Transcript
Speaker: What used to happen, you would have a column and then you have a new version of it and then you have the old one for like a backwards compatibility. but you know that you shouldn't be using it, but only people in the know using that table. So everybody's aware of that. Whereas now you're pointing the eye and it's whatever it feels is the most comfortable thing to do. But but AI doesn't look at it as a wrong data. AI, again, even with a semantic layer, the problem you might have, unless you force AI to go through semantic layer, like for example, Looker MCP would require you to go through ncp a semantic layer, you can't access SQL directly. But if you, for example, have a DBT semantic layer, depends on whether it's enforced or not enforced.
Speaker: Today's episode is brought to you by Omni. Most companies I speak to want AI analytics but are failing to put projects into production. That's where Omni is different. It's the semantic brain that grounds AI in the heart of your business logic, giving you governed answers, whilst also the depth to identify root causes. It's intelligence everywhere, from your spreadsheets to their chat feature, even within your product. Omni moves you beyond the dashboard.
Speaker: Don't just take my word for it. Trust teams like Perplexity and Synthesia that are already using Omni to deliver intelligence that people trust. Check out them in the show notes or visit omni.co. That's O-M-N-I dot co.
Speaker: Now, back to the show. Hello, everyone. Welcome to another episode of the Stax Data podcast. I'm your host, Harry Gollop. Most data teams measure success by what they ship.
Speaker: The dashboard's live, the model's deployed, the pipeline's running, job done. Except it definitely isn't, and particularly not in today's age, because sometimes you find out that nobody's actually using what you've built, or you're not able to quantify what the impact of what you've built actually is.
Speaker: Today's conversation is about why that keeps on happening and what actually fixes it. And often it's not not better tooling, it's not more headcount, it's not a different data stack, it's it's something much more human. And it's about whether you and your data team genuinely understand the business um it's supposed to be serving and whether you're willing to be pragmatic enough to meet the business needs rather than maybe what's the the most fun or or technically exciting to to build. so
Speaker: Today I'm joined by one of the leaders of of data in Carwell, Roman, who heads up growth and analytics for the business. He's had a fascinating journey, which I will let him introduce and I think offers a really unique perspective into some of these challenges and how he's tackled them at Carwell. Roman, it's it's great to have you on the show.
Speaker: Thank you very much for having me here. Yeah, so my name is Roman. I'm a head of marketing and media analytics here at CarWow. I have joined CarWow for about five years ago. So I have been here for very a long time. I'm looking after everything when it comes to bringing data in from the platforms, building pipelines, as you said, ah wrapping the data around and then sending it back to either platforms or stakeholders. But the the bigger part for me is basically supporting business driving decisions and making sure that people understand what that data means and what it does. I also to work with a consultant as a consultant and help small and medium clients to sort their problems and data stack. And with with with AI, especially nowadays, it becomes very, very difficult to to drive decisions because getting some data is now very easy, but then making that data useful is insanely difficult.
Speaker: I couldn't agree more and I think it's something that we've seen ah across the industry over the last 12, 18, 24 months. When we're working with our clients, the the need for candidates and data folks to be able to talk about their their impact is has never been higher.
Speaker: um I think it's it's often the can be harder to to do. so um Yeah, Roman, like why are the most effective data teams stopping to think about delivery and starting to think about impact? What's your what's your view on this?
Speaker: Well, I think in general, everybody in the business should be aligned to driving results and impact. But the difficulty about the analytics team specifically is that in in most companies and in the Cairo as well, we are mainly enablers. We let other people do the best decisions. We help them not to make mistakes where it needs to be and help company grow.
Speaker: What's really difficult is how do we connect our analytical knowledge and and move from just answering a question to actually driving the conversation and being there. And with the current market and current global economy situation and generally where all the companies are going right now,
Speaker: it becomes more and more important to have that one all aligned goal where both analytic partner and the stakeholder and sometimes a wider group are all driving towards one goal.
Speaker: And then you see more often embedded teams, you see more shared OKRs approach, and it actually becomes very, very difficult otherwise to function because it's it's no longer possible to have like 25 people analytics that just do analysis. You need to make sure that everybody is 100% on the goal and rocks that essentially the company decides is number one priority.
Speaker: And how does that actually work in practice, Roman? like I think we hear often that many folks are looking to drive outcomes, but they don't necessarily know how to get there. So, yeah why what what is your approach to making sure that you're going to be working on the the right projects and the right work for a team?
Speaker: I would even probably start a bit earlier, to be fair. The number one thing that I build with my team, with everybody, is making sure that we build a very trusted relationship with people and with stakeholders. And that means that like there are two different approaches you can take. ah What you see a lot is people don't want to bring analytics people to the meeting because and they're going to ask questions and they're going to drive the conversations and make it so complex that it's easier to agree on something without them being in the room. And this is absolutely like foundation new wrong way of solving problems because we want to be there partnering. And then when when I work with my team, um my number one goal is like your
Speaker: target is to build a relationship in which at every single point the stakeholder feels that bringing you in the room will benefit the conversation. it will bring something to the table. And that starts with you understanding the business problems they have, not the data problems they have.
