WEBVTT 1 00:00:02.730 --> 00:00:04.530 When you're putting a new tire on your car, 2 00:00:04.530 --> 00:00:06.350 you don't want to tighten one bolt all the way 3 00:00:06.350 --> 00:00:07.560 and then tighten the rest. 4 00:00:07.560 --> 00:00:09.080 You want to tighten them all a little bit 5 00:00:09.080 --> 00:00:10.684 and continuously tighten. 6 00:00:10.684 --> 00:00:11.994 What does that have to do with 7 00:00:11.994 --> 00:00:13.684 artificial intelligence and fashion? 8 00:00:13.684 --> 00:00:16.879 Find out today, when we talk with Arti Zeighami from H&M. 9 00:00:16.879 --> 00:00:19.913 (upbeat music) 10 00:00:19.913 --> 00:00:21.870 Welcome to Me, Myself, and AI, 11 00:00:21.870 --> 00:00:24.860 a podcast on artificial intelligence and business. 12 00:00:24.860 --> 00:00:28.050 Each week we introduce you to someone innovating with AI. 13 00:00:28.050 --> 00:00:29.250 I'm Sam Ransbotham, 14 00:00:29.250 --> 00:00:32.150 professor of information systems at Boston College, 15 00:00:32.150 --> 00:00:33.480 and I'm also the guest editor, 16 00:00:33.480 --> 00:00:36.203 for the AI and Business Strategy Big Ideas Program, 17 00:00:36.203 --> 00:00:38.272 at MIT Sloan Management Review. 18 00:00:38.272 --> 00:00:40.120 And I'm Shervin Khodabandeh, 19 00:00:40.120 --> 00:00:42.180 senior partner with BCG, 20 00:00:42.180 --> 00:00:46.000 and I colead BCG's AI practice in North America. 21 00:00:46.000 --> 00:00:49.110 And together, BCG and MIT SMR 22 00:00:49.110 --> 00:00:52.420 have been researching AI for four years, 23 00:00:52.420 --> 00:00:54.680 interviewing hundreds of practitioners 24 00:00:54.680 --> 00:00:56.800 and surveying thousands of companies 25 00:00:56.800 --> 00:01:00.120 on what it takes to build and deploy 26 00:01:00.120 --> 00:01:02.680 and scale AI capabilities, 27 00:01:02.680 --> 00:01:05.488 and really transform the way organizations operate. 28 00:01:05.488 --> 00:01:07.866 (ethereal music) 29 00:01:07.866 --> 00:01:10.350 Today we're talking with Arti Zeighami. 30 00:01:10.350 --> 00:01:13.250 He leads artificial intelligence at H&M. 31 00:01:13.250 --> 00:01:16.260 Arti is joining us from Stockholm, welcome Arti. 32 00:01:16.260 --> 00:01:19.270 Thank you, thank you very much for having me here. 33 00:01:19.270 --> 00:01:20.790 Really we'd like to hear 34 00:01:20.790 --> 00:01:21.870 a little bit about your background. 35 00:01:21.870 --> 00:01:22.780 Why don't we start there? 36 00:01:22.780 --> 00:01:24.100 How did you, how'd you get interested 37 00:01:24.100 --> 00:01:26.066 in artificial intelligence? And 38 00:01:26.066 --> 00:01:28.570 for our podcast today, I'm actually wearing a nice shirt 39 00:01:28.570 --> 00:01:31.370 because we're talking with a fashion person, 40 00:01:31.370 --> 00:01:32.330 even though this is audio, 41 00:01:32.330 --> 00:01:35.138 I can assure everyone that I look fabulous. 42 00:01:35.138 --> 00:01:36.230 How did you get interested 43 00:01:36.230 --> 00:01:38.067 in artificial intelligence though? 44 00:01:38.067 --> 00:01:40.010 I think my interest in the area 45 00:01:40.010 --> 00:01:43.680 started many years ago as a, you know, a young teenager, 46 00:01:43.680 --> 00:01:46.850 when the, like a lot of the people I was, you know, 47 00:01:46.850 --> 00:01:50.260 studying math and physics and loved all those stuff, 48 00:01:50.260 --> 00:01:53.390 and I read these books by Isaac Asimov, you know, 49 00:01:53.390 --> 00:01:57.070 the science fiction author. And I was reading about this, 50 00:01:57.070 --> 00:01:59.732 this theory that he had about psycho history, they call it, 51 00:01:59.732 --> 00:02:04.732 which was about how you can start somehow predict the future 52 00:02:04.880 --> 00:02:06.400 by looking at the past, 53 00:02:06.400 --> 00:02:09.077 and apply mathematical models on top of that. 54 00:02:09.077 --> 00:02:12.620 And I was so intrigued by that, when I was 15, 16, or 17, 55 00:02:12.620 --> 00:02:14.350 I was like, this is something amazing. 