Episode 48

Roman Bunin on AI's Influence on Data Roles and Modern Hiring Practices

TIM: Roman Welcome to the Objective Hiring Podcast. Thank you so much for joining us.

ROMAN: Hi Tim, thank you for having me. I really appreciate that you invited me. It's going to be an interesting podcast, I believe.

TIM: I believe so, and it's absolutely our pleasure to have you on the show, and where I'd love to start, if we can, is just a bit of future gazing. It feels like AI is changing so many things about every aspect of the world right now. What I'd like to get is your thoughts on what impact you think that's going to have on data professionals. I can imagine the roles of a data analyst, a data scientist, and a data engineer are going to change. It's changing already. Where can you see those roles evolving over the next five years? What are your thoughts?

ROMAN: I think it's going to be evolved in terms of every profession is going to evolve in terms of we are going to use AI as an extra bonus, extra stuff, improving our mind and our ability to think and automate a lot of stuff for sure, but I, to be honest, don't see it as a game changer just because here, for sure, everything is going to be easier. You will, it will be easier to write the SQL or any kind of code; it will be easier to, I don't know, maybe to provide some experiments or whatever, but in the core I believe that the main structure and the main responsibilities are going to be the same, and at least for right now maybe I'm wrong. maybe in the five years I'm going to be Oh my gosh, I wasn't right so hard just because for now I don't see it like it can replace human beings, not at all. Basically, it's going to help us, but I don't see any big changes, to be honest.

TIM: Yeah, I feel like I've worked in data for a long time, and I've been almost like a slight AI skeptic in a sense. I feel like a lot of the stuff's been maybe oversold and overhyped. That said, I feel like my view on this has changed recently in the last six months or so, where I feel like we're getting to a tipping point where, oh my god, actually the changes I can now imagine happening are profound. I think a lot about the hiring use case, less so than what an individual analyst, data scientist, or data engineer would be doing, but I just feel like for the large language models in particular, they seem to be now good enough at so many different things that I see a huge opportunity for drastic changes and improvements, including solving problems we wouldn't even think of because we haven't even gotten to that level yet. and now the possibility to solve some of those things is coming to fruition, so I feel like we're going to start to see some big changes. Personally, I'm not sure about the data roles exactly. Like I've heard some people say a lot of the data engineering work maybe that would be the first set of things we can more easily automate. A lot of us just—and I'm using air quotes here for anyone who can't see this—it's just pushing data between systems. So maybe that could be automated a bit. I spoke to someone recently who was trying to actually automate the analytics bit, so they're investigating where the tooling was to add an AI layer on top of dashboards to actually generate insights, but now it seems pretty primitive at the moment from what they've uncovered. I feel like that bit would be harder than the pipeline bit to automate personally, but

ROMAN: Okay, just because recently I got the task, I was preparing for a presentation, and I decided to do this presentation in a video-like style, so I was creating a lot of videos using AI, and I was like thinking that, oh, it's going to be so easy; I'm just going to put the prompt, and it's going to be really flashy, cool videos that I'm going to use, but it happened to be that I was using the most advanced model that is in the market right now. and it was so shitty to be honest. I was feeling like I was gambling, to be honest, because I'm pressing a button, something happens, I pay money for it, and after that I receive a result, and this result could be cool or could be really, I don't know, a nightmare, and that's why I really changed in this perspective. I changed my mind just because I was thinking like, okay, this stuff is going to replace designers or video editors or whatever, but I tried it myself, and I don't know, maybe 20 slides or something like that, and I created about a thousand tries of it, so, like, it's enormous cherry-picking that went on. So that's why I was like, Okay, I'm a bit of a relaxer, just because, like, I tried myself, and for this particular purpose it was not working, and for data analytics I was also trying to do, but usually it came across as the poor quality of metadata, and if you don't have good data, it's working as well as with real data analysts; it's also quite hard to understand what's going on.