Speaker: Because the data problem is decision making, right? You try to answer one question or another. Should I go for this marketing challenge? Should I switch your budget this way? Should I optimize the conversion rate? But behind that, you might see that the main stakeholder is actually really balanced by p and l constraints. Like you would say we need to spend more money here because it makes sense, but then the P&L restricts on how much they can actually account for that. So that completely shifts the conversation. Your recommendation analytically can be right.
Speaker: But then in the in the situation where the company is right now, where there's stakeholders right now in strategy, that might not be the best approach. And you only know that if you have a constant communication with them and you have a constant relationship that you consistently build over time.
Speaker: does that How do you actually do that though, Roman? You said obviously don't look at it as a data problem, look at ah yeah understand their business problems. I think one of the things I often see is it's really how do you actually go about understanding and having having that commercial lens? Have you got any advice or strategies um to really digging into that commercial element?
Speaker: I think every business is very unique in that sense. What helps me there is that I came to analytics from being a person, being a person in marketing. I used to manage my marketing campaigns myself.
Speaker: And then when I work with my team, I can walk them through every single aspect of it and say, well, this is what you think, but this is the actual reality from the other side. um What we build internally is obviously we embed people in the team. So if I have a person who is responsible for paid search, that person is going to come in every single meeting that that team has. it will be part of They will be part of the OKR planning. They will have a specific goals on building relationship with the director level, which means uh regular catch-ups understanding what's there but once a quarter going to be exact level conversation where we're going to sit down together and say okay uh on the cmo level what's the three number like what's the number three priorities for the next quarter at the board level that you have to like discuss that you want us to solve and those problems are usually like let's say okay we wanted to drive diversity channel diversification which means you don't want to spend all the money in Google, you want to spend money in different channels, you have less dependency on one traffic source. This is something that is very familiar for many, many companies of this size.
Speaker: But then when you come down to stakeholder level, every single person going to have their own targets. though It's really hard to to make changes to that level. So you do need to be having relationship across the whole vertical.
Speaker: And so one of the things we're going have, we're going to director level set of work yards and targets when we help solve those particular questions. And like, if we were to diversify which of the channels we're going to address. And this is one level of conversations. And then on the day to day level, we are part of, again, stand ups and all of the day to day things.
Speaker: One more thing that we also do is we make sure that everybody on the stakeholder side, in our case, it's marketing media, feel comfortable raising any single concern they have.
Speaker: Like, one number one rule I have in the team is if anybody says the data is wrong or they feel the data is wrong, they should be comfortable coming to us and not assuming they're doing something wrong.
Speaker: Because if you don't have trust in what we're building and or data that you see, you will never actually have trust in our relationship. And yes, 9 out of 10, they forget the filter or they they approach or they look at a different thing that they shouldn't be looking at.
Speaker: But then they feel comfortable raising that and they will feel comfortable with the data they are provided with. So it's giving that stakeholder, I suppose, the security and the confidence that um too that they can come to you.
Speaker: um they they can that that That you're not there to to be a blocker or to, i suppose, yeah may undermine them. But how I think one of the other things that we've mentioned before is like, that's around the, yeah once you've built the product, how do you get in the room in the first place? Because I think that's so important, right? I think often we see teams where, you know, that there might be a new product being launched and they the product team come and, or I suppose in growth in your sector, you know, we've we've just run X campaign, how did it perform? And that's then a very different, but you can't answer a question as well when you're you're coming in at the end. so So how do you make sure you're in the conversation at the beginning?
Speaker: It's a very difficult question because it's a case by case case a situation. It's says a stakeholder by stakeholder. You need to remember that foundation, you're all here trying to achieve the same goal.
Speaker: And then you will be invited in the room when you help people solve their problems and they happy you help them to achieve the targets and the goals they have. And this is what my goal is to understand how we align on those.
Speaker: What people don't want you is to come there and say, well, everything you did is wrong. and The question you're asking is wrong. Everything you do is wrong. This is not the way approaching it. Because you might be right with like judging everything, but it doesn't actually give people any feeling of like, how do we resolve it?
Speaker: You need to come there with a mindset of here's how we can move move forward. This is the steps I can do for you. you need to be very careful not accidentally grabbing like 100% of the work on the table that can be solved by other teams. There is like another aspect of that.
Speaker: But fundamentally, you start with small things and then you grow step by step. In my case, for example, we always like my, like the huge thing that moves conversation forward is education piece. So when we sit down with stakeholder, explained to my team is like, they would not have as much analytical knowledge as you do, they will not ask the questions that you do, because you are way more You feel like you know me what much more about it. So your goal is to help them get on that path, to take them on the journey with you and help them understand how the data is structured and understand them what metrics mean and what you can and cannot do that.