56 00:02:14.350 --> 00:02:15.850 This guy made it up by himself, 57 00:02:15.850 --> 00:02:18.450 and you know, what if this can be for true? 58 00:02:18.450 --> 00:02:21.630 What if I could work with something like this in the future? 59 00:02:21.630 --> 00:02:23.650 And, you know, it took me a long time to get there 60 00:02:23.650 --> 00:02:25.599 because there between school came, 61 00:02:25.599 --> 00:02:27.210 and then I started working as a consultant, 62 00:02:27.210 --> 00:02:30.000 and I was doing programming, I was in architecture, 63 00:02:30.000 --> 00:02:32.860 I was a business developer, I was a strategist. 64 00:02:32.860 --> 00:02:34.980 I did, you know, different startups and all that, 65 00:02:34.980 --> 00:02:38.548 and then I ended up in a fashion retailer. 66 00:02:38.548 --> 00:02:41.930 Then I got this opportunity to start looking at, 67 00:02:41.930 --> 00:02:45.110 advanced analytics as an AI, as a capability. 68 00:02:45.110 --> 00:02:48.180 I don't have a formal data science background, 69 00:02:48.180 --> 00:02:50.640 I do have an engineering background from engineering school, 70 00:02:50.640 --> 00:02:53.170 I even started business school parallel to that, you know, 71 00:02:53.170 --> 00:02:55.100 and then life brought me here. 72 00:02:55.100 --> 00:02:56.950 Somehow it's like the universe brought me 73 00:02:56.950 --> 00:03:00.107 to artificial intelligence and somebody made fun of this. 74 00:03:00.107 --> 00:03:02.060 "Yeah, your name is Arti, that's already Arti, artificial. 75 00:03:02.060 --> 00:03:03.852 It's probably that's why," so, yeah. 76 00:03:03.852 --> 00:03:05.344 Oh yeah-- 77 00:03:05.344 --> 00:03:07.317 That's pretty funny. 78 00:03:07.317 --> 00:03:09.495 Yeah, I always joke about that. 79 00:03:09.495 --> 00:03:11.570 Joking aside, I would say, 80 00:03:11.570 --> 00:03:15.420 like what you described as a diverse and colorful background 81 00:03:15.420 --> 00:03:20.340 and experiences, you know, architect, engineer, strategist, 82 00:03:20.340 --> 00:03:22.650 how does it help you now? 83 00:03:22.650 --> 00:03:25.136 That diversity versus, if you'd been all focused, 84 00:03:25.136 --> 00:03:26.960 do you have a point of view on that? 85 00:03:26.960 --> 00:03:28.240 Yeah, absolutely. 86 00:03:28.240 --> 00:03:30.910 I think it has helped me tremendously. 87 00:03:30.910 --> 00:03:34.440 I think one of the major, you know, 88 00:03:34.440 --> 00:03:37.382 parts of working with introducing a new capability, 89 00:03:37.382 --> 00:03:42.382 such as AI, into a, you know, old industry, like in retail, 90 00:03:43.003 --> 00:03:44.960 which has done certain things in a certain way 91 00:03:44.960 --> 00:03:48.380 for so many years, is about shift of mindset. 92 00:03:48.380 --> 00:03:51.000 It's about transforming people's way of thinking it's, 93 00:03:51.000 --> 00:03:54.750 you know, it's very little AI, it's very little tech. 94 00:03:54.750 --> 00:03:58.640 I usually refer to that 10% AI, 20% tech, 95 00:03:58.640 --> 00:04:01.210 but it's 70% people and processes. 96 00:04:01.210 --> 00:04:04.170 So you try to shift people into thinking differently, 97 00:04:04.170 --> 00:04:06.780 to ask different questions, to, you know, 98 00:04:06.780 --> 00:04:09.880 look at the world differently and working as a consultant, 99 00:04:09.880 --> 00:04:12.380 I think that was one of the biggest help I ever got, 100 00:04:12.380 --> 00:04:14.890 because as a consultant, you always make sure that, 101 00:04:14.890 --> 00:04:17.570 it's not about you shining, it's about your clients shining, 102 00:04:17.570 --> 00:04:20.070 and it's about making them the hero of the day, 103 00:04:20.070 --> 00:04:21.600 and it's always been about that. 104 00:04:21.600 --> 00:04:23.840 And I kind of sort of even internally would, 105 00:04:23.840 --> 00:04:25.767 you know, my colleagues and my peers always said, 106 00:04:25.767 --> 00:04:27.610 "Listen, let's make sure that 107 00:04:27.610 --> 00:04:29.670 we are almost like internal consultants 108 00:04:29.670 --> 00:04:32.080 because it's about helping our colleagues 109 00:04:32.080 --> 00:04:33.650 to achieve their goals." 