 Welcome again, my name is Zhenya and today I have an honor to open up our event with Day2Dream and Decision Making. But before we start, let me just ask you a question. Has anyone of you ever struggled with making decision in your life? I know I have. Thank you. So let's hope that the next 15 minutes will bring us somewhere and help us. And let's start with the terminology. So data-driven decision making, the first part is about data. And data nowadays is everything and everywhere. Just let me show you what happens in the world every 60 seconds. Instagram followers upload more than 46,000 photos. YouTube streams more than four million videos. Uber makes more than 50,000 car rides. And all this information about videos, photos, destinations, likes, dislikes, it then starts somewhere and can be used in the future for decision making. And big companies, they really do it. So like Facebook really relies on data to make a decision like which color the button should be. But not only big companies make decisions, as we have seen each one of us is facing decision making every day. Actually an average individual makes 35,000 decisions per day. And sure, the majority of them are just subconscious decisions, but still at least some of them are decisions we actively think about. And I'm not only talking about some serious stuff like where to invest my money, but also like every day, small stuff, for example, which route should I take to work? Or on Tinder, do I swipe left or right? And all these areas can actually profit from data-driven decision making. Also on our workplace, we all make decisions daily. And if we take a little bit futuristic view on the labor market, at some point maybe data-driven decision making will completely take over and even replace experts in some areas. So maybe some professions would even disappear. And maybe in the future, the artificial intelligence would completely take over. But it's a bit pessimistic thought, so let's go back to our topic. If we take the necessity to decide and back it up with data, we have data-driven decision making. It's an approach that is based on science, more precisely statistics, rather than on intuition or our gut feeling. And to show you how this stuff works, let me use a real-life example that I was working at as a freelance statistician a couple of years ago. I had a client, he was a winemaker and he wanted to expand his business. And he wanted to buy a new wine yard. There were two good offers on the market, A and B. They cost roughly the same, but as it's often is in life, he only had money to buy one. So he came to me with the question, which wine yard should he buy? And based on his professional knowledge, the minerals in the soil, they are really important for wine quality. Sounds logical. And there are lots of minerals, I know, but let's just concentrate on one, on zinc. So to produce good wine, you need to have zinc somewhere between 10 and 20 ppm in your soil. So actually it's easy. Our guy shall just go to wine yard A, check the soil to wine yard B, check the soil decision made. But like wine yard, it's huge. And it's just impossible for one little winemaker to analyze every molecule on every square centimeter of the soil. So he took samples, soil samples. And this is what we statistician call the problem of general population. In our case, a wine yard, it's just so big and so huge that you cannot analyze it. It would take enormous amount of time and enormous amount of money. So you take just a small part of this huge thing. And this is what we call sample. So our guy would be facing this huge wine yard and the red circles that you see there are other samples he took. Then he had the samples analyzed in a lab. And the data that I got for my analysis looked like this. So you see the two columns, one for wine yard A and B. And on the lines you see the zinc values taken on different spots. And from here, like my part of the game began, actually a statistical approach to make decision. And the cool thing about statistics is that no matter what question you ask, you always go through four steps. First formulate your question, then decide on the risk that you are willing to take. Then apply statistical procedure and finally make a decision. Well, data-driven decision. So let's go through these four steps together and start with the first one asking the question. So what our guy wanted to know is which wine yard is better. But with statisticians instead of normal word question, we use a fancy one and we say hypothesis. And we distinguish between two hypothesis, the now hypothesis and the alternative. And it's pretty funny but even though it's called alternative, this is actually what we want to prove. And now hypothesis is what we want to disprove. So in our case, the guy really wanted to know that one wine yard is better because it would mean like they cost the same but if one of them is better for the same money, he will produce better wine. Let's assume he wanted to show that wine yard A is better. And this would be our alternative. Our now hypothesis, what we don't want is to know that they are both the same because it would mean our guy can buy any one of them and should have never invested money in my services to start with. The second step, risk. Statistics is a science where you are never 100% sure. I'm sure you understand this. It's like you can never blindly trust a weather forecast. There is always some risk involved. And risk is the probability that the decision we are making is wrong. Luckily, we can control risk. So it's a good thing. But there is always some tradeoff involved. Because look, if we can control the risk, then cool, let's make it as minimal as possible all the time. But to do so, we need to take more samples, better samples, better data, stuff like that. It's again time and money. So it depends on the situation you are in. If you are inventing a new cancer drug, then I say go ahead. Just minimize your risk. Invest as much time and money as you can. But if you are thinking about which music should I play in my shop to facilitate the sales, then maybe do some quick analysis and accept the fact that it will have a slightly higher risk. So now we go to the third step, the most technical one, applying the suitable statistical method. But don't worry, I'll start lightly and I'll introduce you to this handsome guy. His name is William Gossett and at the beginning of the 20th century, he was a chemist at Guinness lab in Dublin. But not only that, he was also a sort of a byproduct. He was a world famous statistician. He was hired by Guinness to evaluate the quality of the beer. So like he already had sort of a dream job. And to do so, he invented a special statistical procedure that we still use, even in our time. Because of the trade secret of Guinness, he wasn't allowed to use