Speaker: And it will never happen overnight. It will take weeks and weeks of education. But once people understand it, they feel very grateful and they're very happy to sit down with you and and like and and and work with you.
Speaker: Interesting. And that's around, i I suppose what you're saying there is the about how do you influence behavior rather than necessarily taking taking orders. um so you don't need to just take an order from your stakeholder. You do need to influence them, but you need to bring them on that journey with you. Is that is that what you're saying?
Speaker: Yeah, I mean, the worst case scenario, one of the worst case scenario can be is your ticketing system, right? People come there, they drop a ticket for your team, they don't want to chat about it, they just want you to get the number out. You do the thing, you get the number out. and And the reason why it's bad is, A, nobody really validates whether that request should be taken as a priority. You never had a conversation about it. But second, you also don't feel that you're driving an impact. So for the team member, it's hard to feel they're helping business grow. They're essentially moving data around, they answering the questions.
Speaker: um A significantly different approach would be you are at the table like, okay, we need to hit this goal. How do we do that? And you're like, well, we need to make a user journey analysis. We need to understand which pages don't perform. There's this idea as an audience that we can, for example, exclude those like cohorts of people or exclude those cohorts of people and then the marketing person's gonna say well this is what we can do this is we cannot do and then together you understand the overlaps and it's an open conversation it's just as much an open conversation as for example in my team the number one communication i give to everybody in my team is that if you feel that i'm wrong as a manager you should be comfortable telling me that i'm wrong
Speaker: Because if you don't feel comfortable that I'm wrong, you don't have accountability, you can't actually influence the things that you're owning and that we are in a very wrong relationship. The same happens with the stakeholder.
Speaker: Number one thing is that they should be able to comfortably raise the concern, but they should also be comfortable with us raising concern with them. And that is only possible if none of us feel that we are threatening. And again, that sounds weird, but we're all people, it's companies, it's career path, it's journeys. So it becomes really difficult when everybody has a very strong opinion that they try to get to get their own KPIs and they don't align. So how do we fix that? We put people in the room and we try to put them in the same boat and say, well, all of you try to share the same problem and try to solve it.
Speaker: What about that final stage, Roman? i think that's um I like that sort of analogy there of, I suppose, shared problem solving when you're trying to come to ah yeah a root cause analysis or trying to understand a requirement.
Speaker: What about when you're building something that is a little bit longer lasting and and you need to drive adoption and behavior change and you want to see these changes How how do you work with stakeholders and teams and to make sure that you are yeah you're you're shaping that behavior and it's it's longer long longer lasting rather than just a flick of a switch?
Speaker: There are a couple of examples just about that that came to my mind. The most recent one was the AI adoption tool. So obviously everybody's trying to get on by either Claude or Codex, depending on what you use. The thing is, there are two different types of behavior you see. one is People that are miles ahead of everybody else. They are driving adoption, they're trying their best, they know everything before you even know. that in the morning on Monday like, look at this thing that changed the industry and you're like, wow, I'm i'm feeling really behind. And then majority of people, they are on opposite side where they they they being told that AI is amazing, but they constantly feel that it's just really hard to get your hands in and like try to do something with it.
Speaker: AI is insanely overwhelming. So how do you solve that? Well, you can come there and you say mandate everybody to go and do that. Surprisingly, that doesn't work well because do what is like it's not an instruction you can go and try. if If I ask you to go and cook me something, that something is not something like a recipe you can follow, you you will be really like stagnated by what to do.
Speaker: ah We in our team decided to do well, we're going to build something like i said, a set of skills that help people to get some basic answers. thing And then we're going to person by person.
Speaker: Try ask them to do that with us for half an hour and then see what problems they encounter. Because once you get them that feeling that they understand what to do, they start push push pushing boundaries, they start to to feel safe and comfortable. And ah surprisingly, the same approach works with every other thing. Again, you build a tablet dashboard, we build a Omni dashboard, doesn't really matter.
Speaker: If you build it without having any feedback from the other side on how they use it you will never be able to drive adoption because that's your vision, how they should perceive the metrics. And then, while in a lot of cases we're wrong because we we don't do what they do, like what stakeholders do all the time. Same happens with any other change.
Speaker: Sometimes it's a bit more complex. though The same kind of example in my in my mind was attribution change. So a few years back, we went there, we changed our signup based last click to like a multi-touch attribution.
Speaker: And for those that don't know, attribution essentially answers the question, which channels or which people's visits drive which conversions. And when you move from something like a last click, when there's a one-to-one relationship, a very simple model to like a multi-touch thingy,
Speaker: Nobody understands it fundamentally because numbers are now spread across different visits and people just want to know which visit drove which conversion. You're going like, it's a good question. You can't answer that because of multiple of them.