TIM: large language models in the last 18 months has maybe been I've been a bit impatient that's how I view it now in that I've been trying to get answers to problems deal with the inevitable set of hallucinations or misinterpretations or running out of memory and forgetting what I told it and all these kinds of common issues we've all encountered But I wonder whether I just needed to stick to it a bit more and know how to hack around these issues, as well as the fact that because the models are being updated so frequently, my experience from three months ago maybe is irrelevant to now, and I wonder whether, yeah, like I personally, upon reflection, felt I was being a little bit harsh and expecting an amazing 10 out of 10 and being disappointed with a six out of 10. whereas maybe I should have expected a zero out of 10 and been impressed with a six out of 10.

ROMAN: Yeah, it's like in that comedy show, then it was like Wi-Fi in the airplane, and the customer was disappointed that, okay, the speed is not so good.

TIM: Yeah, exactly. Yeah, you're flying in a plane with internet. Oh my God, come on, get a grip.

ROMAN: Maybe you're right. just because also I am expecting, yeah, I'm expecting quite a lot from it. Yeah, for sure, it's really funny how you do it. How do you think about these LLMs? Just because then they came out, I tried it, and they Oh my gosh, it's so cool. It's black magic or whatever, but like in a month, you're already okay. Yeah, just a new LLM. Okay, no matter.

TIM: Yeah, yeah, exactly. I've gone through that as well, where I'm thinking of it as so far being great use cases. One personal one I've had is in getting feedback on something in a depersonalized way, so I'll give you an example: these very podcasts. I was trying to get feedback on how to be a better host. and so I asked some people on my team to watch a couple of videos and provide me with some feedback. I watched some myself, but then also just whacked in a whole bunch of transcripts into ChatGPT and asked it for a stack-ranked list of feedback, and it was interesting that not only did it seem like pretty reasonable feedback, it also gave me action items on how to fix the problems, but also it was delivered in a way that I could consume it easily. I was defensive at all because it wasn't a human giving me the feedback; it was an AI, and so I wonder whether that will end up being one of the great use cases of these large language models in reviewing processes and giving feedback to someone in a way that they can actually take and accept and act on as well. What do you reckon?

ROMAN: Have you found this feedback useful for yourself?

TIM: Yep, I have, and so I've got a posted note underneath my monitor that says to be more concise in my questions to have smoother segues between questions and to make it a bit more personal, so that's what I'll try to do in this call, and that's based on feedback from our AI overlord.

ROMAN: Okay, okay, okay, I see, yeah, yeah, maybe it's true. I have tried something like that, but it's really fun nowadays. You can just ask ChatGPT, What do you know about me? How do you present me? and it's funny. Yeah, yeah, and also I found the cool cases in data as well just because I have to make documentation and preparation for dashboards. Just because if you upload an image for a dashboard, it can write quite good documentation. It's ridiculously better than a data analyst does, just because they don't like to do the recommendations that always are like drudgery for them, and ChatGPT is going to work cool on that, so it's quite funny. and also I started to analyze business customers requirements with ChatGPT, and also quite okay, it's going to take some time. That's why if I don't trust it really fairly, but I would say that it eases my day-to-day work not quite a lot, but yeah, a good chunk of work just analyzing screencasts or whatever they send me. Yeah, I feel that this kind of automation is good, but I don't feel it like a game changer. Yeah, it's going to go. We're going to do our work faster and smarter, but they're going to do it anyway.

TIM: I think we'll see. I wonder whether there's a segment of tasks where it is probably still worse than a human, but comparing it to human performance maybe is slightly misguided because a lot of the time the problem is the human can't do it at all; for example, imagine in hiring interview feedback writing interview notes. It's not like candidates are currently getting feedback from an interview and it's perfect. It's the problem that it's not happening at all. Having an AI-generated transcript where you automatically grade candidates based on interviews and send that feedback is better than nothing.

ROMAN: Yeah, sure, it's my one friend of mine. He's a manager; he owns a company, a data analyst company, a data consulting company, and he was hiring people. He's using a lot, and he just asked people to write something down, so it's like a screening interview, but it's made by text, and after that, he puts in ChatGPT and does it like that. Yeah, and it saves a lot of time for him, and it's fun.

