Episode 24

Ariel Hanemann on Hiring Data Scientists in a Cookieless Environment & in the AI Era

TIM: Ariel Great to see you! It'd be great to start by understanding a bit more about IntentIQ, learning more about the company and the product.

ARIEL: Yeah, so first of all, thank you for having me here. It's very nice to be here. So Intent IQ is an identity resolution company in the EdTech business. We have existed for more than 20 years now, and we have over 170 patents; some of them are licensed to Google and Microsoft. Generally, what we do is identity resolution. We specialize in the cookieless environment, which is a harder environment. I think the big money is in this environment. Okay, it's much harder; the scale is really huge, and actually it brings me to also what we're talking about, which is recruiting data scientists, because when you have to do big scale, there are some features you need in the data scientist that sometimes maybe it's hard to assess. Okay, you can't just give them a task that includes big data. Okay, so yes, and generally we sell our clients, who are usually the publishers, Ed sellers, and Ed buyers.

TIM: And this cookieless environment is becoming more and more common now. I hear every year about the death of the cookie. I guess it hasn't quite happened yet, but it's certainly on its way.

ARIEL: First of all, it's on its way; it's currently mostly with Apple, the Apple devices, the iPhones, but the big fear of this whole industry is what will happen when it will happen also in Chrome devices in Chrome browsers, so yeah, it's becoming big. It's good that we already have the solutions efore it's deprecated. yeah

TIM: Okay, that sounds fascinating, and you guys are ahead of the game, clearly in a good position to capitalize on the death of the cookie. The death of the cookie is your win, basically.

ARIEL: Yes, not nice to say, but yes.

TIM: Okay, and so you touched on already there some of the challenges in hiring data professionals into your team, so one is just the sheer scale of the data you deal with is atypical by the sounds of it, and at least among non-tech companies for sure, so it'd be good to hear more about why the scale matters. And then what are the other challenges that you've had in the past in hiring data scientists?

ARIEL: First of all, the skill matters because it's another, let's say, quality that the data scientist should be able to do. Let's say I divide it into two parts: you have the parts of the creativity, the algorithm development, understanding features, and everything, and then you have the—you can say—data engineering part. and the data engineering part is you can be good at one but less in the other, so when you're looking for both, it makes it a little harder. Second of all, when I want to test them when I have candidates and I want to test the candidates, you know, for how well they will be fit for our company, so it's easier to test. You know, with small tasks, it does test creativity and understanding, but you can't really give it as a task, so usually you need to just ask questions that will say how much experience the candidate has in big data. Another option is to just take data engineers and data scientists, but I find it harder to work with. Okay, because then you need to explain yourself twice every time, so I really prefer to hire people that know both, which makes it a little harder.

TIM: Yeah, and I would have thought that at a certain scale, those roles would have to specialize. If your company was, I don't know, 10X the current size of it, do you feel like then you'd have the dedicated data engineers and the data scientists would be separate?

ARIEL: Yeah, first of all, we started doing it lately; we are now also hiring a data engineer, the first data engineer, because of the scale. I still think that you cannot have just a data scientist that doesn't know anything about scale, although the data engineering will do most of the job because sometimes you can't do things. The way you want when you know that scale will be involved, okay and you have to change it accordingly. You should know what you're doing, okay?

TIM: So, be like, I don't know, a database manager working under the understanding of how the data is actually used, so making sure the database is optimized for the website that they understand its user base, not the fact that they're a front-end engineer themselves, they're a designer not doing it, but they have to have some contextual understanding; otherwise, it's just not going to work.

ARIEL: I came from algorithm development, and in the beginning I wasn't aware that you can't just do anything you want; some things the scale actually limits you, so it's important to understand it.

TIM: And is it the case, then, that your team currently has this kind of profile of a data scientist or data engineer? You would take someone who's relatively stronger on one or the other side, and then you'd be upskilling them where they're a little bit weaker. Is that your approach?