Speaker: ah What we had to do is we had to do a several months education process. So we had to start with marketing, we had to bring exec on board, we had to bring directors on board. We made a video recording internally that helped people understand how how how to do f FAQ.
Speaker: And still, it took us months and months and you need to be patient. It is difficult to transition to something that people fundamentally don't understand, but it impacts their work. And they should be uncomfortable with it because obviously they own the metric they don't understand. And it's your goal to help them get there. So it's either you do a very long term transition, but in any case, you need to sit down and understand what's the day to day problems and get that feedback back to then iterate from that.
Speaker: That is a really interesting point. So you're saying that, um yeah, it's really part of your job to make sure that your the stakeholders, the business, do understand the, suppose the metrics, but also sometimes they don't have to understand the full thing behind the hood, but but how they they are getting to this number. Why is why is that beneficial? is Is it because they can then empathize with some of the challenges? but Why is it so important?
Speaker: I really struggle to see the other way around it. Like, I understand that there are different companies and different approaches, but unless you go through this education piece, and by the way, it goes both ways. Like as an analyst person, my goal is to explain the technical limitation of data and how we can use it and what we can build. But as a business owner, their goal is to explain to me how the business works and what happens in a lot of cases, I make assumptions that I just plainly wrong or i don't take into account certain things. Again, there multiple different industries, but um
Speaker: like one of the companies I used to work for was the travel companies many years ago. And essentially you buy tickets from place to place on the train tickets, for example. And I was like, what's problem with that? Like why we're not hitting the numbers? and Like, well, because it's very highly, don't know, seasonal market. And then seasons depend on what country you advertise for. And if, for example, people travel in Europe,
Speaker: um and they're from America, they will be only traveling during the high season. But if you sell the same tickets to Europeans, they will travel all around the world because the travel for them is way, way simpler. So you need to split the markets in the absolutely different ways and they're going to surge very differently. It sounds obvious because now explain it obviously, but when you deep down in processes and you look at the numbers in Excel sheets, none of this makes sense because you see the numbers go up and down and you make assumptions that all of them should go up at the same time.
Speaker: But they can't go all at the same time because fundamentally they're driven by different processes. Interesting. I think, Roman, one of the other things that I see people struggle with is pragmatism um over perfection and when do you where do you join the line between balancing the business needs and the here and now and and especially in I can imagine in organizations like Carwell and Scale-Ups which are are trying to move very quickly and adapt to the environment and business the business stakeholders are are having to make decisions on the on the fly sometimes more with
Speaker: with with with their gut because they can't wait for a for a rock solid answer. But then there's also the times where you do need to have the a bit more around the perfection and the perfect areas when it comes to the technical side. So how how how do you break down balancing that need of a business in the here and now versus the the longer term sort of work and rigor?
Speaker: well it's a really good question. I think myself, I was on the journey. I come from like a bit of a marketing, but also engineering background. So historically, I was like, we need to build a rigorous, long term durable system to solve every problem.
Speaker: Business doesn't always need a rigorous system. You don't need to spend a year building something that will be redundant in two years from now. It's absolutely wasted effort. But there are some things that have to be build like bulletproof basically from day one.
Speaker: I'm a person who has a mindset of get things done and get people unblocked as soon as you can with only one exception. If that process you build or whatever you're trying to answer will have a huge business impact. Let's let's imagine every once in a while, every analyst have to build an exact level of review like a or or board review or maybe mna which is emergent acquisition of course in that case you're go to triple check the numbers and this is absolutely a different approach but when you build a reporting you also need to take into account how what is the risk factor on you getting things right getting things wrong and like approaching them all of this needs to be taken into account The good news with adoption of AI, building a good DBT structure or whatever the system you use becomes significantly simpler because all of that is now afforded to the AI agent. um
Speaker: The bad news is now everybody scales their code base like five, 10 times faster than they used to. So it becomes really difficult to approach. Internally, we use we have a system, which is attribution, which is essentially answering number one question, where people come from. And that system then yields answers and that goes into Google Ads. And then if anything wrong with that particular system, app company has a real revenue hit and real cost hit.
Speaker: So if anything changes there, we have a multi-step review process where every single person needs to go through validation. There is a checklist. There's a huge amount of things you need to do and test you need to fulfill because there's a very high risk.
Speaker: And then there are rules. You will never be allowed to merge changes on Fridays because you don't want to people jumping it on Saturdays and there's going to be delay in response. After that, there is a reporting layer.
Speaker: That layer basically answers the question, how we doing? If it is broken on the weekend, it's bad, but it's not gonna hit to the point where people are gonna struggle with it because most people don't work on the weekend, as they shouldn't, um and it's fine to check it on Monday. So we have a different rule there on like how much you can change and you you are allowed to make changes on Friday if you feel like it, and you have to validate something to less things.
