This is an interesting and entertaining video that is correct in spirit, but not always in fact. One page outputs for HR or talent managers to review makes a lot of sense. But, to suggest that having only the Big 5 personality traits and some cognitive measures 'under the hood' is all anyone can do in actual prediction is both overly simplistic, and not aligned with modern science. There are at least 4-5 big, major cognitive dimensions. If we look at the modern Hexaco framework of psychological mapping of personality, there are at least 6 major traits and 24 sub-dimensions with some element of demonstrated validity. But, there are dozens to hundreds of other traits that have been studied in isolation and in specific job instances - that may or may not fit well into these 24 sub-dimensions. There are also many relatively well-researched models of organizational preference such as OCAI, and extremely well-studied models of vocation strong interest frameworks and general values based frameworks, not to mention competing many major theories of motivations and human drives. So, to insist that all of the aspects of a person that might influence job performance can be reduced to the 'Big 5' personality dimensions plus some cognitive is not just wrong, but is also fairly reductionist.

There are significantly more features beyond those listed above that can be 'model' features.

Updates by Frank Schmidt and others looking at 100 years of research have found that, for instance, interviews (both structured and unstructured) and reference checks also have statistically significant validity - the size of which varies across the job and company.

Even if the 'Big 5' and 1 Cognitive dimension were all we could measure, how we combine them and build our predictive models will greatly influence what we find - and there have been books and thousands of academic articles on the subject of backtesting and building models.

Nearly all 'traditional' academic historical findings were based on linear models in fairly small samples. With modern data processing and machine learning techniques it becomes possible to build company and position specific models based on comparing top and bottom performers with dozens to hundreds of underlying traits. It is now also possible to do audio and facial emotion scoring and Natural language processing based on text analytics using approaches like topic modeling, sentiment analysis and bag-of-words approaches. All of these are methodologically sound and rigorous approaches - where the effectiveness can be checked and measured.

So, suggesting that 'output' for users should be very simple and clear is 100% correct, but suggesting it must be simple and clear because all we can ever use as valid features are a small number of inputs combined with simple linear type models is fairly 'old school' and not likely to be supported by real data over the next decade. Nevertheless, I enjoyed the video.