ARIEL: Usually yes, and usually because I come from algorithms, I will usually take the people with the soft skills first of all the creativity and the things that I can assess. It's really harder to assess candidates about their big data skills; sometimes I also ask for someone from the databases to help me. in this, but generally my exams that I give the candidates are usually based on algorithms and data science and feature extractions, but maybe special features, or let's say intelligent features, and I will not rule out candidates who have less experience in the big data issue, but I do think it's important.

TIM: So other than finding candidates who maybe have a bit of a crossover of these two areas and then dealing with the high scale or large scale of data, are there any other obvious big challenges that spring to mind for hiring data talent at the moment?

ARIEL: I think one of the hardest challenges was that we made a couple of iterations until we made it good. I'm in this company for four years, but before I was in another company, and in the beginning, so it's really hard to assess, first of all, how they will work with you. Okay, I will be the day-to-day work. So in the beginning we used to give an exam that you take, and we will send you an exam to email, and you will work on it alone for three to four days, and then you will explain what you did, and this resulted in sometimes situations that the assessments were really off what it was supposed to be. You can find a candidate that will look really good in this kind of assessment, okay, but when he comes, you will see that he knows maybe 20 percent of what he showed in this test. I don't know exactly why; maybe when you give him all the time he wants, he can also consult with his friends. and this is a big issue because people need work, and I don't blame them for trying, so we stopped this, and then we passed to some kind of a shorter exam, but the candidates would come to our offices and do it, so in this case, you know that they are doing it alone, and they did it, and sometimes they did it well. and so this was better, but still it lacked one important thing: to see how we work together. I think that specifically in my team, I work very closely with each data scientist. We sit together every day, and brainstorming is a big part of the work. Okay, so when you do a lot of brainstorming, you need to have this ping pong to feel that the candidates or the employee understand you and you understand him. and there will not be some nodding because no one really understands each other, so we changed this to a task with I, me, and him working together. This is a little help because I already know the solution, but you need to actually mimic a real workday. This is, I think, the best way for me to assess a candidate. You mimic what you will actually do in work, not first of all in the specific task. The task can be similar to what you do, okay, but try the best you can to actually mimic what you do in the daily routine, and when we started this, we had really great employees currently.

TIM: Yeah, I think that's the trick, isn't it? Making sure the hiring process is as close as practically possible to the real work, because it's very hard to do really well at something that's 99 percent of the same job as you're about to get and then not be good at the same job. Surely it's going to be a better predictor. I've won the job performance I guess most people would say the devil's advocate to that is the level of effort that you would need to go through because you're in there in that final meeting; you're actually engaging with them; they're not doing the take-home test, and they're in time and just presenting; you're actually sparring with them. So it must be a challenge to make sure by the time you've got people at the bottom of the funnel that at least the odds of them being good are very high, so you didn't end up doing a lot of interviews. Have you experienced that or iterated through that?

ARIEL: Of course, this is a great point, and exactly what you say is hard. Sometimes there are times that you have two candidates a day, so you can almost not have time to work, so it's really important that you know your focus and you take only the ones that are good. and this is when solutions like yours can help First of all, I think in the screening for listening, the screening part is very important. You don't want to lose anybody that can be good, but you need to screen for CVs, and after you screen for CVs, you need to understand that the CV actually reflects what the candidate knows. So a lot of this kind of in the beginning we wrote a short test for our—we didn't have HR yet, so we had a secretary that will give them the exam, a really short exam, and then we'll read it and see if they can continue, so yeah, this is, I think, very important, and the best way will be to have some kind of digital screening or something that will work. and then you are left with a few good candidates, and with them you can do this assessment.

TIM: Yeah, and it would be interesting to see where these assessments go because, yeah, from our perspective, we've always been in this interesting space where we've got a product that's quite structured, which has some upside but some downside as well because it's not going to give you the same depth of insight as a customized assessment based on your data where you have that, as you said, that ping pong sparring with them. You're going to learn a lot more; you're going to get a lot more breadth and depth of understanding. But I guess it's about having almost like the right tool in the right position in the funnel. I wonder whether we'll get to a point with large language models where they could interpret this unstructured input and output a lot better, and you could almost have something that is almost as customized and in-depth as your assessment but still maybe with the benefits of some kind of automation.

