There are 4 stages: HR, on-the-spot business case, take-home case, hiring manager round. At case stages, they give random problems that aren't related to the job description necessarily. You are expected to devise solutions to those and they ask you a lot of questions around it, fine. I made it to the final round with the hiring team where they ask behavioral questions, and they had a negative attitude from the start. They asked questions in a way that shows they haven't even properly read my resume, criticising my answers saying "why did you give example from this project?", although that project was very similar to what job description entailed. I've found their attitude a little disrespectful. At the end they compiled negative feedback from all rounds, and then they ghost you when you ask if you've passed the previous case rounds to reach the final round or they just did all the interviews regardless of pass/fail. They also haven't provided any positive feedback, leaving a placeholder in their email. They should treat candidates better, bring a better attitude to the interviews.

I am sharing this review after participating multiple times in the Machine Learning Scientist (MLS) interview process at Booking.com over the past three years. A little about myself, I have a PhD in AI/ML and several years of post-PhD industry experience, and I entered the process with genuine interest and respect for the company that I love it. While Booking.com attracts strong candidates and works on interesting problems, my experience suggests that the MLS interview process suffers from significant randomness and inconsistency which is largely driven by interviewer assignment and unclear application of internal rules. *** Role categorization and interview validity According to discussions with recruiters, MLS roles are divided into two categories. Candidates who pass two technical interviews are told those results remain valid for six months. However, in practice: After passing two technical interviews for one MLS role and being rejected at the final stage, I was automatically excluded from consideration for other MLS roles I applied to shortly afterward. This happened even when the roles were described as belonging to a different category. According to their rules, for a new MLS application within 6 months, in the same category I expected to directly go to the final fit round, and if different category, I shouldn't have been rejected because I failed the fit interview recently! When I raised this inconsistency and referred to the stated rules, the discussion was never resolved. This creates strong confusion and undermines trust in the process. *** Interview randomness and interviewer dependency A major concern is how strongly outcomes depend on who conducts the interview. Examples: A “fit” interview that I mentioned above unexpectedly turned into deep technical questioning focused on an unrelated, Non-ML short-term project from years earlier. No ML questions were asked. The rejection feedback later cited “insufficient ML knowledge.” In the 10-minute presentation for the second technical interview, you have to use the best of that 10 minutes and therefore you need to leave some topics for the discussion part. Interviewers sometimes focused on secondary or tangential topics, leaving no time to discuss those topics. Feedback later penalized the absence of those core points—even though the interviewer controlled the direction of the discussion. This suggests interviewers are not consistently trained on: - How to steer discussions - How to distinguish signal from noise - How to fairly evaluate within tight time constraints *** On-the-spot cases and domain expectations In one on-the-spot case interview focused on recommender systems, I explicitly stated that recommender systems were not my core expertise and that I would reason at a high level. The role itself was not a recommender systems position. Despite this, feedback cited “lack of recommender systems depth.” Expecting deep domain expertise in an unfamiliar area during an "on-the-spot case"—rather than a take-home assignment— feels misaligned with the interview format. My overall impression is the MLS interview process at Booking.com appears to have: High interviewer variance, inconsistent application of stated rules, and misalignment between interview format and evaluation criteria. In this randomness, strong candidates may be rejected not due to lack of skill, but due to randomness in interviewer focus and questioning style. Booking.com is a good company, and it is clear that they hire many excellent people. Based on repeated firsthand experience, I see signals the false-negative rate in the Machine Learning Scientist hiring process could be high. The process could benefit significantly from stronger interviewer calibration and training, clearer separation between fit, technical depth, and domain-specific interviews, more consistent enforcement of interview-validity rules across roles, and more structured guidance for managing discussions. Imo they need to somehow provide a fairer and more reliable experience for candidates. I have invested a substantial amount of time in this process over the past 3 years. While I remain interested in Booking.com, its great product, and its lovely culture, I do not currently plan to apply again in the near term unless I see clearer evidence from others' experiences that the randomness in the interview process has been addressed.

I first had a standard recruiter call then the technical one where we went through a case study discussion. We went over a typical scenario at booking. The interviewers were focusing a lot on AB testing. They were not part of the team where I would work and new nothing about it. I could not ask questions and I found it pretty strange.