TIM: For sure, and I wonder whether also once we start to see the large language models combined with at-scale programming, so, for example, using iterating through a variety of models and taking the average of all those models, so going through simulations and using it at scale as a step in a process, maybe that's when we'll start to see the massive benefits rather than an individual using it for one-off queries and one-off tasks once it's part of that whole programming engine as well, then it might be

ROMAN: Yeah, agents and all this stuff seem to be a big deal, and like, also it's claimed that GPT itself is some kind of mix of agent engines, but we're not sure. Yeah, it's true; it's true. That's why I really like—I hate that I'm working in a company that sells GPUs.

TIM: Yeah, what's that saying? Gold rush: make sure you're selling doubles. Is that right?

ROMAN: Yeah, that's our mission for nowadays. Like, our CEO is like highlighting it's a lot like, Yeah, we are showing we are selling shelves.

TIM: Roma What about in terms of hiring? So let's say we just even think about AI now in hiring. Have you seen any use cases? For example, have you noticed candidates using AI? Have you been tempted to use a bit of AI at some point in the process? What are your thoughts on the hiring application of AI? Yeah. Yeah, I've seen a couple of candidates who were trying not to cheat but used it during the interview process, so I caught it immediately, so you can see that then the answers came like general, and they are really not related to the subject a lot. It means that the person is just using GPT or something.

ROMAN: I would say that in my experience it was like I didn't like it because if a candidate would use it in a smarter way, it would be like only being like Google or whatever when he found some information and used it in the correct way, I would be okay, so I don't have a problem if you just… it's just a tool you can use, and there is no problem for me about it. But I came out with my particular case, and it was like, no, it was like, for using this tool, I would say so, yeah, but I don't understand, to be honest, why companies are still having this SQL LI Life SQL coding or whatever interviews it, because nowadays it's really not bullshit, but like I would say it's not a necessary skill. Yeah, you need to understand everything about how it works in the hood, but you don't try SQL anymore just because you can write it, but if it's simple, you can just ask ChatGPT if it's going to be better or the same quality if you do it with Stack Overflow or whatever. Yeah, talking about the process itself, I believe in our system we have some kind of built-in AI features. I haven't used them a lot, but as far as I see, there is some kind of score of the candidates beforehand, so it's analyzing CVs and LinkedIn pages, and like maybe it's cool for recruiters themselves, but for me as a high-rank manager, I nevertheless go through the full CV or full LinkedIn page. LinkedIn page, so like in my process, it doesn't it hasn't changed much, I believe, but yeah, no, to be honest, I don't have really good examples. The one I also only set for you is about my friend, but he was like using it in really like MVP style; he was using like Google Sheets with the chat GPT function. and he was just going through the answers, and it was quite easy for him to do, but I don't have any particularly good examples of how to use them. Maybe you have one because you are in the field.

TIM: Yeah, I've spoken to a few companies recently that, like your friend, have hacked out their own little skunk works, especially in a couple of areas. One is that application stage, so taking in the application question answers and the CV and doing some kind of scoring, some kind of matching against the job description and some other requirements to come up with some kind of number, basically. And then the other use case I've heard about is, yeah, in the interview itself, running a Zoom or whatever interview, getting the transcript, passing it through Claude, and using the answers to then grade the questions rather than having to do that manually. I feel like they seem no-brainers to me because The current way of doing those steps, I feel, is so flawed that it's hard to imagine how it would be worse with AI, personally.

ROMAN: Okay, yeah, I think so. Yeah, if you change it, yeah, if it's a funny thing, then we have to make garbage. It's going to be garbage anyway, so yeah, if you just think about it, if you want to really improve it a lot, yeah, yeah.

