AustraliaTechnology and telecom2026 – ongoingAppen
Machine learning classification and auto-resolution of contributor support tickets
Appen cut contributor support resolution from two weeks to under 24 hours with a machine learning ticket system
Problem
Appen's contributor support received more than 11,000 tickets a month from over one million contributors. A Contributor Success team of two people could not keep up, and tickets waited about two weeks for resolution.
Approach
Provectus built a system on Amazon SageMaker that classifies each incoming ticket, assigns priority and routing tags, and resolves routine issues without an agent. It connects to Appen's existing ticketing system. Tickets the model is unsure about go to human agents. Agents' manual corrections feed a monthly retraining cycle. The first training data was labelled on Appen's own annotation platform.
Results
Takeaway
Routing low-confidence tickets to people and retraining on their corrections let a very small team keep automatic resolution safe. Next steps named were extending to 180+ languages and alerts for at-risk contributors.
Categories
- Industry
- Technology and telecom
- Business function
- Customer service
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Partner co-development
- Evidence
- Vendor case study
Sources
Source summaries
Provectus's case study describes the SageMaker-based ticket classification and auto-resolution system it built for Appen. It reports about 80% of 11,000+ monthly tickets resolved automatically, resolution under 24 hours instead of two weeks, and a 10% rise in satisfaction. The company did not say how satisfaction was measured.
Full details are in the original sources.