110 00:04:33.650 --> 00:04:34.600 Ultimately, if you're talking about 111 00:04:34.600 --> 00:04:37.099 transforming people's mindset, it's about the rhetoric's, 112 00:04:37.099 --> 00:04:40.390 it's how you make them understand what you're trying to do 113 00:04:40.390 --> 00:04:42.243 and how you make them understand 114 00:04:42.243 --> 00:04:44.160 what you're trying to help them with. 115 00:04:44.160 --> 00:04:46.460 So, you know, it goes back to the Greeks rhetoric, 116 00:04:46.460 --> 00:04:47.821 ethos, pathos, logos. 117 00:04:47.821 --> 00:04:51.290 That's a long way from the ancient Greeks to, 118 00:04:51.290 --> 00:04:53.610 from Greece to Sweden here, I guess. 119 00:04:53.610 --> 00:04:54.810 Is there something specific 120 00:04:54.810 --> 00:04:58.100 that AI has been able to help with at H&M? 121 00:04:58.100 --> 00:04:59.494 Absolutely, you know, we as a company 122 00:04:59.494 --> 00:05:01.340 have always been analytical. 123 00:05:01.340 --> 00:05:03.873 If you go back to how the company was brought up, even, 124 00:05:03.873 --> 00:05:06.010 you know, the company is a family, 125 00:05:06.010 --> 00:05:08.340 that started as this, the personal family, 126 00:05:08.340 --> 00:05:11.650 the third generation is now chairman of the company. 127 00:05:11.650 --> 00:05:13.740 Erling who started this back in 1947. 128 00:05:13.740 --> 00:05:15.200 It was very analytical on the, 129 00:05:15.200 --> 00:05:16.398 on this way already back those days, 130 00:05:16.398 --> 00:05:18.620 you know, there's a story is about, 131 00:05:18.620 --> 00:05:20.770 something they called "Following the bags," 132 00:05:20.770 --> 00:05:23.200 you know, when he was trying to enter a new city, 133 00:05:23.200 --> 00:05:24.892 he send out people with pens and papers 134 00:05:24.892 --> 00:05:26.320 and then will walk in the street, 135 00:05:26.320 --> 00:05:29.060 looking at the bags of the people or where they were going, 136 00:05:29.060 --> 00:05:31.450 and if they were crossing an intersection, 137 00:05:31.450 --> 00:05:33.557 or this side or that side, and they didn't understand, 138 00:05:33.557 --> 00:05:34.390 "Okay, what sort of intersection 139 00:05:34.390 --> 00:05:35.960 that should be on the store?" 140 00:05:35.960 --> 00:05:38.285 We start to look at this artificial intelligence, 141 00:05:38.285 --> 00:05:43.011 back in 2016 and try to understand what it entails for us 142 00:05:43.011 --> 00:05:47.310 as a retailer, and we entered a very small area 143 00:05:47.310 --> 00:05:49.300 and we did a proof of concept in that area, 144 00:05:49.300 --> 00:05:52.510 within the personalization to understand, you know, 145 00:05:52.510 --> 00:05:56.210 how we can enhance the communication, the personalization, 146 00:05:56.210 --> 00:05:57.670 the offering to our customer 147 00:05:57.670 --> 00:06:00.470 in a way utilizing AI analytics. 148 00:06:00.470 --> 00:06:03.431 And we saw that based upon the amount of data that we have, 149 00:06:03.431 --> 00:06:06.040 you know, the vast information that we have of our product, 150 00:06:06.040 --> 00:06:08.010 of our sales, of our customer, 151 00:06:08.010 --> 00:06:09.960 we can be really precise on that. 152 00:06:09.960 --> 00:06:13.740 That by itself was not pivotal, it was more of understanding 153 00:06:13.740 --> 00:06:16.490 how you change the mindset to set, you know, 154 00:06:16.490 --> 00:06:18.570 you want people to come in Monday morning 155 00:06:18.570 --> 00:06:20.380 and ask different questions. 