No sooner had we sent out the CVAT test job to candidates, than we got a huge wave of feedback that CVAT open access is not working for many due to technical reasons. There were, of course, some candidates who were able to complete the task or were more creative and completed it in another program with similar functionality! When inviting candidates for interviews, recruiters of course took this into account, but since it was not possible to create the same conditions for everyone, we refused to evaluate the job using CVAT.
Resumes were screened to determine the priority of interview invitations. HR colleagues will wonder: «Screening in the penultimate step, not at the very beginning!» This is due to the fact that project manager in data markup is a new profession, so the right applicants are people who may have never worked with markup before, but have great potential due to their unique background and experience in various fields. When evaluating candidates, what mattered more to us was not so much a person's background, but rather their developmental vector and career goals in data markup. And when it came to prioritizing interviews, we paid attention to the candidate's «past» and additional skills.
So, based on the screening results, the first 25 applicants were invited for an interview with HR and operations managers. A series of three days of intensive interviews ensued, focusing on the skills, motivation, growth prospects, and values of the candidates. Also, some applicants again had the «I really want to be in IT and it doesn't matter what I do» story — unfortunately for candidates, these were stop-factors. At the end of this stage 5 offers were put up, all were accepted.
Ultimately, the hiring funnel for the project manager position looked like this:
308 — responded (100%)
295 — filled out the questionnaire (95.8%).
70 — passed the questionnaire (22.7%)
25 — interviews (8.12%)
5 — offers (1.6%)
As you can see, a fairly large cut of the conversion occurred just at the stage of the survey due to the large number of irrelevant responses.