Speaker: um So that's the framework we use internally. If it is, what's the revenue risk and what's the decision making process there? Now, when it comes to getting, not building the system, but answering the business question, let's say you need to answer the question, what's your CAC LTV, whatever the numbers you want to get to.
Speaker: yon You should take approximation where possible and you should move faster where possible if you don't have a business risk for that. Usually there is no
Speaker: let me put it this way, more analytical approaches with more safe to say, if you're trying to answer the question, which landing page is better, it doesn't matter if it's 5% off compared to everything else. You just need to know which one is better.
Speaker: um There is a huge ah framework in the business that every single decision needs to be A-B tested. And in a lot of cases, yes, if you're big if you're Google, if you have a search page, you should A-B test it on a really small sample sizes. But if you're just launching new business out of, let's say, youre launching a new vertical,
Speaker: Moving fast is way more important than A-B test every single decision you make on the way. And then post, pre-post might be a better choice. and And this is the framework that you need to adopt within your within your data team.
Speaker: You need to have a comfortable level of risk where it's more beneficial for the business to move faster than it is to be absolutely 100% right.
Speaker: I think that's great. It's the cost of doing nothing. Or being slow. Is another way I think of of looking at right? um What is going to be the the cost and the impact if we don't get this the this right or this to you straight away? And and then you're weighing up that with the the the trade-off of of of and of time and and what that more accuracy or or deeper ah validation might be. So I think that's a really good way of of looking at it like I think in most things in in in data it's all about the trade-offs right of what what what is the impact of the of the different decisions and think one of the things that obviously we've you've touched upon there Roman but also comes into this sort of pragmatism versus perfection is how AI is is obviously changing that equation
Speaker: um yeah things is much is much quicker to rapidly prototype and quicker to build, but that can obviously have a huge impact on um tech debt. you know People love to build, they they don't like to to to to delete. and i think the yeah Where's your view on how AI should be used to change the was the economics of of data work?
Speaker: I think I'm going to start with a believe limited with the words like I'm a AI believer. Like I believe that the AI is an amazing thing. It's here to really bring a lot of good changes to our to our positions. And yes, it's going to change what we do. It's going to merge a few roles. Our roles going to be more technical on the way because you can now deliver so much more and they could be wider. But also the work becomes much more fun when you can build whatever you need on the way very, very quickly. um
Speaker: I think there are two different parts to AI. One is the analytics engineering part, late engineering part, to getting things done to to do the piping. And it's essentially like coding, as long as you know what you do,
Speaker: It's almost bulletproof, like it can drive insanely good outcomes that are very easy to test and very easy to validate and very easy to automate. And majority of the time you spend is actually thinking and planning what's going to be the good infrastructure rather than building it. And in my experience, we use AI for about 95% of our pull requests internally, we write enormous amount of code with it.
Speaker: It's It's almost failure proof. Like its it's very rarely hallucinates in the way that it drives the drive outcome. As long as people driving the AI actually know what they're doing.
Speaker: Well, the second part of it is analytics. Analytics is way complex way more complex because again, ai essentially is a people's pleaser. It tries to please you with the most probable answer.
Speaker: And that answer, not always what you actually needed to get to. And it really, really depends on the question you ask. and It's hard to make sure that it gets 100% things right.
Speaker: because a few things going in the way. So the first thing that going in the way is the the person who asks the question. So if you if you get AI, raw AI, let's say Claude, and you give it to a stakeholder that has no idea about how the data is structured, they can ask an amazing question, but that question is so biased fundamentally that the answer is going to be just as biased. And I have seen this happening all over the company. and this is fine. This is just the way people think because they're not that They don't understand everything underlying every underlying data data point. um
Speaker: Solving that is really difficult and it it comes with building guardrails, building semantic layer, building limiting on what people can self-serve with AI, people can't self-serve with AI. The more data proficient the person is, the better the questions they the more can get the more you can get with ai But the second thing that happens, and this is essentially the framework around it, we are right now in situation where majority company of the companies have built their data layer for people.
Speaker: So it is actually expected for you to know a lot of things to be able to navigate that. Whereas majority of the execution and questions now will be coming from AI. And if you would be building, let's say, a table for AI, you would not build it the way we used to build it.
Speaker: And what I mean by that is you need to name every single column in the most simple probabilistic answer so that AI looking at the column would get it most probably right. Yes, semantically it helps, but the more semantic layers you you you throw into the AI, the more context to window, the more likely get things wrong. So your whole goal is becoming building the data as simple as possible.
Speaker: A great example, I had a table that were assessing clients' performance, essentially people coming to buy advertising on CarWow, and then that we want to know how much they get out of it.
Speaker: And we had a column which said revenue. And then the AI went there, assessed the column and said, well, these spots grow that much revenue. They're absolutely amazing for clients, except that was a karma revenue. It's what we earn. It was actual cost for the client. So it's the opposite way of looking at it.
