Sebenz.ai workers can do any task that can be done by a person on their smartphone with a few minutes of instructions. It must be easy for an independent person to verify that the task was done correctly because Sebenz.ai workers verify each other's work (we make sure there is no collusion!). Many of the tasks involve labeling training data for machine learning or fixing the output of machine learning models but we can do any kind of digital task including finding data on the web and answering research questions. We help converting tasks normally done on desktop into a format that can easily be done on a phone (usually by breaking the task up into smaller pieces). Have a look at our industry & sections to get some ideas and if your use case is not mentioned, please contact us and we will create a custom project for you...

We charge per task output which is calculated based on how long it takes to complete and how many Sebenz.ai workers need to complete the task to achieve high quality results. Before starting your project you can send us some sample data and we will give you a rate for your tasks. In general, our rates are significantly lower than other mechanical turk / data labeling services because the minimum wage in Africa is low and we reduce overall costs (like the need for offices) using modern technology and the fact that our workers already have devices. We created sebenz.ai in the first place because we were not satisfied with the pricing or quality of results from other companies in the market.

Sebenz.ai workers earn at least minimum wage and their earnings increase as they get more experience and accumulate skill which lets them do more difficult tasks and work as task verifiers. The founders of Sebenz.ai are African and we care very deeply about empowering Africans to earn a fair wage and improve their standard of living. Paying our workers fair wages while being able to offer attractive rates to our customers will always be our main objective.

When we set up your task, you give us several examples of how your task should be done. We use these as a test for each worker and calculate their skill on the task. As we allocate your tasks to workers, we show each task to several workers and use their skill to combine their answers into a single high quality result for each task using our skill-weighted consensus mechanism. Workers earn more money if they have higher skill but lose skill faster than they gain it if they do a bad job so they are economically incentivized to always do a good job. We also create verification tasks where randomly assigned workers check other workers’ tasks in a way that makes it impossible for them to cheat or collude. These testing, consensus and verification mechanisms make sure that we provide superhuman results.

We take data privacy very seriously and use several mechanisms to ensure that your data remains secure including not showing a significant portion of your data to any single worker and using digital shredding to scrub context from images and audio files. For example if you play a 5 minute audio recording of a call to a single worker to transcribe, you have to trust them to keep the content of the call secret. Better is a trustless approach where the call is shredded into clips that are a few seconds long which are transcribed by several different workers such that no single worker hears context or a significant portion of the audio being transcribed.

Response time depends on how long your task takes for a single worker to complete but we return results immediately once we have a skill-weighted consensus between the workers assigned to your task. We strive for near real-time response and if your task takes an average of 10 seconds to complete then you can expect a response in approximately 15 seconds. For tasks that do not require real-time response, we give you a small discount for batch response in which case we send you an email as soon as your set of tasks have been completed.