12:52 am UTC, customers using Document Understanding in the U.S. region experienced intermittent failures and delays when processing documents.

The impact was limited to one Document Understanding extraction flow in the U.S. region. Document Understanding services in other geographies were not affected.

Root Cause

The incident was caused by an autoscaling issue in an internal component responsible for making extraction models available to processing nodes.

During a traffic increase, this component did not scale out. As a result, some processing nodes spent too much time loading or retrying model loads instead of processing extraction requests. This reduced availability for customer traffic and caused intermittent extraction failures and delays.

Detection

The issue was detected by automated monitoring on May 14, 2026 at approximately 4:17 pm UTC. The incident response process started shortly afterward, and the public status page was updated after the customer impact was confirmed.

Response

Engineering teams confirmed that the impact was isolated to Document Understanding extraction traffic in the U.S. region.

The team mitigated the issue by increasing available processing capacity where possible, prioritizing customer extraction workloads, and manually scaling the internal model-loading component. After this mitigation, model loading performance improved and the extraction errors rate returned to zero.

Recovery was verified through service health dashboards and error-rate monitoring. The incident was marked resolved on May 15, 2026 at 12:52 am UTC after the service remained stable with no ongoing customer impact.

Follow-Up

To reduce the likelihood and impact of similar incidents, UiPath is taking the following actions:

• Improve autoscaling algorithm for the model-loading component so it responds more effectively to sudden increases in demand.
• Add monitoring and alerts for model load delays, model load timeouts, and processing nodes delayed by model loading.
• Review Document Understanding capacity buffers and scaling assumptions across regions.