ARIEL: Yeah, I'm sure that this is currently still not the case, but as things are going with the LLMs, we will all be replaced, so I'm sure that eventually this will be a very good way to assess, and yeah, but I think in the next few years we are somewhere in the middle of this automation. This is up to you guys to find what is the best way. It's currently, I think, the screening process is very important because it can take more than a month to find somebody, and every day it's really hard.

TIM: And one thing We're hearing a lot about speaking of screening; a lot of companies are getting inundated with many CVs and many applications where they feel like a lot of them have been written or optimized using ChatGPT or some kind of large language model, and then maybe in some cases the candidates are also applying with some kind of automation tool to many different roles. Is that something that you've noticed as well in your market?

ARIEL: is relatively new, but I think ChatGPT doesn't really change much of the original problem, which is that the CVs are always, you can say, exaggerated. Maybe I would say that with ChatGPT, everything is easier. They don't even have to be good at writing CVs; this is another thing that the machine can do for them. but I think that usually CVs can be very misleading. This is a big problem in the beginning when we interviewed; everything looked really good in the CV; it was really okay, so I want this guy and this guy, and then you understand that this is really not the case. When I look at a CV, I try to look only at the bottom line. If it's if you have a degree in computer science, then you have a degree in computer science, or if you work with a specific model, XGBoost or Deep Learning, then I know I will assume that you are telling the truth, but when you go down to the high resolution of the CV, usually it's not true. So this is a very big problem, and another thing is the automation. When people can just, by the way, this is always, it's also the other way around; sometimes the job description can sound better than it is, okay, because the companies also want to attract good candidates. but I think the automation process where people can just send their CVs is also a big problem because when you have 20 CVs, it's one thing, but when you have 200, then you It's not something that a person can do on its own; you must do it some way automatically, and you don't want to. You know you want to choose. Okay, you don't want the ones that wrote better CVs with ChatGPT to be the ones that you examine.

TIM: Yeah, it's going to be really fascinating to see where this is going. I feel like the problem's going to get worse before it gets better. Personally, I'm not trying to be cynical, but that seems to be where it's going, and one other layer that companies have told me recently is that there's a lot more applications. They seem to be more likely to be written with Chachi Petit than not, which may be in itself okay, but the other related problems are Oh, all these CVs look amazing. They all seem to match the job description very well. If I have a hundred CVs that all look equally good, which one do I interview? So that's the same problem. And then to your point before around CVS, they have always been a dreadful indicator of who is worth interviewing. There's a sense I haven't got any measurement for this, but there's a sense that maybe that's even worse now. I wonder if because as a candidate I have to lie on my CV, that's me writing that lie. For most people, that's going to be difficult to do. You might exaggerate; you might not lie completely, but now it's almost like you can outsource the morality to the large language model and go, I didn't lie; it hallucinated. It's almost like a get-out.

ARIEL: Yeah, first of all, one solution is to bring morality to the LLMs. One day, maybe, yes, I agree, and also another thing is that it's really hard to make your CV fit the specific job. On many jobs, when you go, when you send it yourself to three jobs, maybe, but if you're going to do a hundred jobs, then everything is automatic. You don't even have time to read what the ChatGPT wrote for you; you just imagine that people sit in the interview and are surprised that from their own CV yeah

TIM: I have a master's degree in statistics. Oh wow, when did I do that? That was amazing. I'm very proud of myself. I did it in my sleep. I'm interested in actually flipping it around because of what we're seeing happening now. Imagine you yourself were going for a job in a few months. Would your strategy differ? Would you use this sort of almost spray-and-pray approach to applying to job boards, or would you use a different approach? How would you attack it?

ARIEL: Specifically in our field, I think that I would maybe use ChatGPT by fitting the CV to the specific job for every job. I don't think that I will do it in a machinery manner that I will send to 200 places; I would probably start with 10 that I can actually know what we're talking about. I think maybe I'm old school. I don't know if this is still the case, and I think also that if this is the situation and we are going to use more and more solutions like yours because of what we say about the CV, that you cannot screen with, so you must screen in another way; you must add insert automations. So I think that one of the effects would be that the candidates will be willing to participate in this because sometimes candidates say, Okay, so I have 20 interviews. Okay, it's a lot of jobs; it's a lot of work. I need to do this and this. And I think if it becomes a standard, then when something becomes a standard, then everybody just does it that nobody will complain that if it's hard or if it's time-consuming, because what other way do we have in an automatic world? Everything, the scale increases.

