The problem
The platform had treated moderator attrition as an HR problem and moderation quality as a training problem, when they were the same problem: high-intensity queues without a wellbeing framework drove both. Wrongful-removal appeals were rising and classifier training data was inconsistent across Spanish variants.
What we did
- Implemented a documented wellbeing framework: mandatory rotation off high-intensity queues, scheduled decompression, on-site counselling and peer support
- Standardised classifier training data across Latin American Spanish variants
- Added AI-augmented triage so moderators focus on genuinely ambiguous cases
- Delivered from Puerto Plata with trust-and-safety leads embedded
Results
Moderator attrition fell sharply once the wellbeing framework was in place, and moderation accuracy rose as the same change reduced burnout-driven error. Wrongful-removal appeals upheld fell as classifier data and triage improved. Full metrics are in the downloadable portfolio.
Figures relate to the period stated and were provided by the client. They depend on that client's starting point, so they are not a forecast for another operation.
We had treated attrition as an HR problem and quality as a training problem. They were the same problem. The wellbeing framework fixed both at once.
What stayed in place
The wellbeing framework is now the client's trust-and-safety standard and a documented Corpshore Dominicana approach, and the Latin American classifier data set is a client-owned asset.