0:00:09.746,0:00:13.009 Okay, decisions in the age[br]of digitalization. 0:00:13.619,0:00:16.957 To make one thing very clear[br]at the very beginning: 0:00:16.957,0:00:21.018 Most important human decisions[br]can and will never be automated. 0:00:21.018,0:00:24.876 They are done by gut feeling,[br]under extreme uncertainty. 0:00:25.146,0:00:27.957 I'm not going to talk about these. 0:00:28.557,0:00:31.171 Don't blame me for that. 0:00:31.451,0:00:35.549 0% automation on decisions like, 0:00:35.549,0:00:39.118 these are examples from my life:[br]marrying my wife, 0:00:39.118,0:00:42.132 declining a tax-free,[br]permanent CERN job offer, 0:00:42.132,0:00:45.122 instead of a KIT professorship, 0:00:45.125,0:00:46.843 or founding Blue Yonder. 0:00:46.843,0:00:49.881 In retrospect, these were[br]all very good decisions, 0:00:49.881,0:00:52.169 but they weren't clear beforehand. 0:00:52.169,0:00:54.515 So I'm not talking about these things. 0:00:54.785,0:01:01.057 But I claim that 99% of all operational[br]decisions in enterprises 0:01:01.060,0:01:02.813 can be automated. 0:01:02.822,0:01:04.729 And they will be automated. 0:01:05.055,0:01:09.264 They will also be simultaneously improved[br]by quite some margin. 0:01:10.566,0:01:13.273 That is what I am going to report on. 0:01:13.605,0:01:18.474 And these are especially repeated,[br]regularly repeated, similar decisions. 0:01:19.174,0:01:22.095 Let's take an example from retail. 0:01:22.975,0:01:26.222 Assume you are a store manager[br]and you have to decide: 0:01:26.222,0:01:28.816 Of these articles,[br]how many do I have to order 0:01:28.823,0:01:30.804 so that I have enough tomorrow, 0:01:30.814,0:01:34.525 but not so many[br]that I have to throw them away 0:01:34.532,0:01:36.050 once the shelf life is passed? 0:01:36.063,0:01:40.055 Ten? Zero? 100? 0:01:40.055,0:01:44.549 That's a decision many people[br]have to make every day, again and again. 0:01:44.549,0:01:46.214 Or, again, you are store manager: 0:01:46.214,0:01:49.954 Do you want to decrease[br]the price of this product today? Yes? 0:01:50.214,0:01:53.511 By 5%? By 10%? By 20%? 0:01:53.511,0:01:56.787 That's a decision that people[br]have to make very often. 0:01:56.787,0:02:01.005 Or to send this expensive catalog [br]to that customer or to that customer? 0:02:01.005,0:02:03.672 Is it worthwhile? Will it be good for him? 0:02:03.672,0:02:06.086 Will it be good for us? 0:02:06.086,0:02:08.676 These are questions I'm talking about. 0:02:09.727,0:02:12.740 Don't you believe the 99%? 0:02:13.850,0:02:19.509 First, you have to know how most[br]common decisions are taken in real life. 0:02:19.590,0:02:21.390 And that's about something like that, 0:02:21.390,0:02:28.366 so 90% of these decisions are:[br]Do nothing, or do what we always do. 0:02:28.518,0:02:30.039 What people always have done, 0:02:30.044,0:02:32.050 even before I came here 0:02:32.053,0:02:38.895 and wanted to bring into this business[br]everything I have learned at university. 0:02:38.895,0:02:43.296 First, you are told, "No, we do it[br]this way - ever, always." 0:02:44.216,0:02:48.534 About 9% apply business rules[br]in one way or another. 0:02:48.534,0:02:52.187 But many of these business rules,[br]if you really look at them, 0:02:52.187,0:02:53.345 are not very good. 0:02:53.345,0:02:58.367 Almost none of them[br]has a real proof of value. 0:02:59.147,0:03:03.119 [For] only about 1% of decisions, 0:03:04.149,0:03:08.428 [does] somebody sit down,[br]use his brain, and think about it. 0:03:09.338,0:03:13.196 And even this 1% is far from optimal. 0:03:13.196,0:03:14.664 For this, you have to know 0:03:14.664,0:03:17.650 that the human decision-making[br]system has two systems. 0:03:18.420,0:03:21.630 The so-called system one[br]is fast and intuitive, 0:03:21.630,0:03:23.820 but it has many biases. 