TIM: Yeah, it's just going to be fascinating to see how it plays out. Do you feel like there are any bits of the hiring process that can't be replaced or are not going to be replaced by technology or AI anytime soon?

ARIEL: I think if the work environment and the jobs themselves, let's assume that they will stay the same, then I think it will be very hard to replace the ping pong part with the brainstorming part when you really assess this is something that, because we are not machines and because you need to feel that when you brainstorm, you understand each other. As I said before, I think this part is the hardest one to replace. Maybe it can be helped with automation somehow. For sure, the stages before maybe we can make them shorter and more automatic. This is really the hardest part.

TIM: I wonder if there'll be, like, an Ariel interview bot in a year or so, a deepfake version of you that has that sparring. It's trained to you and the types of questions you would ask and how you engage with a candidate. Maybe it's 90 percent close enough that it's almost good.

ARIEL: And then it will also decide if it is to hire or not, but I think when we reach this stage, then why would it even need me for anything if it can do this? I'm sure it can do anything else I can do. This is a more philosophical argument for the future.

TIM: It is indeed. What about actually just thinking about candidates using these tools in the hiring process? What is your actual view on that? Do you view it as cheating, or do you think they have to use them? Why would I not want them to use these groundbreaking technologies in their job? or is it somewhere in between? How do you think about candidates using these tools?

ARIEL: You could always use Google before CGPT; also, when you gave an assignment to a candidate, you could always use Google. I would never send him an assignment and say, Listen, do it, but don't ask Google anything, not nothing, just work without your computer, because in the day-to-day job, he will and also if it is a code coding assignment, I have no problem that it will use Chat GPT because this is how it's done now; this is how it's done today. The only thing is that it makes it harder to write the assignment because you need to write an assignment that the candidate will show with the existence of the Chat GPT; still, you need to see that the candidate is good enough, which is now harder. especially by the way for coding because maybe data science is a little complicated, easier, so yeah, I think it's okay that they use it for this part for the CV. Also, anything that I would do, I think it's okay. What I didn't like before is that it looks like many times they used a human friend to help them, and then you're actually interviewing the friend, but the friend is not the one who is coming to work for you.

TIM: Yeah, and I feel like that's clearly worse because once you get in the job, presumably you still have access to the AI, whereas your friend isn't sitting next to you 24 hours a day, so

ARIEL: I'm sorry, I'm saying if he did it, he will probably also want the salary.

TIM: Those pesky humans—they want to get paid, don't they?

ARIEL: Yes, that's what's good.

TIM: One thing we've noticed in the hiring market at the moment is there's almost a sense of increasing distrust, I would say, on both sides. Maybe this is mirrored in general society; maybe it's not just related to hiring, but lots of candidates would say Oh yeah, these job ads are bullshit, or there are lots of fake job ads. Companies are just trying to boost themselves, and companies on the other side are saying, Hang on, the CV is nonsense. 10 percent of what 20 percent of this is a lie is this as big a problem as it seems, or have we overhyped this already? What do you think?

ARIEL: think that it is a problem, and I think it is increasingly Being increasingly worse if you're talking about high level, generally what's going on with the culture today and the networks and everything, but specifically when it happens in the social networks and everything becomes distrustful, and everything is made by machines, so you can't know who really made it, so first of all, yes, I'm sure I think I see it; it really was these days, but the two things will maybe have to improve it. First of all, standardization, I think it's maybe easier said than done. If there will be some kind of standardization, you can just say anything you want, or there will be some kind of a baseline. Okay, so I did this and this in the CV and also in the walk. This will be very good, but for this we need a company like yours to be so big that everybody will just straighten up itself according to the new, let's say, morals or And the other thing maybe will be some kind of a peer review. Today the reviews for anything for the taxi driver and for the restaurant and everything, so this is something that unless we can fake it, this will be something that will help it because let's say a big company, they will have a lot of reviewers, people that work there, people that interview there. So it's harder to lie or to exaggerate, and this could be some kind of possible solution currently without this solution. I think you're right. I think people feel that the disinformation generally that is currently in the world is also entering our hiring process, and it's hard; it's even hard mentally.