TIM: I think so because I'd love to get your thoughts on this, but I feel like part of the fundamental issue with the way hiring has been done is that it relies, at least in a couple of stages, especially the screening stages and maybe the first interview, on taking the candidate's word for it. You look at their CV. They've written about themselves, and you say you have this experience, you say you have these skills; I just have to believe you, but then, as any hiring manager would say, so often that is inaccurate; it's exaggerated; there's lies, or maybe candidates just haven't pitched themselves.

ROMAN: Yep.

TIM: Maybe they're delusional; who knows? But it's just so hard to guess based on a CV or an application whether or not the candidate is worth speaking to. Have you experienced that yourself?

ROMAN: Yeah, for sure. Yeah, for sure. And like nowadays, it's a funny thing that is because it's like AI battles. One is preparing a candidate CV for the job description, and the other is matching it and saying, Okay, yeah, is it true? Is it not? Yeah, that's why it's for sure. We have a lot of not fraud, but like I would say some kind of bending the skills and history according to the description. on the other hand, it was done previously as well, so it's not, but yes, as you mentioned, yeah, a lot of people do, and yeah, for sure, I would say that this is one of the most cases, and you see the really cool CV, and then you're trying to touch the person and, like, asking some hard questions, and everything's ruined. I would say that's the general problem. Personally, I believe we can imagine how to play in this field, but for me personally, it's not even about technical skills because nowadays, at least in the data field, a lot of interview stages are related to those few technical sites. Oh, do you know? I don't know SQL. Do you know Python? Do you know R? Do you know I don't know Tableau or Power BI? Whatever. And they were testing on the technical skills. For me personally, it's the least important skills, like just because, okay, I can't. If a person can't learn a new tool in a couple of months, look, I would never hire him anyway, and it's a funny thing: recently I conducted a survey in my community. It has about 2000 answers, and I asked people what the top three skills for data analysts are, and I divided them into the groups of juniors and seniors, people who I'm asking. And junior people were like, Thinking, okay, it's SQL, it's Python, it's all of this technical stuff, but anyway, 44 percent of the answers were related to common sense analytical thinking and also dealing with business customers. So even junior people understand that it's like 44 percent of the success is not doing the technical skills but in this soft skills I would say, but in contrast senior data analysts, they think that it's even more important, and it's about 62 percent of the soft skills and only 40-something percent of the technical skills. and we can see that it's quite rubbish that we are expecting people to have analytical thinking, like critical thinking, all of this stuff, but we are doing our interviews on technical skills, and in this matter I would love to find the way that AIs and all this stuff can help us just because it's really hard to check these skills. and I do it using business cases I don't know what big consultancy companies do; nevertheless, you should provide at least three to four interviews in this style to truly understand what the person's skills are in this area. Yeah, in my personal opinion, if I would apply some I don't know about efforts in automation; I would automate this part of the soft skills assessment, and there are a lot of psychometrics around it. I'm personally quite a fan of a Hogan test. Have you heard about the Hogan test?

TIM: No, I don't know.

ROMAN: No. It's about It's about like it's psychometric; I don't know how surveys work as well. and you pass it, and one of the great things about it is that it's statistically significant in terms of that they put a lot of people who were going through these interviews and through this service, and after that you receive not just score your 9 over 10 or whatever but you receive a percentile. How do you relate to the general population so you and the top one percentile are saying the same thing, or you're a bottom percentile, or whatever, and it doesn't mean you're better or good if you're in the middle; you're just like other people, but then you're getting to extremes, either low or high percentiles? You just can understand you are like something special about you and so on, and it works quite cool, and I think that if we can, it's quite used a lot in hiring C-level positions, the Hogan Hogan assessment, but if we can make it cheaper, it would be nice, and if you apply it to, like, the day-to-day recruitment process, it would also be cool.

TIM: What does it measure, then? Is this measuring an ability, or is this measuring a personality? Because I know about IQ tests, and I know about the personality model; this sounds like it's measuring something slightly different.