0:03:25.100,0:03:27.858 System number two is slow and rational, 0:03:27.858,0:03:29.379 but it is hardly used. 0:03:29.379,0:03:33.281 Why? Because it's work, it takes[br]a lot of energy to use it. 0:03:33.661,0:03:39.130 Most of the time we don't use it,[br]we use system one rules for everyday life. 0:03:39.730,0:03:46.329 And even, and especially, scientists[br]and experts use their "gut feeling." 0:03:46.594,0:03:50.916 They are not always thinking[br]and deciding rationally. 0:03:52.067,0:03:53.729 And it's very important to know, 0:03:53.732,0:03:56.973 system one, the fast one,[br]cannot speak statistics. 0:03:57.546,0:04:04.245 It cannot really judge risk and chance. 0:04:04.695,0:04:06.437 That's a big problem. 0:04:07.217,0:04:09.572 Daniel Kahneman, Nobel Prize winner, 0:04:11.512,0:04:15.953 has done research his whole life[br]about how we think. 0:04:17.333,0:04:21.454 His book, "Thinking Fast and Slow"[br]is really worth reading. 0:04:24.094,0:04:26.216 He describes all of this. 0:04:26.216,0:04:30.094 He describes how our system one[br]makes decisions, 0:04:30.094,0:04:32.995 and it is all explained by evolution. 0:04:32.995,0:04:37.333 So it was right: we are here now[br]because we did it that way. 0:04:37.513,0:04:42.984 But there are many, many effects,[br]for example, the IKEA effect: 0:04:42.997,0:04:45.453 We value something higher 0:04:45.462,0:04:49.199 when we have worked[br]for it ourselves a bit. 0:04:49.216,0:04:53.244 There are many[br]so called "cognitive biases." 0:04:53.244,0:04:56.409 And it's very funny to see 0:04:56.425,0:04:59.903 that somebody, like myself,[br]who thinks he is very rational, 0:04:59.916,0:05:03.198 has all these biases every day. 0:05:03.578,0:05:07.330 So, it's like that,[br]we can't do anything against it. 0:05:07.330,0:05:08.816 So, our question was: 0:05:08.816,0:05:12.484 How can we make more rational decisions[br]than we are actually doing? 0:05:12.484,0:05:16.124 The origin of this idea[br]comes from CERN and KIT, 0:05:16.124,0:05:19.024 so from where I do my research work. 0:05:19.444,0:05:22.484 The development is now done[br]at Blue Yonder. 0:05:22.484,0:05:26.536 The basis is really big data,[br]Bayesian statistics, machine learning, 0:05:26.536,0:05:30.085 data science, stochastic programming,[br]causality reconstruction. 0:05:30.085,0:05:32.285 In other words, it's really science, 0:05:32.285,0:05:36.724 how scientists work[br]in order to get insights from data, 0:05:37.134,0:05:43.479 and then optimize decisions on the basis[br]of these insights for the future. 0:05:44.009,0:05:46.896 It's a purely scientific[br]thinking and ethos, 0:05:46.896,0:05:50.785 that is also not true[br]in most enterprises, as I have learned. 0:05:51.465,0:05:53.657 We are trying to do[br]this technology transfer 0:05:53.657,0:05:57.205 to reinvent business processes[br]with scientific methods. 0:05:57.585,0:05:58.727 This is a key. 0:05:58.727,0:06:01.836 The key to better decisions[br]in the digitalization era is: 0:06:01.836,0:06:04.146 data and algorithms. 0:06:04.406,0:06:08.119 So, first, you need[br]relevant objective data. 0:06:08.119,0:06:11.111 So, it's mostly your data[br]or your company's data. 0:06:11.301,0:06:15.226 And you need good, [br]mathematically correct predictions 0:06:15.226,0:06:20.156 for near future events on the basis[br]of this data, including uncertainties. 0:06:20.156,0:06:21.966 That's a very important point. 0:06:21.966,0:06:26.286 And this is handled by an area[br]that is called "predictive analytics." 0:06:26.756,0:06:31.417 Then you also want to know, if you decide[br]this or that, what does it mean? 0:06:31.417,0:06:34.447 So you need cost functions,[br]utility functions: 0:06:34.447,0:06:37.380 What is the consequence[br]of this or that decision? 