How a moderation pilot on your data works
A pilot is a check on a sample of your content before you roll anything out permanently. You get a concrete result on your own data, not promises.
- Usually 2–6 weeks for the API and 4–10 weeks for a custom model.
- A content sample is enough, for example 1,000–10,000 listings.
- At the end a report and a decision: we roll it out, or we stop at the report.
What you bring
A sample of your content, for example 1,000–10,000 listings or reviews. If you have historical moderator decisions, they help us check agreement and tune the model. If you do not, we start with the API and threshold setting.
Four steps
1. We set the rules: what is allowed, what is not and what goes to a human.
2. We run the system on a sample of your content.
3. We review borderline cases and errors together.
4. We measure the result and prepare a report.
What we measure
The share of content that lands in ALLOW, REVIEW and BLOCK. How decisions compare with your moderators on the same sample. How many false alarms (honest content blocked or sent to review) and how many missed violations there are. How much work is left for your team in the REVIEW queue.
We do not quote accuracy or savings figures in advance. They depend on your data and rules, which is why we run a pilot.
Data and privacy
Where we process data during the pilot is agreed at the start. We use a public cloud only if you agree to it. After go-live the system can run in your environment. We agree on the pilot scope and terms with you before it starts.