his real name when he published his research. And the guy had a really good sense of humor. So he published a student. And this is what some of you may have heard of as the student T test. Let's have a look at the logic of his testing. And first of all, like beer quality is a huge topic. Like how much alcohol, the color of beer, bitterness, is it waterish? I don't know what else. But let's concentrate on just one thing, beer bitterness. And also because we're in Munich, not in Dublin, let's not use Guinness, let's use a good beer that we all know. And let's say you have a box of our with thinner and you drink one after another. The first beer you drink has bitterness 13. Then you take another beer from the same box, but you see something is a little bit different. Maybe it has bitterness 15. The third bottle maybe has nine and so on and so on. But they all still come from the same box. It's just like every person is different, like every beer bottle is also a little bit different. And this was the huge idea of students. So if we go back the beginning of 20th century, let's do the experiment ourselves. Let's say we buy two boxes of the same beer of a good one, our with thinner. And then just for the sake of science, we also buy one box of Spaten. And we start our experiment. We take one bottle from each our with thinner box. And the bottle from the first box had bitterness 12. The bottle from the second box had bitterness nine. So the difference was three. It's okay. Then we repeat our experiment. The next two bottles, the difference between those two. And we repeat it and repeat and repeat it until two boxes are empty. Then the heart part begins. We need to try this to guys too. So we take one Augustiner bottle, one Spaten bottle. They differ in bitterness. Sure. And we see here that the difference is already high, like it's seven now. Maybe in the next run of our experiment, it will be even line and so on and so on and so on. Next what we do, we use like a number line. You see here from zero to 11. And let's see where our differences length here. So maybe this guy three would land some way here. The difference two would be some way here. The seven would land here and you understand what I mean. And what students noticed is that all the green guys, oops, sorry. All the green guys, they land somewhere here in one space to the left of this dotted line. And all the green guys, they also land clustered somewhere else to the right of this dotted line. So actually he created this dotted line afterwards, just like a border. And it was a huge discovery. So like if you take the same beer that doesn't differ, you are here. If you take beer that differs, you are here. Of course there are some strange guys of strange color, but this is exactly what I was talking about previously when I said that there is some risk involved. And the cool thing that student did for all of us, he really calculated where this border lies. So in this case it was 6.6. Student created a cool tool, a table where depending on your sample size, like amount of bottles you drink and the risk that you are willing to take, you will see where the borderline lies. And it applies to every situation you can face. So we have already seen these guys on the previous slide. And this guy we will see on the next slide because it's slowly time to go back from beer to wine and return to our main example. Our wineyard guy was dealing with soil samples. And from wineyard A imagine this is the soil samples that I got, like amount of zinc. First sample, second sample, third sample, and so on. For wineyard B I got the same. Then I calculated the difference between these two guys, between these two guys, these two guys, and so on and so on. And I put these differences as points on my number line. And what I could see, first of all, this guy 9.9, the borderline, it was like a gift to me from student. I just took it from his table. And I saw that all of my guys, all of my differences, the majority of them, not all, they concentrate here. So like in the difference part. And already here, I could make a decision that cool, there really is a difference between two wineyards. But my client wasn't really interested in all this technical stuff. He just wanted to know what to do. And this is the fourth step. When we translate all of these scientific statistical stuff into normal language. And for him, it was just purchase decision made easy. We could prove with help of science that wineyard A is significantly better than wineyard B. So the client was happy and I was happy. Just, we saw now, very briefly, how the data-driven decision making work. Let me sum up the most important points. Data-driven decision making facilitates making decisions through statistics. There are some things that you need to consider, like sampling is important. Of course, your data is important and bear in mind that there is some risk involved. And this approach is gaining more and more attention and is now more widely used in areas that we could never thought about before, like Tinder. So maybe at some point it will really change our lives. And I don't want to think that some professions will die out or anything pessimistic. But I'm sure that it could really change our world and also our label market. So that was all from me. Thank you very much. We will have a few minutes for some questions. Don't throw anywhere. We would like to ask a question to Jenny about her talk. There's a mic. So my question would be, do you change the hypothesis along the way? For example, when we looked at the ZIN values in the soil, the hypothesis was that which vineyard was better. But if we define the sentence in a better way, we can say which vineyard is better for my business, like for your employer, for the one who wanted to buy the vineyard. So after looking at the values, we can see that in my opinion, is it working? Yeah, I can hear it. So after looking at both values, in my opinion, both were bad values. So would you think that you can go back to the hypothesis and say that none of them was better to be both? So my hypothesis should be changed? Sure. It's really what happens. Like sometimes you collect your data, you want to make some cool decision, and then you see, actually, it's much worse than I thought. Sure, it happens. And this way, would it be some hypothesis of your main hypothesis or... No, I would say you just, like these four steps that I outlined, you just go this again. So it's like this vicious circle. And maybe you go there, you go there, you go there until at some point, you reach a conclusion that makes sense both from a statistical standpoint and also from your standpoint, like a businessman. Thank you. Maybe just another small question if anyone wants to go? Okay, I guess not. So then round of applause for Yenya. Thank you so much for...