Transcription
The first report after the break is from Dima Kazakov. He will talk about the profession of a Data Analyst.
Hi everyone, guys. Testing, testing. Seems like everything is working. I really love speaking at JinnDay because, as a rule, it has the most active audience. Prepare your questions. Vadim said it correctly, today I will talk about the profession of Data Science, with a greater emphasis on Data Analytics. I will talk about prospects, what skills are needed, and of course, examples from our practice. That is, how we work. Yes, Dim, can we ask people to be a little quieter? If they are talking too much amongst themselves, because it's a bit noisy here, and those who are listening might not hear well enough. Okay, thank you very much. For those who don't know me, probably many don't. My name is Kazakov Dmitry, I am the Director of Analytics at Kolesa Group. I have over 10 years of experience in data analytics, fintech, consulting, marketing, and product analytics. And I think my knowledge and experience will hopefully be useful to you in some way. I would like to share useful information with you today. So, yes, what will be the program today? We will have three main parts. First of all, I will generally explain what Data Science is, what it consists of, so that we all understand this field similarly. Also, I will explain how this field is applied in our company, that is, what the daily life of data scientists and data analysts is like here. And I would like to tell you about the portrait of a good data analyst, a data scientist, so that you understand in which direction you need to move. Now, a small interactive session, guys. Generally, raise your hands, who knows what Data Science is, at least generally? Great, great. And now, raise your hands, who wants to connect their career in the future, that is, wants to try to enter this direction? All great! I am very glad that there are quite a few of you, so my information will be useful to someone. So, let's go. The first block is Data Science, what is it? I will try to explain it simply, with examples. There are many definitions of Data Science, actually. There isn't any single correct definition, actually. Different specialists call it differently. Therefore, I personally like this rather general definition: it is the science of methods for collecting, processing, analyzing, and modeling data. So, all stages of working with data are covered here. And what is most important, Data Science studies this in order to apply it in practice, to find some useful insights. Here are two key insights. You should have two theses: Data Science is about various stages of the data lifecycle and about practical application for some benefit. Very often, when people try to explain what Data Science is, analysts draw these pie charts because Data Science is a certain mix of skills. That is, you can't say that you just studied math and that's it, or that you just studied Python and that's it. I am a data scientist. Actually, a data scientist is, first of all, a combination of three groups of skills: Computer Science, first, that is, knowledge of programming languages, knowledge of how databases work, knowledge of how, for example, backend and frontend work, because all of this is needed in the work of data analysts, data scientists, data engineers. Secondly, these are business skills, that is, so-called business sense or understanding of the business part. That is, an analyst cannot just know Python well, or a data scientist cannot just know Python and that's it. They need to understand very well how the product works, what metrics exist, where we are moving, what business problems there are, what KPIs we have, what the strategy is. So, all of this is important knowledge without which a data scientist or data analyst cannot work. And thirdly, of course, a large group of skills is knowledge of mathematics. In some sub-segments of data-related professions, mathematics is needed less, in some more. But in general, without mathematics, you will always hit a glass ceiling. By mathematics, I mean, first of all, mathematical statistics, probability theory, linear algebra, solving optimization problems. These are probably the key ones that are needed. You need to know them. And traditionally, Data Scientist is called the intersection of all three. But in reality, there are quite a few professions that also fall under this umbrella term Data Science. For example, at the intersection of Computer Science and the subject area, there is traditional development or, for example, Data Engineering. So, these guys don't need to know, for example, statistics, A/B tests, or regression analysis. They need to know how to code well, maintain databases, build data warehouses. Also, somewhere here are BI analysts. Guys who build dashboards and reporting systems using BI tools. They also often don't need to know mathematics in depth, but they should be proficient in, for example, working with databases for tools. These guys are also located somewhere here, closer to the center. The next point is Product Analysts, another type of data-related analysts. These guys are much less, or let's say, further from Computer Science, from programming, from databases. Often, they work with ready-made tools like Amplitude, Google Analytics, Yandex.Metrica, and so on. So, they sometimes don't even know how to code, but they understand the business side very well, build funnels of actions, analyze KPIs, product pain