0:06:38.900,0:06:42.320 And then you want[br]to optimize this decision 0:06:43.040,0:06:46.787 using the prediction[br]and the utility function 0:06:46.787,0:06:48.878 to give a recipe for what you should do. 0:06:48.878,0:06:51.172 That is called "prescriptive analytics." 0:06:51.172,0:06:53.537 And finally, the King's discipline: 0:06:53.537,0:06:56.999 If you [have got this] far,[br]then you can automate it, 0:06:56.999,0:07:01.439 then it's completely automated. 0:07:01.439,0:07:03.968 When can we use predictive analytics? 0:07:05.878,0:07:10.170 In between purely random[br]processes, like the lottery, 0:07:10.800,0:07:15.798 and completely deterministic processes,[br]like a mechanical pendulum. 0:07:16.078,0:07:20.011 Most processes that we are interested in[br]are in the continuum between those. 0:07:20.021,0:07:21.623 So they are partly predictable, 0:07:21.645,0:07:24.542 but there's a lot of randomness[br]and uncertainty inside. 0:07:24.542,0:07:28.652 Most people think that the world[br]is very deterministic. That's not true. 0:07:28.816,0:07:31.223 There is some determinism inside, 0:07:31.230,0:07:34.300 and it's our work[br]to get this component out. 0:07:34.329,0:07:38.073 But there is always lots[br]of uncertainty inside, 0:07:38.090,0:07:42.683 and in evaluating the value of a decision, 0:07:42.683,0:07:45.122 we also need to take[br]into account the uncertainty. 0:07:45.122,0:07:46.717 That's a very important point. 0:07:46.717,0:07:49.716 So, that is what[br]predictive analytics can do. 0:07:50.036,0:07:54.021 Going back to our store manager[br]and his problem: 0:07:54.021,0:07:56.831 How many articles to order? 0:07:56.831,0:07:59.161 There may be many influencing factors 0:07:59.161,0:08:02.532 on how many articles[br]will be sold tomorrow, 0:08:02.532,0:08:05.741 from weather, weather forecast,[br]price, promotion, 0:08:06.811,0:08:10.632 day of the week, holidays,[br]seasonality, and so on. 0:08:10.694,0:08:13.791 But the store manager has[br]to make his decision based on, 0:08:13.812,0:08:17.483 or take his information from,[br]many individual events: 0:08:17.493,0:08:21.815 namely many articles, many locations,[br]many points in time. 0:08:22.325,0:08:26.655 And all together, there are many[br]such decisions to take. 0:08:26.949,0:08:28.830 The output of predictive analytics 0:08:28.848,0:08:32.265 should be something[br]which is not only a value - 0:08:32.279,0:08:34.611 it will never be 80 pieces, 0:08:34.625,0:08:37.679 because we don't know whether[br]80 pieces will be sold tomorrow - 0:08:37.696,0:08:42.092 but a distribution[br]of all possible futures: 0:08:42.119,0:08:45.094 79, 80, 81, 82, and so on. 0:08:45.106,0:08:48.738 The output is a probability[br]for each of the possible futures, 0:08:48.738,0:08:50.496 a probability distribution. 0:08:50.496,0:08:53.256 If you have that for each[br]of the single cases, 0:08:53.256,0:08:56.518 then you can calculate[br]what the value of a decision is. 0:08:56.518,0:08:58.358 That's prescriptive analytics. 0:08:58.358,0:09:01.867 Use the predictions to take decisions[br]on the basis of utility functions, 0:09:01.867,0:09:04.597 which are somehow known[br]or have to be found. 0:09:08.257,0:09:12.117 That is still a management decision:[br]What is a value of this or that? 0:09:14.857,0:09:20.726 We have to decide what is a value,[br]what we want to optimize. 0:09:20.726,0:09:25.259 Prescriptive applications give recipes,[br]prescriptions, for every single decision. 0:09:26.489,0:09:31.331 The recipe might be:[br]Take alternative two, buy 38 tomorrow. 