156 00:06:20.380 --> 00:06:22.900 And to do that, you need to get them more analytical, 157 00:06:22.900 --> 00:06:26.650 and in order to penetrate that into the entire organization, 158 00:06:26.650 --> 00:06:29.260 we took an approach that was a little bit different, 159 00:06:29.260 --> 00:06:30.854 because a lot of people asked me back in those days, 160 00:06:30.854 --> 00:06:32.922 you know, "How did you pick your AI use cases?" 161 00:06:32.922 --> 00:06:35.440 And I said, "I don't have AI use cases. 162 00:06:35.440 --> 00:06:38.700 I have like business challenges that my colleagues have. 163 00:06:38.700 --> 00:06:41.390 These are, I look at the portfolio of our project and see, 164 00:06:41.390 --> 00:06:43.862 we have a lot of problems, we have lot of challenges, 165 00:06:43.862 --> 00:06:46.526 and then those challenges are have entailed into projects, 166 00:06:46.526 --> 00:06:48.770 they're going to change and fix those things, 167 00:06:48.770 --> 00:06:53.550 and here I can come in and help in a very small part of it." 168 00:06:53.550 --> 00:06:56.160 And I think that was also a very important part, 169 00:06:56.160 --> 00:06:58.300 how you infuse something new in an organization, 170 00:06:58.300 --> 00:07:01.103 because a lot of people, again, take one part, 171 00:07:01.103 --> 00:07:02.180 a very small part, 172 00:07:02.180 --> 00:07:04.570 and then they do deep dive on that small part, 173 00:07:04.570 --> 00:07:06.300 and they create that, you know, a sexy app 174 00:07:06.300 --> 00:07:09.740 or a cool customer facing stuff, and that's fine. 175 00:07:09.740 --> 00:07:12.400 But then you make one part of your organization 176 00:07:12.400 --> 00:07:14.092 to become very, very good at that, 177 00:07:14.092 --> 00:07:17.410 and then the rest are still, you know, lagging behind. 178 00:07:17.410 --> 00:07:20.540 I believe you need to elevate everybody a little bit. 179 00:07:20.540 --> 00:07:23.640 So instead of, it's almost like putting a tire on a car, 180 00:07:23.640 --> 00:07:25.757 right, you don't screw one bolt really hard 181 00:07:25.757 --> 00:07:29.130 and then do the next one. You just do every by a little bit, 182 00:07:29.130 --> 00:07:30.740 and then tighten everything up. 183 00:07:30.740 --> 00:07:32.650 And I think that has been a really good approach 184 00:07:32.650 --> 00:07:34.050 for us to do that to everybody, 185 00:07:34.050 --> 00:07:36.610 and I'm enhancing stuff in the beginning of the value chain 186 00:07:36.610 --> 00:07:39.700 with fashion forecasting, with quantification, 187 00:07:39.700 --> 00:07:42.310 how you quantify, how many pieces you buy, to 188 00:07:42.310 --> 00:07:43.863 how you allocate the garments throughout 189 00:07:43.863 --> 00:07:47.235 the whole value chain, to how that puts prices on them, 190 00:07:47.235 --> 00:07:49.750 and maybe also working with personalization 191 00:07:49.750 --> 00:07:52.860 and all those fancy customer facing stuff as well. 192 00:07:52.860 --> 00:07:57.210 And for us, AI has not meant artificial intelligence, 193 00:07:57.210 --> 00:08:00.310 we have always talked about amplified intelligence, 194 00:08:00.310 --> 00:08:02.930 because we're amplifying an existing knowledge 195 00:08:02.930 --> 00:08:04.840 and competence of our colleagues. 196 00:08:04.840 --> 00:08:07.730 So, it doesn't have to be the AI that does the decision, 197 00:08:07.730 --> 00:08:08.563 it could be combination, 198 00:08:08.563 --> 00:08:12.530 and we see that when we do the combination of AI and machine 199 00:08:12.530 --> 00:08:16.250 and human, the gut feeling of the data, the art and science, 200 00:08:16.250 --> 00:08:18.300 that's when we get the most out of it. 201 00:08:18.300 --> 00:08:21.850 I see a lot of things that we do today is that mixture. 202 00:08:21.850 --> 00:08:23.320 I want to pick up on, 203 00:08:23.320 --> 00:08:26.500 what do you referred to as amplified intelligence? 204 00:08:26.500 --> 00:08:28.340 I think it's very elegant, 205 00:08:28.340 --> 00:08:32.240 and it sort of also underlies the theme you started 206 00:08:32.240 --> 00:08:36.930 this conversation with around organization, 207 00:08:36.930 --> 00:08:41.792 people, you know, 70% is the people and organizations. 