points, user pain points. These are the guys who are located here, closer to the center, to our desired center where everyone wants to get. Data Analysts are guys much closer to Computer Science. That is, they already know how to jump, how to work with databases. At the same time, they also know mathematics well and are well-versed in the business context. And here's a conditional center with an asterisk, which is conventionally called Data Scientist. Here are guys who are good in almost all skill groups. As a rule, they already know Python well, they know databases well, they know the business side, mathematics. A/B tests, experiments. These are universal full-stack guys. For example, there is also such a profession that many guys strive for now, ML engineers, because it's hyped. These guys are often further from the business side. They don't always understand, for example, how a user behaves, or why we have specific goals in the team, what our KPI strategy is. So, ML engineers are more towards the engineering side. But as you can see, there are quite a few professions. And from myself, I would like to say that it is very important for you at this stage, when you are studying, when you are still a student, to determine what you like more, because becoming a universal person who knows engineering, data engineering, and product analytics is not possible. You must understand well that, for example, if you want to communicate more with the business, test hypotheses, find growth points, help the business, then you should go into BI analytics or product analytics. If you are closer to engineering, building complex data pipelines, then it's either data engineering or ML engineering. Of course, if you like a little bit of everything, that is, you like to play with data sometimes, sometimes communicate with the business, then try to get into data analytics or data science. This is such a core, such an intersection. Okay, I said that Data Science is a field that covers all stages of the data lifecycle. I will now try to explain what stages there are, generally, what traditional stages there are, and what professions appear at these stages. Traditionally, any [music] data lifecycle begins with primary collection. For example, we have the Krysha product, and we sell or rent, help rent apartments. And for something to start collecting, we log various data, for example, about the announcement, photos, price, or about user behavior. That is, logging clicks on a button, seeing a certain screen, interacting in some way. All of this is logged. So, we have a rather large volume of data, about 200-300 GB of data per day, just from events. And approximately the same amount comes from the backend. After we have collected it initially, we need to build some storage system for it. It's not enough to just put it in logs and let it lie there like garbage. We need to build a very good data warehouse system, where to store it in a convenient form so that everyone can work with it. After that, the first stage of working with data begins. That is, if you just collect data and just store it, you are not using the data at all yet. It is very important to start working with it. And here, the first stage of preprocessing begins, some ETLs are built, for example, second-level aggregations, some data enrichment. These are already the first stages of analysis, you could say. After that, there is a division, so to speak. So, there is one stream of working with data, rather analytics. That is, descriptive things, finding some relationships, insights, and modeling. This story is more about ML. That is, sometimes we use data to, for example, predict whether this announcement on Krysha is fraudulent. We have such systems, we constantly fight with fraudsters. They use data to tell whether this person or this announcement should be blocked as fraudulent. After that, an important stage is communication. That is, I said that Data Science is about application to extract some benefit from data. Here, the extraction stage begins. We can simply tell the manager about the analysis, tell them that we have such a level of fraud, the dynamics are falling or growing. This will already help make some decisions. Or, we can also say that we have developed an ML model, and we need to tell the business about it, sell it, so to speak, pitch it. After this stage, as a rule, the application stage follows. That is, after we discuss all the nuances with the business, we, for example, roll something out to production or apply a certain insight. For example, after testing, we say, no, we need to roll it out to everyone. This is the application of data. And the last important stage, which, unfortunately, is often forgotten, is to conduct an analysis of the changes. That is, it's not enough to just apply and forget. You also need to conduct an analysis of what it gave us, for example, A/B tests or something else. Here is a diagram of the approximate lifecycle of any data in a company. How are these data professions superimposed here? So, how does this happen? The first three are, as a rule, the territory of data engineers. That is, guys who build streaming systems, collection, processing, storage of data, their construction, data marts of the second level, when we aggregate something. This is traditionally the territory of data engineers. Below is a small, narrow section, this is where ML engineers are engaged. As you can see, they cover not so many stages. Often, they just take some