Speaker: ah But AI didn't have that context. This is revenue. It says revenue is good. The higher the revenue, the better. So this is the answer. And then people would make those decisions. Whereas if you would be building it now, you would actually either call it the cost, but you would called call it revenue for car out so that it's self-explanatory. So yeah, this is essentially comes to two different parts. Well, how good your data system for the AI native solutions? And second, how good the questions that people ask and how many guardrails you have in and the place. I'm a strong be believer that people should use as much
Speaker: AI is possible and it will create a lot of questions for the data team and we'll have to go through this for a moment when we all learn how to navigate that. But it's better that nobody uses AI and then you're not getting the advantage of AI that is there on the market.
Speaker: Yeah, I couldn't agree more with you. i think um you have to be moving with the times. um You are going to fail. There are going to be mistakes. and There are going to be challenges, but you you definitely learn from them. And I think the organizations that a bit more risk on, um on this and are aware of the risks are going to be the ones that are going to find find the way through to hopefully hopefully value because I think it's it's it's clear you you make a really interesting point there about I suppose your your data estate and the importance of it being
Speaker: I suppose everyone uses the term AI ready, right? um I think one of the things I heard another another leader say is back when when when he was a bit more when when he was a practitioner, he said, like, I was the semantic layer for the business. And um I think that's a prime example of of that kind of um question. you As a data person, you'd look at that column and you've got that semantic knowledge of what that column means, um whereas the AI takes it at face value. So I think, yeah, you have to make sure that in a world which we're moving to, that everything has been
Speaker: extremely well documented in a clear and concise um way so that semantic knowledge isn't locked up in the in the heads of your your your data team. 100%. one more thing that just came to my mind is that you actually have to go and clean up the mess much more often now. What used to happen, you would have a column and then you have a new version of it and then you have the old one for like a back backwards compatibility, but you know that you shouldn't be using it, but only people know, like in the know,
Speaker: using that table so everybody's aware of that. Whereas now you're pointing the AI and it picks whatever it feels is the most comfortable thing to do. We have a few of those in one of the core tables with essentially like a backward compatible things. But now, yeah, it constantly goes there and picks that up and we're like, oh, we need to solve that. It's just really, really bad for the business. It's it's the wrong data to use. But but AI doesn't look at it as the wrong data. AI, again, even with a semantic layer, the problem you might have, unless you force AI to go through semantic layer, like for example, Looker MCP would require you to go through MCP semantically, you can't access SQL directly. But if you, for example, have a DBT semantic layer, ah it really depends on whether it's enforced or not enforced.
Speaker: and The problem is like it might take that into account, but it also might ignore the context unless you explicitly ask it to do that. This is where the whole infrastructure coming in place and how do you enforce that and how do you create those situations where people, what people should self-serve, what people should not self-serve.
Speaker: Yeah, I think it's, um you know, AI is, in if you look in the media, it's coming for everyone's jobs. um And yeah, I think it will come for for jobs. But I think in the the data space, I think there's it's it's adding more complexity, more engineering rigor.
Speaker: um that is needed to actually get value from it and the AI can't do that. You need that human human and in that loop to create an environment where the AI um can thrive. think you know it sounds like, Roman, you're doing just as much analytics engineering, software engineering around build designing and building these systems as probably more than ever compared to to your days as an analyst. Is that a fair sort of assumption?
Speaker: Yes, that is 100%. And that's my general belief that for, especially for the small, medium-sized companies, what you're going to see is roles are going to be merged. Instead of having like data engineers, alice you're going to have like analytics engineers, which prepare their old data sets, but they run analysis on it.
Speaker: What happens a lot, and this is an interesting take actually, that Now you can come up with an idea and if you have right set of skills and right setups and yada yada all around it, you basically come there, you prompt AI to write something, you go to the meeting, then you come from the meeting and it's already there. You can go and try it out. Instead, you used to go and spend a few hours building it. Now it's just built for you. You can go into cloud, auto mode, you can the planet. There's a lot like depending on your technical knowledge,
Speaker: There is a lot of guardrails you can build yourself just to make a really high level execution, really high quality. Now, what why why we don't see like tripling, quadrupling outcomes in the businesses? Well, there are a lot of other things. One, writing code is only just part of the job.
Speaker: Like you need to ask a good question and in a lot of cases you need to sleep over the problem and you need to take some time and you take beers and then go to the pub with your friends and then you're go to be having a third game like, aha, that's the moment when you realize the question you need to ask because we as humans, we're not all, we need time to process things.
Speaker: The second thing that happens is lots of time in companies are spent on syncing and talking and going to the meetings and doing documentation and lots of things are still important. You still need to have your time with execs and you can't but expect everybody in the business just use AI to answer their questions.