ROMAN: It's something like that, yeah, but they do measure three things: they do measure your motivation—what do you motivate? What motivates you? Also, they measure your weaknesses—what about what you should improve, like in communication or whatever—and also they measure your not skills, but I forgot the third. So they have, inside this survey, three separate parts, and I forgot what the third means, but I took this test recently. I took it five years ago, and they took it just three months ago. Whatever, and I compared myself to what has changed, and it was also quite fun. Yeah, but basically it measures your motivation and strengths and weaknesses.

TIM: That's a good share. Thanks for that. You also mentioned this survey, and given you have it, I remember now we were chatting about this audience you have on Telegram. Is that right, of these data professionals? That would be really interesting to just hear a bit more about that. because that's a really significant sample size, and you're asking interesting, very relevant questions. So you said you asked around 2000 people what their top three most important skills were, and you divided it between senior and junior analysts, and these were the respondents: where they were senior and junior. and you're asking them about what they thought was most important for their roles happen but not for their roles, but not for the rules but for the data analyst role as well as general, so like senior people also were saying not about the position but about the data analyst in general, yep you I don't know about this part.

ROMAN: I've had some surveys related to salaries, but it's more about the market that maybe if not relevant to Europe and all of the stuff I haven't asked it anything like really specific about the structure or maybe skills or whatever, maybe it's a good idea to also have something like that. The only one, yeah, I can think about it was about salary and skills, and we were analyzing how skills influence the salary, so basically we were trying to understand what skills are the most important ones, but yeah, in monetary form, so yeah, yeah, also have such it was like about 1000 respondents, I believe.

TIM: So we've touched on already this kind of changing world where AI is maybe starting to get involved in the hiring process. Candidates are starting to use it maybe to write CVs, sometimes to take tests. You pointed out already that high-end process probably needs to adapt because what's the point of testing someone's SQL skills as an example if you're now going to use a large language model to write 99, for example, and you're saying the soft skills are increasingly more important, so then what does that mean for the next few years, almost getting back to that first thing I was I was asking you about which is the development of the roles; is it then that these roles are going to become less technical, or is it just that the technical skills are different because now it's going to be that you're going to have to be amazing at using large language models rather than interacting directly with the code itself? So it's like an abstraction layer, like where can you see that skill development going?

ROMAN: Yeah, I think it's going to be like some kind of division inside the profession itself, so I believe that most people are going to be like some kind of product managers or product owners, and they just will have a lot of tools on the site to analyze data, and it's going to be a mix of product management and business product analyst roles. So, I think like it's already happening to be honest, just because product managers are more and more data sophisticated right nowadays, and they really dig into the data, so yeah, I believe that one of the streams is going to be more about product-like people, and the other stream I think is going to be more about technical stuff just because it's also quite a funny problem. I was thinking about it, for example, if you have okay nowadays you don't need to write complex Python SQL just because just do it for you, but how do you do really complex stuff? You will never—you will nevertheless need to have these coding skills, and I don't know applied mathematics, applied statistics, all of this complex machine learning stuff, and all of this high mathematics, and so on. and the problem is how are we going to get this specialist because they will need them in any way? Yeah, but if you're a junior, you have never written SQL yourself, how will we get these really advanced professionals, and that's why I think I believe that is going to be segregation? just it's going to be less technical, and it's going to be a lot of people who are doing less technical work, and it's going to be less people who are doing really complex and advanced work, so something like that, the same stuff as we've seen in science, to be honest, like I don't know, 100 years ago, you just can, I don't know, put a steam something in the 200 years ago steam in the some of Ken or a tank, and just okay, it's a steam engine, okay, we're doing something, but it was Not so technically advanced and also technically complex But nowadays we have a data scientist who is working with nanomaterials. I don't know something really complex, and I think that this segregation is going to just grow, just like the people who are creating LLMs nowadays. They're like, I don't know what they are; they are just because they are thinking in just the other way we are." and I think that this cohort of people is going to enlarge on one hand, but yeah, some elite data and the list of ML engineers

TIM: Yeah, it's really interesting to think about where this would go. Like, part of me thinks, okay, so now I can see a path in the next year where most programming would now be you just writing a prompt to an AI, and that's going to do 99 point something percent of it. Like, we must be close to that point probably.