TIM: It's hard, and I think it's especially hard because models like ChatGPT are so good at making something that seems a first pass good, especially if you don't have any knowledge in that area yourself. So I could ask it to write me a—I don't know—a legal document. I'm not a lawyer, and I'd look at it and go, Wow, that probably looks as good as a real legal document to me. Maybe a lawyer can scrutinize it and quickly go, Hang on, this is wrong; this is ridiculous; this is wrong, so I feel like that's its danger: it could be the world's greatest spam engine.

ARIEL: The question is who will read all this spam. We also need the Contra the Chachapiti that will read all the spam and the emails for us, and also we are, let's say, we work in data science, but we are—this is our expertise, but reading CVs is not exactly my expertise. It's more maybe of an HR expertise, so I can also be fooled by these tools, and yes, I think it's like if you look at deep fakes maybe 40 years ago, if you saw a picture, then you were certain it's true, then you add Photoshop, but now a pixel doesn't mean anything, so in the same manner, CV will not mean anything. It will only make it harder; the level will be just higher and higher, and all the CVs will look as perfect as all the others, so it's not even worth sending a CV.

TIM: Yeah, funny you mentioned that because I started to then think, okay, if I were a candidate, what would I be doing now? I probably would stop thinking about an inbound funnel of applying through a job ad. I'd be thinking about, Oh, who do I know? I'd go back to trying to get my going through the back door, basically trying to leverage my networks that I've built up. but if you're a junior candidate or you come from a disadvantaged background or a lesser university or you haven't been to university and you're starting from zero, then you have to basically fight it out with everyone else.

ARIEL: Yes, which will be for sure really hard just to get an interview. It depends on the market, but yes, that's why we need this kind of baseline. You can be first in your class, and then you compare it to everybody knowing that you were first in your class, so you have this kind of comparison. It's really hard to let the employees, the employers, see you in the sea of CVs that everything looks the same and not real.

TIM: Yeah, I think you're right. I think we would need some new data set; yeah, maybe it's the peer reviews. Maybe it's that people are going to have to be publicly sharing their work more, like 99 point whatever percent of what we do would end up in Confluence documents, Jira tickets, emails, and blah blah blah blah. I imagine a co-pilot kind of tool that's sitting across all of these must have a very good understanding of what you've actually done in the past five years. If it sits in that ecosystem, maybe it could be the one creating this real CV about you, and it's actually getting your real accomplishments, which, to be honest, would be hard for anyone to remember anyway unless they were going through and documenting that as they go. Maybe it's just we need new data or new data brought to the fore or something.

ARIEL: I think this is a very good solution. It will also be a standard, and everybody will use it, and you don't even have to think about using it because it's always in the background, and it will assess you. The problem is what happens when you're on the lower part of the candidates, but I guess that's always a problem.

TIM: Another angle, actually, which we haven't spoken about, is I feel like the way hiring is traditionally being done has not been the fairest. There are clearly biases that already exist. I'll give you one interesting example just in Australia, and I'm sure there are other similar experiments in other countries. So some researchers at the University of Sydney compiled tens of thousands of CVs and split them into three groups. The groups of CVs were similar to each other except their names were different, so in one group it had Anglo-Saxon first and last names, and the second group had an Anglo-Saxon first name. Chinese last name The third group had Chinese first and surnames; they then applied at scale to thousands of jobs in Sydney and Melbourne and measured the rate at which those CVS got a callback, and to cut a long story short, the first group had a 12 percent callback rate, and the third group had a 4 percent callback rate, so basically if the only difference between you and anyone else in Australia is you have a Chinese name, you have only one third the chance of getting a callback, which is ridiculously unfair. If I had Chinese children, I would be very annoyed with that fact, and so I feel like there are a lot of systemic issues. Could AI improve those? Could AI even exacerbate those? What do you think?