Speaker: It's going to happen more often that people self-serve, but it's not going to be to the point where everybody will understand how Google Ads works or how j four works or how do you need to make sure that your ICO complied for cookies. There's a million things in there that still takes a lot of time. And lastly, just because you run more tests doesn't mean that the test quality is improving as well. You will run into a situation where running more tests, for example, does not quadrupling your outcome.
Speaker: It might take twenty thirty percent advantage. or what it used to be, you're still going to hit the sample sizes and idea stagnation and investigation and thinking. There's a million things involved in that which are tied to our human physical world that you can't AI automate.
Speaker: But there is amazing things that AI does. Like one of my favorite things that it can now do my Kanban board without me actually needing to put every single ticket. I ask my cloud to put it and it's going to pre-fill every single field and it's going to take a Slack message as a context. So annoying parts of our world that and now can be offloaded to AI and we can focus on the really cool things and asking really cool questions.
Speaker: What's your advice for folks to stay at the forefront? you know What skills should they be perfecting in this ah in this dynamic time? It's a really good question. ah I think the first thing I'm going to tell to everybody is it feels like everybody around you is doing building a multi-billion business with AI and you're the only one on the on the in the back. That is not true.
Speaker: We're all learning and we're all trying to understand how to apply it best. And we're all in that situation where it's really difficult to feel. It's like, you know, you sometimes you open Instagram and everybody's living a dream life. It's just not true. That's just a social media kind of fact. The same happens with AI.
Speaker: Take your time and focus on one thing. The second thing is AI really comes down for me in that intersection between technical and business thing. The more technical understanding you have, the more you can do with AI, the more you can deliver with AI. Let's take an example.
Speaker: For example, if you have zero understanding of SQL and you need to ask AI to query a database, you cannot validate anything about what that actually does. Whereas you have a basic understanding of SQL, you can maybe double, triple the quality of outcomes.
Speaker: Again, putting the semantic layer aside right now, just a separated, isolated example. So don't think about the AI as it's going to replace it. Think about it as an exoskeleton. It drives, it multiplies what you can do. If you can't do anything in the field, you're probably not going to get to the 10x, you're probably get to 1x. But if you are really good at it, you can 10x your output.
Speaker: And finally, is like a very talented junior assistant that you can get. So what it means, you can ask the junior assistant to do anything in the world, but the way you ask will hunt and influence what that assistant is going to do. You will never hire a person who has never been on the job and ask them to go and fix a thing. You're going to say, well, do this first, research that, come bring me this, I will validate that, then do that, that, that, that. That's the way you would delegate.
Speaker: And what we see a lot is people who are naturally good at delegation and formulating what the outcome should look like, they can get so much more from AI than people coming to AI and saying, can you fix a thing for me?
Speaker: And it happens that majority of of us professionals working with other professionals talk to each other in a way. Hey, Harry, did you fix a thing? Yeah, Roman, I fixed a thing. Good.
Speaker: What was it? It was that thing. This conversation makes zero sense to you. It has zero context. It doesn't understand how to get there. You need to explain the context. You need to know how to build the context from scratch. And this is where people get the most out of AI is that they can delegate it properly. and they can communicate properly.
Speaker: I think a good analogy that just comes from, I think, your excellent explanation there, Roman, is, you know, these people that are really getting the most, they're they're acting like the the head chef, right? They've got all of their different different chefs who who focus in on different areas of bringing together a plate of food um and they have all been trained um repeatedly on the process of creating the sauce of frying the steak or whatever it might be. um And then it it is all yeah delegated and and concocted together by the the head chef to to then go out onto the onto the past. Maybe it's a nice analogy which could probably pop pop into a a LinkedIn post.
Speaker: Yeah, i understand why agree with you. it's It's really about driving that orchestration and understanding what you want. And and and don't ever think too much about, or somebody runs like whole hundred different agents. You're going to get there. At some point is' going to click and you're like, oh yeah, I can get so much done with it.
Speaker: the way harder is to make sure you consistently use AI for everything you can because it's that skill, that muscle of like, oh, that can be delegated. That is really hard to build if you're like at step zero because you open ChatGPT, you open this chat, for example, like, okay, how do I make this chat make my job? Well, you can't. But eventually you learn how to prompt, eventually you learn how to FCP works, eventually you learn how the plot code works, eventually you come to the point where you can drive sub-agents, you can go agent view and a lot of other fantasy worlds and it's driving amazing outcomes. But as every other skill it takes a lot of time to get there.
Speaker: Excellent. Roman, it's been a pleasure speaking to you. I think we've covered a lot of ground really from about the importance of having that mindset sift from output to to impact. um You do that through partnerships with the business and really understanding that commercial lens behind what is the real real driver and having that pragmatism um over when to move fast, when to slow slow down, um tying that back to to your business outcomes. And um a really interesting point on AI and I suppose how that's affecting the industry, the data roles and
Speaker: Yeah, I suppose how how to it succeed. So thank you so much for talking us through some really, really interesting areas. But before we let you go what would be your biggest sort of takeaways from the episode that, you know, if the listeners can go and and and and take away one thing, one or two things, what is it that that you're really hoping that they would learn?
