Whose CV is most likely to be picked out from a pile of job applications by recruitment professionals? After analysing a staggering 441,769 of them PhD researcher Colin Lee from Rotterdam School of Management, Erasmus University (RSM) has written an algorithm that uses big data to predict who will be invited for an interview ‒ with an accuracy of up to around 80 per cent.
What’s unique about Lee’s research is that all of the 441,769 applications have been judged by real recruiters. This allowed Lee to study what recruiters actually valued in a CV, versus what they said they valued. Without even judging the applicant’s accompanying cover letter that recruiters usually get to see, Lee’s algorithm can now predict who gets invited to job interviews with an accuracy of 69 per cent. If the vacancy requires no cover letter and just the CV is analysed, the algorithm success rate goes up to 81 per cent.
To be able to process so many applications, Lee used software that automatically scans digital CVs for a wide range of attributes, including experience, age, distance from the workplace and education. Contextual factors were also taken into consideration, such as ‘did the candidate apply in time’ and ‘was the candidate already employed by the company’?
Lee designed a very detailed model of the job market that described every occupation in terms of the most common work activities performed in that occupation. This involved cataloguing equivalent occupations across industries. The tasks of a business analyst in the construction industry will differ from those of a business analyst in healthcare, for example. Then he matched up the characteristics of the applicants receiving interview invitations to the the occupations the applicants applied for. This allowed him to predict what kind of people are most likely to be invited for a job interview in various occupations across sectors.
Lee's results show that recruiters are, as expected, concerned with how many years of work experience the candidate has. Unexpectedly, recruiters care very little about whether or not the applicant's skills and education are closely related to the job in question.
Lee predicts his model will be useful for screening large numbers of CVs. It can help recruiters distinguish between applicants that should be invited for an interview and those borderline applicants that need more careful consideration, he says.
Predicting candidates for new roles
Lee’s model can also predict which candidates are suitable for newly created occupations. This would be especially useful in innovation-rich sectors such as technology or IT. Business models change rapidly in those sectors, and new occupations frequently emerge. With his model it is only necessary to establish what the main work behaviours are expected to be, and from there, it can predict who would be the most likely candidate.
Adding performance data
This use of big data to model the job market will become even more valuable once former applicants’ job performance is added to the database. This would make it possible to predict a candidate’s future performance simply by scanning their uploaded CV, Lee concludes. He defended his PhD thesis on 11 march 2016.