data, train models, make predictions, tell the manager how it works, and deploy to production. But often, even deployment is done by developers. So, know that if you think you want to connect your career specifically with ML engineering, remember that ML engineers are a very niche profession. They often occupy a small piece. But, for example, data analysts or analysts of one kind or another, on the contrary, are guys who cover a very large piece. And therefore, their demand is always higher. That is, they are usually needed more in teams because they start even in the territory of data engineers, helping to collect some tables, data marts. Moreover, they are involved in their enrichment, writing some SQL, analysis is always on their shoulders. Communication with the business is on their shoulders. Helping backend, for example, to apply it, to implement, say, A/B tests or something else, and moreover, to calculate it using tests or new experiments. And data scientists, I remind you, are central. They are somewhere in the middle. They sometimes help deploy models to production. Sometimes they just work in the analytics territory. This is a very simplified scheme, but so that you just understand who does what and think about which role would be interesting to you in the future, what you would like to do. After this scheme, well, logically, I would probably talk about how it works in our company. Well, I will actually do that. It is very important to say that we, well, a feature of our team is that we grow data analysts as full-stack. So, there are two directions: there are full-stack analysts, and there are very specialized ones. In our company, we have full-stack. What does this mean? It means that our analysts are essentially data scientists themselves. They can do ML, they can make dashboards, they can brainstorm with managers, generate hypotheses, they can conduct A/B tests, they can conduct ad-hoc research. So, the skill of being full-stack is very important. So, guys who are thinking about how to join us, know that we need a set of skills, various ones. It's not enough, for example, to just know SQL well with databases to join us. So, what do guys traditionally do? I will now list some, not all, but the most frequent types of work that are done in our company. The first is ad-hoc research. What is this? This is research on demand. A manager comes, for example, a Krysha manager, and says, "Damn, the volume of complaints about fraudsters or about irrelevant prices, for example, in rentals, has increased." So, we just see some symptom, and we need to analyze why this is happening, what happened. The analyst takes, pulls up various databases, generates hypotheses, rechecks them using various analysis methods, like searching on a small relational database or something else, and after some time, after a week, for example, brings and says, "I've studied everything, I have answers to your questions." This is ad-hoc research. This is a very frequent case in the daily practice of our guys. The second frequent task is working with data engineers when we help them build data marts, storage systems, data warehouses, so that access to data is more natural, so that all necessary data is stored, so that it is fast. So, our analysts are also actively involved in this. A very important stage, specifically in our team, or generally, types of tasks are experiments. So, unfortunately, I often see at interviews and generally that many analysts don't know this point of tasks, although for me it's a bit wild. So, experiments are the gold standard for answering the manager's question: did a feature work or not? Did we do something that helped us solve the problem or not? So, if you don't have experiments, you can't truly give an honest answer to the manager. And so, our guys actively use experiments, specifically classic A/B tests. Here the picture is more about A/B tests, but also methods like advanced Cohort analysis, methods for accelerating A/B tests, PSM, and many others. So, our guys spend about 20% of their time on this. Here, by the way, is a QR code. I don't know, you can probably screenshot it. It's specifically about our experimentation system, our experimentation culture in our team. So, this is a video from Kolesa Conf. I think it will be useful to read, to watch, rather. Okay, I hope those who wanted to open it have opened it, and we are moving on. BI tools. As I said, our analysts often have to build reporting systems. So, sometimes it's not necessary to conduct complex analysis for a week to quickly see an answer, for example, about the dynamics of something, that the funnel is falling. For this, it's not necessary to overload an analyst. It's enough to create a reporting system, a visual representation of key metrics, answers to questions. And our guys also actively do this. For example, in our stack, it's Power BI. Here, by the way, is also a QR code for free Power BI courses from Microsoft itself. So, in my opinion, Power BI is quite easy to learn. So, in a few weeks, you can improve quite well. Testing, testing. Yes, for those interested, you can try to switch. We are moving on. What else do our guys do? Here we enter the territory of ML. So, very often managers come with the question: what will happen next? So, for example, next month, or what will happen next with this user? So, this is already the territory of some forecasting, prediction. So, traditionally, at the end, somewhere at the end of October, November, we