ARIEL: First of all, it's surprising. It's really surprising. I think that if you want to solve this, first of all, AI can be trained not to look at these things. The question is why the employers look at these things. Do they actually think, or maybe it's something in the back of their minds, or they actually think that they are better off with somebody with I'm going to say I was Anglo-Saxon name, and even if you start applying without names, okay, then if they really think that they don't want Chinese employees, it will not be okay, so you will get the interview, but you will not get the job. First of all, for sure it's not fair, and it's something that needs to be changed. I think the fact that you bring up AI can really help, and maybe the CEOs and the management should somehow enforce this use of AI to stop this. Yes, it's surprising, and I think maybe it's in a high-level view; you are talking about names and nationalities, but in a low-level view, it can be something that doesn't really have to be, Are you Chinese or not? It's just how I like these kinds of people, or these kinds of people—people that are more like me in some way that I don't even know how to say, but I feel it. This is also something that AI can solve. I don't know if it's a I don't know if you want AI to solve it because sometimes you actually work better with somebody that you want to work with.

TIM: It is, and I feel like, in theory, the AI model could be a dramatic improvement to this process because, as you say, it could only, I think, be part of the actual challenge with the CV, which is that so much of it is noise. I don't need to know at that stage their ethnicity, their religion, what they look like, where they went to school, or how old they are; like, that is all irrelevant. It's shoved in front of my face, so even if I try my best, I'm still human. I can still see these things even if I'm unaware of them, so at least an AI should not have that emotion, and/or it could be fed only the data, like maybe there's a two-step process: it extracts the information, removes the noise, then makes a decision. Like, surely that could be a big improvement, I feel.

ARIEL: Yes, and I think, yeah, if there is a possibility that you will give the name and it will learn the exact thing we don't want it real, also it'll be nice if the CVS will just not include this. You will have some kind of candidate ID, and nobody will know it. So maybe it'll make it easier. Eventually, you will reach the final interview, and also there is also a question if so eventually when you interview, your goal is to find the best employee that will make the most value for the company, and this not only means how it works, but it also means how it works with other people. and how the entire team will work, so then you start thinking maybe I do want somebody that will fit the team more socially. Okay, I can't say what exactly it means. Okay, let's say people that you like to work with, okay, maybe you will prefer somebody with less experience, but you feel that is nicer and easy to work with. It's not even I think this is also even okay.

TIM: Yeah, I feel like that's fine as well. I think that companies, when they hire, should just make that like an explicit criteria, so let's say you've got these technical skills you want them to have; it may be a bit more measurable. I would have no problem with a company then saying, You know what? 30 percent of the weighting is going to be likeability, and everyone who interviews them is just going to rate them on a scale of one to 10 on how much they like them. That's fine to me if you say you hate them or you love them. But if that matters to getting along with people and having a good team culture, then fine, but I feel like a lot of the time it's like an unconscious thing in the back of their head rather than explicitly saying on paper this is how much I like them.

ARIEL: Yes, and I wonder how, because if somebody really doesn't want to hire Chinese employees, then maybe when it comes to this, when they ask him if they liked him, it would like him less, so the question is how is it actually preventable if we can actually completely prevent this unfairness?

TIM: Yeah, I feel like it's possible to chip away at it and not get rid of it completely. You're right. If someone ultimately is either explicitly or implicitly against Chinese people, eventually they're going to rate them lowly, at least in that scenario, though that would be really interesting to get to the end of a process. and have five interviewers rate a candidate on likability; four of them give them an eight, and one of them gives them a two because then that person who's given them a two would be like Okay, why have I given them a two? What's behind that? What was it like? If you dig deep enough in the psychological reasoning, hopefully you'd almost uncover the bias in a way I imagine.

ARIEL: Yeah, it's interesting that we're data scientists and we work with algorithms and mathematics, but so much psychology and human behavior is inside this field, and maybe it's more of an HR field, but you can't let HR do everything because eventually you're the one that is working with the employee, and now what we're talking about is that when you hire a manager, then one of the skills that he needs is some kind of hiring ability. So you need to take this also into account.

