Two factors affect the accuracy of the result: the quality of the analysed source, and the accuracy of our AI pipeline.

Our chrome extension allows users to analyse any website, but we only publish results from good sources on our public HowDiverse.is website. A good source is a page that contains photos reflective of the analysed organisation, such as their About us or Careers page, Leadership page, Linkedin company profile, cast pages for films and series etc... Our pipeline can also identify and analyse photos containing hundreds of faces such as a photo from a company event. These all are examples of sources that reflect the company's diversity (or lack of) accurately.

Assuming the source is accurate and reflective of the subject being analysed, then the second factor is our AI pipeline itself. It is not 100% accurate, but its accuracy (~97% for gender, ~75% for ethnicity) is better than most humans. So we are happy with it, especially that we also manually revise the results that have a low confidence score to improve the model and the results in the future. So it will only get better.

So we believe that we provide a very good indication of how diverse organisation are. In fact, in the absence and reluctance of many big organisations to publish such data, our crowd-sourced analysis is often the only data available, which gives the public a baseline of transparency to hold organisations accountabile regarding their DEI goals.

Last, this is only the start. We encourage organisations to "claim" their profiles, and provide their own diversity data. We are also building tools to allow organisations to gather this data by allowing their members to self-identify beyond binary gender and ethnicity. We believe that these tools - not the use of AI alone - are the future of this platform.