ROMAN: In terms of its data-related code, I can believe so if it's like production-ready code for programmers who are creating applications and all this stuff. I'm not sure, but okay.

TIM: Yeah, it is, and I'm going to throw that out there as a possibility at least, and so then you start to think, Okay, oh, so that means we'll need fewer programmers because now the act of writing the code has been automated away, but I wonder if the opposite is going to happen, which is now the barrier to entry to writing software is drastically lower. There are thousands of higher-order problems that we haven't even gotten to solving yet that we could solve, and surely there's going to be this wave of AI-powered companies that are going to replace, like, internet 1 or internet 2 companies and do the thing they're currently doing but just in a drastically more efficient, fast, or whatever way so that actually maybe this is the best time ever to learn how to code because you now have this leverage tool, which if you have just some level of programming knowledge plus AI, you're going to be like a superpower god, whereas if you knew no programming or you had no access to AI, you're going to be left behind. What do you reckon?

ROMAN: Like Sam Altman recently said, there are going to be companies that have billions of dollars and huge revenues, but they're going to be run by a few people just because you can do a lot of it yourself nowadays, and I have an example of that. There is a moment when he's traveling and doing that stuff. and he's a solo entrepreneur. He's like the only one in the company, but he has about 10 products, and as for the revenue, the annual revenue is about 3 million or something like that, so it's only one person. It's not bad, yeah, for him and for sure, like doing his living, and I believe that's true. Yeah, yeah, honestly, it's so hyped, but we went already during this, and during the industrial revolution in the UK, for example, like then it's okay if you don't need to work seven days anymore, like we need just to work five days just because I don't know if listeners know it, but like previously it was a seven-week working week in all of the world. so, and that's why It happens for sure; we will need less people to do the same stuff. For sure, it means that it's not going to collapse, but it's also going to grow up. Like, for example, I don't know, previously you needed to write HTML and CSS, and if you just knew HTML and CSS, you could earn, I don't know, quite a lot of money. and Five 10 years ago, yeah, but nowadays it's if you just know HT mail and Cs, you'll get no money at all just because it's rubbish because a lot of builders and a lot of pre-primed websites and all the stuff, so I think it's, yeah, for sure it's going to be so we are going to automate a lot. We're going to use data a lot, and it's more on the other hand, it's democratization. What is the data, and all of the stuff IT stuff, on the other hand, technical stuff, on the other hand, it's some kind of committee, sorry, so commodity, the IT skills, and all of the stuff is going to be just commodity. Previously it was like, Oh, I know Excel. I know how to use complex formulas to place complex formulas in Excel. Oh, I can even write scripts in Excel. It was like, okay, you're on a lot of money, and nowadays if you just, yeah, just like Excel, nobody even plays it into the CV anymore. Okay, it's everybody expects that you know this skill. So I think it's going to be the same. We are going to solve new problems. We are going to, for sure, for tech people, it's like a bit more frustrating just because our wages are going to be more equal to the other people because for the last 20 years it was like, Okay, we are superstars; we can earn a lot, but yeah, it's going to change. It's because of the immunization of the tool; for sure these skills are going to be not so much valuable, but I don't see anything really bad in it, and honestly in Europe it's already happened because if you're, I don't know, a senior something engineer in a Holland company, for example, in a Dutch company, you earn, I don't know, 90 to 80 K's per year. And you, if you're a tram driver, you're earning 60 or 40, even 50 K's a year, but as the tax is not equal, salaries in the end in the net are receiving almost the same money, so like in this matter it's already happening, and yeah, it's going to go for the, I believe

TIM: Yeah, that's certainly the case in Australia as well. If you're a skilled tradesperson for at least the last 20 years and you had your own business and you were hardworking, you could make unlimited money basically, and yeah, the supply is always restricted and is always low. Yeah, if I were graduating now, I probably wouldn't get into it. I'd probably either go all in on AI and bet that just because of the change in technology, it's so astronomical that you're going to have some kind of upside opportunity. or I'd go into some kind of skilled trade that's in shortage that isn't too backbreaking because there's a ridiculous amount of money there as well.