TIM: Yeah, you're right. That's an underappreciated skill unto itself: the ability to hire and recruit. That probably, for a lot of data leaders, they would never get taught; they just learn it themselves and figure it out as they go along.

ARIEL: I think that in my case, for sure, I started; I didn't know anything about hiring, and I made a lot of mistakes, and with time it improved, and we see, and we changed, and yeah, you live and learn.

TIM: Speaking of HR and talent acquisition, they're an interesting bit of this puzzle. My view is that often talent and HR teams would struggle to do the screening for technical roles because they themselves aren't data scientists; they aren't software engineers, and those roles are a very long way away from something that they know. So, I would say in some sense we've given them an impossible job. I feel what's your view of how HR and talent have done screening for technical roles? What do you view as the way they approach the problem?

ARIEL: I think you're correct; it's a problem. Usually, maybe if they work in the same workplace for 20 years, then they will start understanding specifically what is going on with this field, but generally, yes, they come from a more psychological background and social background, and really, this is not their field. and you see it all the time; they don't exactly understand what you want, and even if you put it in words, okay, so you put it in words, but people can write similar words that are not exactly it, and it's really hard, and I think their part is mostly to see that they fit the organization and that they are okay and handle the process. The problem is that if you have a big company and you want somebody like HR to help you with the process, I agree that they can't give it to you, so what will you do? Will you have some kind of new position that is a technical HR? Because, for example, I'm technical; I don't want to work in HR. Usually the technical people want to work in the technical field. Yeah, I think this is also a place where your solution might

TIM: Yeah, it's interesting. I don't envy talented people because they, in my experience, are sometimes the most overworked, underappreciated people in a company who've been given almost an impossible task to run the hiring process for a wide variety of roles, none of which they've ever done themselves. Yes, impossible, frankly.

ARIEL: Yeah, I totally agree, and it's not fair that we judge them that we don't know what specifically we know in our field because we work in this field, and then they will go to another job with a totally different thing, and then it will be the same all over again. Yeah, it's true that something should be somehow separated between the hiring, the more social, sociological part, or

TIM: And I think that the fact that they have the soft skills, the company culture, and maybe the more psychological background, I feel like that maybe then flavors how they view candidates because they'll always view them through the lens of how do they interact with me? Do I feel like they're a good cultural fit? I feel like they personally maybe overweight that because they're not in a position to judge the technical skills. What do you think?

ARIEL: Yes, I agree there should be some kind of scoring system where they can only talk about this part, the part of the interactions and, you know, the psychological behavior and everything, because really, in the part in the field that they don't know about, they didn't study, you know, that this employee went to the university and studied mathematics, and they didn't I don't think that should even be expected to do This part should be divided.

TIM: And is it a challenge then that people in data roles, or let's say any really technical field, often very smart people, have maybe less good social skills, in inverted commas, that then must mean that lots of, like, slightly awkward geniuses must struggle to get past those initial stages?

ARIEL: Yes, I think that the HR eventually gets used to understanding that this is the level they expect you to be at, and you know that it's—I agree that in the beginning they will say, Okay, but maybe it's not enough in the behavior, and then they understand that everybody is like this in the field. So yes, but also if everybody is like this, it's probably harder to evaluate because these people are harder to evaluate in this manner.

TIM: Yes, it must be so challenging. One final question for you is what about candidates? So, candidates listening to this who we've talked about a lot of doom and gloom of the challenge of getting jobs in this AI era, but what I think would be helpful for them to know is what's going to improve their chances of success. If you think back to the candidates who you've hired and those who you haven't hired, are there certain common factors, like typical reasons why candidates would fail to ultimately get an offer?

ARIEL: Yes, first of all, unfortunately, I feel it's something that it's hard for the candidates to change, specifically in my case, because they don't know exactly what the assessment will be, and I want to see that they understand me and I understand them, so it's something that is hard to teach, but for me, I would say maybe it's naive to be as real as you can. Don't exaggerate, and just do your best and be real. I think for me it helps. It's hard to not show your real self in the interview, so just be yourself.

TIM: That's a great bit of advice that applies to recruitment, probably applies to dating as well, and anything like that, that authenticity I think anyone who can see it would always value that, I think.