0:09:32.601,0:09:34.432 For some very fine tasks, 0:09:34.441,0:09:36.739 the mathematical and technical complexity 0:09:36.743,0:09:39.192 of the whole process,[br]of prediction and decision, 0:09:39.192,0:09:41.724 can already be outsourced completely, 0:09:41.724,0:09:48.052 and you can buy it in some cases[br]as a solution available in the Cloud. 0:09:48.702,0:09:54.042 With predictive analytics, we very often[br]can attain two mutually exclusive goals. 0:09:54.042,0:09:55.581 That's very interesting, 0:09:55.583,0:09:58.469 because usually in management decisions,[br]you cannot do it. 0:09:58.474,0:10:01.552 For example, in purchase[br]order optimization, 0:10:01.552,0:10:06.107 you can have an increased availability[br]during the season, 0:10:06.107,0:10:10.343 and simultaneously reduce what remains[br]at the end of the season, 0:10:10.740,0:10:12.501 for example, in fashion retail. 0:10:12.504,0:10:17.686 Or you can reduce the out-of-stock rate[br]in your meat shelves, 0:10:17.686,0:10:21.456 but simultaneously,[br]you can reduce the meat waste 0:10:21.456,0:10:23.597 at the end of the shelf lifetime. 0:10:24.097,0:10:28.537 Or you can save 35%[br]of your marketing budget, 0:10:28.587,0:10:32.379 but still have more customers[br]to buy your products. 0:10:33.229,0:10:38.503 These are examples that are possible[br]with such technologies. 0:10:38.503,0:10:42.982 We have also automated[br]the work of scientists, 0:10:42.982,0:10:47.553 namely of my own science[br]as a particle physicist. 0:10:47.640,0:10:52.391 So we have seen that[br]usually a PhD student or researcher 0:10:52.394,0:10:54.904 has to do very similar things, 0:10:54.911,0:10:59.373 and once you understand it,[br]you can try to optimize all this. 0:10:59.393,0:11:02.935 We have found that usually[br]a PhD thesis on particle physics 0:11:02.935,0:11:05.746 can be done by 72 decisions, 0:11:05.786,0:11:08.824 and they all can be optimized. 0:11:09.294,0:11:14.345 This was done for one[br]of the most successful experiments 0:11:14.345,0:11:19.103 in experimental particle physics,[br]the Belle experiment in Japan. 0:11:19.423,0:11:22.961 All the data having been[br]collected over ten years 0:11:22.961,0:11:25.374 have been reprocessed with such a program, 0:11:25.374,0:11:27.664 and the output of it was twice as good 0:11:27.664,0:11:32.944 as 400 physicists have done,[br]by hand together, in ten years. 0:11:33.714,0:11:37.684 That was the work of machine learning[br]or artificial intelligence, 0:11:37.684,0:11:43.013 and three PhD students corresponds[br]to about 500 normal PhD theses, 0:11:43.013,0:11:46.065 and corresponds to another[br]ten years of data taking, 0:11:46.065,0:11:49.188 which would have cost 700 million euros. 0:11:50.618,0:11:54.923 Very often we have too many decisions[br]to be made by a human. 0:11:55.073,0:11:58.435 For example, 130,000 articles,[br]five productizes, 0:11:58.435,0:12:03.447 three storage locations, two predictions[br]40 days in advance, and so on to do, 0:12:03.458,0:12:07.535 and immediately you have some[br]100 million decisions. 0:12:07.550,0:12:11.189 so it's always good then to automate. 0:12:11.197,0:12:14.062 Because even if you have[br]100 million recipes, 0:12:14.065,0:12:17.018 who's going to read all the recipes[br]and then decide what to do? 0:12:17.158,0:12:22.429 That is where we see that automation[br]is the ultimate goal of what we want. 0:12:22.759,0:12:26.614 The impact of decision automation [br]can, for example, be seen here: 0:12:26.614,0:12:30.289 So, this is the out-of-stock rate[br]at a German supermarket chain 0:12:30.289,0:12:33.131 at constant overall stock level 0:12:33.131,0:12:34.561 as a function of time. 0:12:34.561,0:12:39.112 So, the original out-of-stock level[br]was about 7%, 0:12:39.112,0:12:44.811 and then when the project started,[br]you see that black line there, 0:12:44.811,0:12:49.112 it got less, that's what we want,[br]it's good if it's small, 0:12:49.125,0:12:52.623 and then suddenly, one day,[br]we really automated it, 0:12:52.623,0:12:55.492 and no humans were allowed[br]to do anything anymore. 