ROMAN: I think if everything happens like it's going to be, if AI took my jobs, I was going to pet sit dogs and cats just because, like, you are not going to be replaced anyway. People are going to have pets anyway, and it's quite good income, to be honest, because I kept my dog for a month, and the bill was like almost equal to my rent price. and I was like, okay, I'm going to do this in this case if everything happens.

TIM: That sounds like a job looking after dogs. I'd pay to do that some days of the week. It'd be like therapeutic, I reckon.

ROMAN: Yeah, it's also just relaxing. I always have the person come to my house and play my PlayStation while working in the office, so maybe I'm doing something wrong.

TIM: Yeah, exactly. It needs a rethink, doesn't it? Yeah, it's funny. Interesting as dogs are, I feel like I should ask you a few more questions about hiring for data people, and in particular, I'd love to throw this one thing at you that has always struck me over the past five or six years, which is that I would speak to a lot of analytics and data leaders whose day job was about making decisions with data. be it in product analytics, marketing analytics, sales analytics, operations, or whatever, fundamentally they were like, Cool, we know that if we make these decisions using data rather than just making shit up, we're going to make more money, we're going to reduce costs, we're going to have a better process, like whatever. There's some inherent benefit to using data to make decisions, yet the vast majority of people I've spoken to who are data leaders wouldn't have that same mentality when it came to hiring. For hiring, they would generally take a very intuitive gut feel approach, so you can imagine a spectrum of pure numerical objectivity to pure gut feel intuition. I'd say 80 percent of them are most of the way towards the intuitive end of the spectrum. Do you have any feelings about why that is the case? Any thoughts? Am I missing something? Do you think it's wrong? Where do you put yourself on that spectrum as well? I'd love to hear your thoughts.

ROMAN: Yeah, look, I would say I have two thoughts about it. The first one is to be honest: data, like we always think we are, like data-driven companies, yeah, data-driven, but data-driven is not data-managed, and when you are data-driven, you will have the driver anyway, and the driver decides whenever it's like if you see something on your that your speed is above the limit, but you nevertheless decide to go this speed. So, the one thing is that, on the one hand, data really helps a lot of companies; on the other hand, there's quite a lot of hype around it, and to be honest, only companies with big operational processes can earn a lot of insights and a lot of stuff on the money on the data just because, yeah, if you are an Uber or, I don't know, whatever, if you are a Facebook, you have a lot of transactions. people visit your site or people using your app or whatever, and it's been if you it's a scale thing if you just improve it by one one, I don't know, 0001 or whatever, you're going to have an enormous effect just because of the scale, but for example, let's talk about our company. We are selling—we are called provider for for GPUs, and almost all of our deals are B2B, and it's really okay. I don't know; there is data not even enough for megabytes in terms of sales analyzing sales, and it's all going through the people talking to each other and all that stuff, and in this case for such a company Data-driven is okay, but it's not going to have an enormous impact on yourself, so my first take is that the data insights and data stuff are a little bit over-evaluated in my opinion, to be honest. And by the way, if you have ever been in a meeting with a C-level executive, let's say it straight: they don't use data as much as we would love them to do; it's a lot of gut feeling of them who are running the companies, so it's going to be the same anyway, so that was the first take. the second take about applying metrics and data to the hiring process I believe it could be a good thing, especially if you're hiring on a scale. Just because if you're hiring, I don't know, one person in a year, you don't have any savings on that and quality improvements, but if you are hiring really on a scale, for sure you can, first of all, automate the process. Yeah, just cast cut the costs, so secondly you can improve it and use metrics, and in our company and the companies I worked with before, we didn't use a pure metrics approach just because, but on the other hand, it was like we always were assessing skills and applying some level to it. Is it poor, is it okay, is it above average, or is it like excellent? So yeah. So cool in some matter, it was data already. Yeah, it was a few data points, but like it was data itself. Yeah, but we were not, I believe, like analyzing it on a scale as well. I'm thinking, sorry, a bit like thinking and getting my mind up. So I would say that, yeah, for sure we can do it, but mostly I believe so for short of the period of the hiring and make it more efficient, but in terms of quality and in terms of matching to the team and all of this stuff, you will need to have a conversation anyway, so for sure maybe we can try to automate this as well, but if it's going to work, I believe we will maybe we are going to be already replaced by EI. okay

TIM: So you don't need the people anyway.