have such a session of forecasting our KPIs for the next year. So, guys take the dynamics for several years, apply time series methods, Prophet, Neural Prophet, elasticity, all sorts of things to understand what will happen with our key metrics, for example, revenue, user base volume, traffic, in the next year. This is a very good support for planning the team's development for the next year. For managers, they see that in this month we are declining, we need to do something more actively. And in this month, we are growing very actively, we need to accelerate even more. This is a very important tool and a very important task, forecasting. One of the types of tasks that analysts often do is segmentation. So, for example, there is the Kolesa.kz product, which is the sale and purchase of cars, commercial vehicles. And it's impossible to create an application that is for all types of users. So, we have guys who sell their car for the first time, these are simple sellers. There are professional sellers who consistently sell 5, 10, 20 cars per week, for example. And they need completely different ways of interacting with our product. So, analysts are the people who find that there are differences in such groups and suggest how to work with them. So, the task of segmentation also arises quite often in our company. A separate type of task that I would also like to talk about is not some technical skill, but how you work with managers and the business. So, brainstorming and active work with your team is also something that distinguishes our guys. I would highlight this separately. Here are some photos of how we work. For example, here the Kolesa team was discussing research, brainstorming on how we will build a particular feature. Unfortunately, I don't remember what we were discussing here, possibly spare parts or the spare parts marketplace, I don't know. Also, an example of how UX researchers present some research on how our users behave, where they have more problems. The whole team is gathered, listening, thinking about how to make life better. Here, we analysts probably gathered, discussing how to manipulate numbers to achieve the desired result. Actually, no, we don't do that, of course. We were probably just discussing some complex methodology. In essence, I have described the approximate most frequent tasks. Of course, we have more specific ones, but this is probably the main volume. We are moving to the third part. I will try to tell you what a good analyst is. So, what should you strive for, what should you learn to, for example, join us or, you know, go abroad somewhere, what is the standard of a good analyst, what do you need to know and be able to do? So, let's go. First of all, I will say that a good analyst is also not just hard skills. Hard skills are an important part, but besides that, there are also business skills and soft skills. That is, without them, you are unlikely to become a good analyst or data scientist. At the same time, I want to say that still, about 60% of success in analytics, in Data Science, is in the area of hard skills, technical ones, because the profession is engineering, it's not a humanitarian profession. You need to know a lot, be able to do complex things with your hands. But about 40% of success lies in the area of soft skills. There is also a small interesting study here, you can also make a QR code, which talks about the importance of soft skills. And according to this study, almost 85% of success factors lie precisely in the area of soft skills, and only 15% in hard skills. But this is probably not just for data analytics, but in general. Let's go. What hard skills, for example, do analyst guys, data scientists need? First of all, programming skills. That is, without programming skills, you cannot perform advanced analysis methods, for example, A/B tests, ML, forecasting, segmentation, access to complex databases. Without Python or R skills, you won't be able to. But I recommend Python. R is a rather specific tool. Working with databases, SQL. That is, without SQL, you definitely won't be able to enter analytics. If you are generally thinking about where to start entering data analytics, learn Python and SQL. That is, this is the golden rule. Start with SQL, then Python. This is a must-have. Thirdly, a third important skill, also rather a must-have, is mathematics. People often say, "Oh, I'm without math. I know how to write SELECT, I know how to build a dashboard, no problem." All of this is enough. Actually, no. That is, often the answer to the question of what worked or what the problem is lies through the application of rather complex mathematical methods, multifactorial analysis, correlation, regression analysis, hypothesis testing. Without mathematics, you simply cannot correctly and honestly answer most business questions. So, be sure to study statistics, probability theory. It will help you a lot. Then comes a skill that is half, well, slightly less important, let's say, experiments. Why? Because often when you join a company, there isn't a culture of experiments right away. That is, you might not need it right away. Secondly, you might go into an area where experiments are often applied, for example, data analysts use them less often than product analysts. Then comes Data Science, Machine Learning. This is already in the area of nice-to-have technical skills. In general, guys who know machine learning develop faster at some point. For example, if I know time series, I