Speaker: Well, there are two big topics we have discussed, so um I um will allow myself a little bit time and say, on the relationship building, i would say the number one thing for the person is to be able to communicate and and and understand your stakeholders, but also be understood. and we're all human beings, building relationship is important.
Speaker: The thing that most people in analytics keep, but it's actually insanely important for both your career, but also building stakeholders relationship is getting the skill of showcasing what you have done and communicating that. Get into habit having a Slack message once a week into broader communication, showcasing what you did and letting people, inviting people to ask questions. And then also get into habit of having regular just ask me questions kind of conversation. Can you buy like, you know, in the kitchen with a coffee, like asking what's what's on your mind? What's the problem you're trying to solve? Once you have these two things, you will achieve essentially a visibility of what you do. The more visibility you have, the more people understand what you're trying to solve and how you're solving it. And it's really difficult, especially if you're like a lot of analytics, people are introverted by default. And like we we kind of like we solve the problem where we just expect it to be understood. There's no reason why it would be. But also you will get to the point where people will want to ask you questions because you're working on something really interesting.
Speaker: And now on AI point, again, i will reiterate and say, if you are at the point zero, your number one goal is to start using it as much as possible. And the easiest place to use it is to use it for non-work related things. Like literally I started going to AI and say, can you find me a good, I don't know, shower hat that is not the the cheap brand. because Like I wouldn't really care about the outcome, but it would,
Speaker: get me like I don't know 15-20% 30% more than otherwise I would like all the good renovation plan what do I need to consider buying a flat if you're not at zero point then this next thing is the more you share with other people the more you see what people do the the more you can do with AI AI is not one of the very few things in life that is exponential Unless you see somebody's doing something, it's really hard to come up with the idea.
Speaker: ah But what we do internally is like we have people who literally showcase, bragging, oh, I automated this piece of work, or this thing just checks my Slack and creates tickets for me. like well, how did you do that? you start learning.
Speaker: um really helps when you work with other people getting there. And I don't think AI is going to take our jobs. I think it will significantly impact and change what the job actually mean all of us. So it's not really choice whether you get on the AI or not.
Speaker: But if you get there and it really helps, you're going to have a lot of fun on the way and it probably still going be very much needed because together with amazing outcome from AI, AI also creates a lot of wrong things and those are all things need to be cleaned up by somebody eventually.
Speaker: Amazing. Roman, it's been an absolute pleasure talking today. I think, yeah, it's been a really fascinating conversation. And to any of the audience, I'm sure if you've got any further questions or or wanted to pick Roman's brains or or ah speak to him about consulting, I'm i'm sure he'd love to speak. So please feel free to to reach out to Roman. we'll We'll put a link to his LinkedIn in the in the comments. And yeah, I suppose last thing from me, Roman, is just thank you for for joining me.
Speaker: Thank you very much for inviting and also thank you very much for organizing the meetups in London there. Absolute pleasure to be with and I really like your brand in general. It's a pleasure to to to be part of it and be here.
Speaker: Thank you. Amazing, Roman. Well, look, we'll hopefully get you up on the stage also at ah one of the meetups um in in the future. i think you're doing some amazing things at CarWow. So, yeah, im If anyone wants to come along to our ah future events, I'm sure you'll be able to catch Roman at them and we'd love to see you there. So thank you for the shout out, Roman. And yeah, we'll see you in a couple of weeks, and everyone. Thank you and goodbye.
Speaker: Goodbye. Hi, everyone. Just a quick one from me. If you've enjoyed today's episode, I'd be so grateful if you could hit that follow button or leave us a rating. Even better, pass the show on to a friend who might also get some insight from it.
Speaker: It really helps us grow the community and continue to share amazing conversations. I also wanted to take a minute to talk to you about Cognify. Those of you that don't know, Cognify is the leading recruitment partner for modern data teams.
Speaker: We help some of the world's best organizations scale data and drive real value from players that they make. If you're thinking about building a team or making a hire and you're struggling with talent or just want some insights on the market, then I'd love to jump on a call with you and tell you a bit more.
Speaker: Equally, if you're looking for a job and want to find your next dream role, then reach out to myself or any other Cognify team. We'd be happy to see if there's anything on our books that we can help you with and give you general advice on the industry.
Speaker: Finally, big thank you to Omni, this season's sponsor. If you'd like to learn more about the AI analytics that Omni can deliver you, then check out the link in the show notes or come speak to me. i can happily point you in the right direction.
Speaker: Again, thanks for listening and look forward to seeing you a few weeks time.