ROMAN: Yeah, you need two people anyway, to be honest. Yeah, just because I know of such a lot of cases when we were hiring people who had really cool skills, really good CVs, and really good previous experience, they go from funk or whatever, but on a personal matter, they just didn't fit the team. So, okay, it doesn't matter what your skills are; if you don't work as a team and you just don't match the culture, like, it's going to be ruined anyway, so that's why I think that if there is a human being in a loop till we replace it, it's going to be some kind of human assessment. anyway, I personally

TIM: What about just throwing this at you? Sorry, so yeah, there's still a subjective layer, and there are still these things that maybe don't have a black-and-white answer, so you could easily measure someone's SQL skills, but you probably can't easily measure how well this person would interact with that team. but you could at least measure it rather than not measuring it, so, for example, you can have every interviewee give their opinion, like, how well does this person match that team on a scale of one to five or something like that, or what about that kind of approach? Is that better than just

ROMAN: It sounds cool. We just started the last couple of weeks. We started to do chair analytics at the company, and I'm responsible for that, and I feel like we'll apply some of those approaches you're mentioning. Just for now, it's like only we just need some operational stuff. Just what takes so we are working with data, but on a matter of how can we hire faster, so we are looking for the hiring managers or the leaders who provide the interviews. Do they, I don't know, lack time, and does it influence our hiring process? Are there some, I don't know, delays on the HR stage and screening stage and all this stuff? So we're using data, but not for assessing the candidates themselves but for assessing the process. And. Yeah, so finding the bottlenecks, trying to Yeah, they're just like,

TIM: Between each step, the conversion rates, yeah

ROMAN: Yes, the convergence rate here is why it was so, why we said that we are not going to hire this person. What are the reasons and all this stuff collected? Yeah, but it's in my terms; it's, yeah, maybe it's so obvious for me just because I don't even consider it like data something just because, yeah, for sure you need to measure the process for sure you go through the stages, and for sure you raise the conversions, and who are the interviewers? Are they the most not rude but the most hard one to pass?

TIM: Yeah.

ROMAN: Yeah.

TIM: Yeah, in terms of interviewer analytics, you could almost call that, yeah, I've had a few people attempting that, but that's rare, like even that level of thought process you've put in.

ROMAN: Yeah, maybe. Yeah, we have. Oh, cool. Okay, yeah, no, maybe it's so natural for me. For our company, we are quite data-driven in this matter. We're quite a data-driven company, so yeah, maybe that's why I have such a culture that, like, okay, yeah, just the process of that should work.

TIM: Yeah, I find anyone who I've noticed so far, the pattern is if you've worked in marketing or marketing analytics, and you already just naturally think of a funnel where there's conversion rates at each step, then hiring is identical to that, and so often I would find people with that background would then apply the same lens, which helps a lot and is such a low-hanging bit of fruit, I think.

ROMAN: Yeah, and I really like to draw sun K just because if you have people on the side, like they're going through the different stages and for different laws, like, yeah, and just a keen on the realization, so yeah, it's my favorite.

TIM: Roma Thank you for the interesting and completely relaxed conversation today about a whole bunch of different topics. I'm glad that we got to discuss dogs and dog sitting at some points. It was not quite what I was expecting, but I think that's a nice value add, and you've made me want to go and pat a dog. That's for sure, if nothing else.

ROMAN: Yeah, thank you. It was fun, and thank you for having me, and it was interesting, and I look forward to what you are going to do with your product and how you want to apply AI for your field because I'm really interested in doing this.