can handle more complex tasks. If I know how to solve clustering problems, I can also do more complex tasks. But in general, this is also rather a nice-to-have. Also, the skill of building dashboards, Power BI. It is also in demand, but it can be learned in a few weeks, so it is perhaps slightly less critical. These are the hard skills. We are moving on. There isn't much time left. Soft skills. So, what do you need to know in the area of soft skills? That is, working with people, some internal qualities. First of all, critical thinking. What is this? This is the skill to consciously understand the problem, consciously and objectively perceive information, ideas, problems, see manipulations in data or perceptual errors. Why is this important for analysts? Because an analyst in teams is the person who asks questions, questions everything, proposes non-standard solutions, often argues. So, it's a rather annoying type, but these are the costs of the profession. Therefore, without critical thinking skills, you will also hit a ceiling. You cannot always analyze a task or hypothesis. In addition to critical thinking, an important skill is systems thinking. That is, seeing cause-and-effect relationships or building systems in some chaos of scattered information, finding main and secondary elements, building hierarchies. So, a good analyst thinks systemically very well. So, if you have any problems with this, improve them. There are courses and books on systems thinking. This is very important. Next, an important skill is effective communication. This is not only for analysts, but also for analysts, because analytics, although it is an engineering profession, you often interact with the business, prove, present, try to explain complex things in simple language, or vice versa, translate a business request into, well, into the language of mathematics or computer science. So, guys with good effective communication grow much faster in their careers. The next important point is product sense. That is, understanding how your product works or how your team works, how KPIs are set, where your users have problems, what value we provide to users. So, without product sense, without business thinking, or without customer centricity, it will be difficult in professions like product analyst, data analyst. For example, data engineers might need less of this, but for guys working with the business, it's much more necessary. And the last important skill is learnability. In general, our profession is formed on the fly. There isn't really any golden standard. Every day, new methodologies are invented. For example, Cohort analysis methods are becoming very popular now, which, for example, no one knew about 5 years ago, or accelerating tests. So, about 5 years ago, no one really did this. Now it's forming, and you need to constantly learn, read, attend conferences, study actively, network. So, without learnability, that is, the ability to absorb this, it will be difficult. These are the most important, but not all, but the most important soft skills for data analysts. And guys, the last point is business skills. I will try to be faster here. This is the first important skill: Project Management. That is, you can be very good at dealing with people, communicating, knowing technical things well, but for example, you manage tasks poorly, you always have problems with deadlines, you don't know how to decompose, everything is postponed. You have poor project management. So, you will always be lagging behind because of this. Learn to build your tasks as projects, with deadlines, requirements, risk assessments, decomposition, and all of this that managers love. Secondly, an important skill in business is presentation skills, storytelling. This is more important specifically for data analysts because we often have to explain complex things in simple language. For example, to apply some advanced deep learning or time series analysis to simply tell the business what will happen next. This is not so easy to explain to guys who are far from mathematics. So, I often see, for example, beginner analysts who come, do a complex analysis, create a table with 50 columns and 100 rows, say something with smart words, and managers understand nothing. As a result, the application of these insights is reduced to zero. So, it is very important to be able to tell stories with data. The next important skill is negotiation skills. This comes from critical thinking, as I said. An analyst in a team is the person who often questions. They must be able to argue convincingly, debate. So, negotiation skills are very important, especially for senior level guys and even more so for leads. And an important skill is product vision, user, well, customer centricity. Because, well, I've also said that without this, unfortunately, you cannot be a good product analyst, test hypotheses, and so on. And the last business skill, without which, unfortunately, you cannot, is time management. So, this is also somewhere close to the project aspect. That is, you must be able to meet deadlines. So, for example, in our company, there is a direct value: punctuality. That is, we highly value guys who can meet deadlines. So, without this, deadlines can be extended, projects can disappoint the business. Without this, unfortunately, our company definitely cannot. Essentially, these are also some business skills that are important, and we have reached the question section. Guys, let's raise your hands.