0:12:55.492,0:12:59.054 Then you see, the result[br]was much better, and constantly better. 0:12:59.364,0:13:02.914 So that's the good or bad thing. 0:13:02.914,0:13:04.934 So, three examples at the end. 0:13:04.934,0:13:08.094 Automatic fresh-meat replenishment[br]for supermarket chains 0:13:08.094,0:13:10.014 does work very well, 0:13:10.014,0:13:14.744 and in some German supermarkets,[br]it is done automatically. 0:13:15.024,0:13:17.106 Also dynamic pricing: 0:13:18.456,0:13:22.018 in retail, it's something[br]that is appearing more frequently. 0:13:22.018,0:13:24.507 With the Internet, it's normal already. 0:13:24.507,0:13:27.006 It's also coming[br]in brick-and-mortar retail. 0:13:27.596,0:13:29.851 It's not bad even for the customer, right? 0:13:29.851,0:13:33.027 Because a retailer can never fight[br]against the customer, 0:13:33.027,0:13:37.335 the customer has to want[br]to come and to buy, 0:13:37.335,0:13:40.165 so it does not mean[br]that the price is always higher; 0:13:40.165,0:13:45.807 it just means the right price,[br]at the right time, for the right article. 0:13:46.137,0:13:52.332 So most retailers do not fight against[br]the customers, but against competitors, 0:13:52.602,0:13:57.072 and the competitors also use[br]methods like these nowadays. 0:13:58.602,0:14:03.982 So finally, another example from research, 0:14:04.192,0:14:07.422 from the collaboration[br]of Blue Yonder and KIT, 0:14:07.422,0:14:11.113 we have built such[br]a decision system onto one chip, 0:14:11.343,0:14:15.221 and this now holds[br]the world record in decisions, 0:14:15.241,0:14:17.662 it does eight billion[br]decisions per second, 0:14:17.732,0:14:21.704 and that is for the next generation[br]of particle accelerators. 0:14:21.704,0:14:27.912 So much data is produced in the sensors, [br]that we cannot read them out anymore. 0:14:27.925,0:14:33.193 So directly on the sensors,[br]there's a chip which decides: 0:14:33.201,0:14:37.093 "Okay, in this event,[br]this part of the response 0:14:37.115,0:14:39.443 seems to be important so we save it, 0:14:39.459,0:14:41.480 we put it in a computer, 0:14:41.498,0:14:44.449 and the others are not[br]even read out anymore." 0:14:44.491,0:14:46.464 So, these are things that are coming. 0:14:46.472,0:14:48.860 And here, it's very clear,[br]here it's simply time, 0:14:48.880,0:14:51.426 no human would have the time[br]to make the decision, 0:14:51.435,0:14:53.996 it has to be extremely fast. 0:14:54.262,0:14:58.478 What I'm talking about here is not[br]the far future, but it exists now, 0:14:58.478,0:15:01.368 and for sure, it will evolve further. 0:15:01.722,0:15:06.678 So, my prediction is that[br]quite some repetitive routine work, 0:15:06.678,0:15:08.478 but also white-collar work, 0:15:08.478,0:15:13.071 so it's not only the workers[br]whose jobs will change in the future, 0:15:13.071,0:15:17.061 but also, like me, white-collar work[br]will be automated. 0:15:17.061,0:15:18.953 Should we be afraid of it? 0:15:19.513,0:15:22.923 I think not, it is normal innovation,[br]it makes us stronger, 0:15:22.930,0:15:29.791 it makes our strategic decisions[br]more efficient and also sustainable. 0:15:30.156,0:15:32.505 I think that's the way[br]towards a better future, 0:15:32.505,0:15:36.690 and I am completely convinced[br]that it is unavoidable. It will come. 0:15:36.696,0:15:38.093 We don't have to discuss it, 0:15:38.097,0:15:40.617 it's only a question of time,[br]it will come, for sure. 0:15:40.617,0:15:41.707 Thank you very much. 0:15:41.707,0:15:43.629 (Applause)