208 00:08:41.792 --> 00:08:42.625 Yes. 209 00:08:42.625 --> 00:08:44.070 And it also ties very well 210 00:08:44.070 --> 00:08:46.820 with the research we just did, which is all again, 211 00:08:46.820 --> 00:08:48.860 talks about the role of human 212 00:08:48.860 --> 00:08:50.580 and sometimes the misunderstood 213 00:08:50.580 --> 00:08:52.760 or understated role of human. 214 00:08:52.760 --> 00:08:55.253 Comment more on that, and particularly, 215 00:08:56.110 --> 00:08:59.450 different ways that human and AI can interact. 216 00:08:59.450 --> 00:09:02.230 You know, across these different business problems. 217 00:09:02.230 --> 00:09:05.520 Right, right. When we did our first pilot, 218 00:09:05.520 --> 00:09:09.150 a test of utilizing AI and advanced analytics, 219 00:09:09.150 --> 00:09:12.160 and end-of-season sales, and that was my first attempt 220 00:09:12.160 --> 00:09:14.680 to try to use it on an actual use case 221 00:09:14.680 --> 00:09:18.380 and an actual business challenge. We saw that very early, 222 00:09:18.380 --> 00:09:20.860 that the AI could actually enhance 223 00:09:20.860 --> 00:09:22.911 what was much better than the human on putting the prices. 224 00:09:22.911 --> 00:09:25.840 I mean, end-of-season sales result. 225 00:09:25.840 --> 00:09:28.050 And important part of that journey was to make sure 226 00:09:28.050 --> 00:09:30.370 that the teams, they were actually applying it, 227 00:09:30.370 --> 00:09:32.300 not my team, not the AI team, 228 00:09:32.300 --> 00:09:34.820 but actually the people that are working the merchandise 229 00:09:34.820 --> 00:09:38.020 online on a very small selected markets, 230 00:09:38.020 --> 00:09:41.560 I let them actually calculate what the outcome was. 231 00:09:41.560 --> 00:09:44.340 So they both were responsible for the test, 232 00:09:44.340 --> 00:09:45.380 for setting up the test, 233 00:09:45.380 --> 00:09:47.250 putting all the constraints they want to me, 234 00:09:47.250 --> 00:09:50.000 and making sure that my algo was not getting anything else 235 00:09:50.000 --> 00:09:52.410 than the merchandise that we're getting on a daily basis, 236 00:09:52.410 --> 00:09:54.430 and then they'd try to understand, 237 00:09:54.430 --> 00:09:57.350 how better it is by actually calculating it themself. 238 00:09:57.350 --> 00:10:01.040 This is perfect, you know, it helps us to enhance our job. 239 00:10:01.040 --> 00:10:03.040 Let's do a next test, for the next season, 240 00:10:03.040 --> 00:10:05.140 mid-season sales, so a couple of months later, 241 00:10:05.140 --> 00:10:06.703 we added the test and we made it a little bit larger. 242 00:10:06.703 --> 00:10:10.440 We added another country, another market, two warehouses, 243 00:10:10.440 --> 00:10:13.660 sell them on the products, and we tested that. 244 00:10:13.660 --> 00:10:15.710 I want to add a little more complexity to this, 245 00:10:15.710 --> 00:10:19.060 But one thing that we want to do is also add a third bucket. 246 00:10:19.060 --> 00:10:21.410 So we're still going to have a few products to look at, 247 00:10:21.410 --> 00:10:23.170 but we want to divide it, not in two buckets, 248 00:10:23.170 --> 00:10:25.058 but three buckets, where one bucket is that 249 00:10:25.058 --> 00:10:26.630 algo putting the price on, 250 00:10:26.630 --> 00:10:28.410 one bucket is the merchandising, 251 00:10:28.410 --> 00:10:31.390 and one bucket is the algo putting the price on 252 00:10:31.390 --> 00:10:32.960 and the merchandise merchandising coming in 253 00:10:32.960 --> 00:10:34.760 and tweak those prices. 254 00:10:34.760 --> 00:10:38.130 Because we saw there some things that algo isn't good at. 255 00:10:38.130 --> 00:10:40.077 And we found that very interesting, and said, 256 00:10:40.077 --> 00:10:41.870 "Let's absolutely, let's do this." 257 00:10:41.870 --> 00:10:44.308 And then, we did the mid-season test, 258 00:10:44.308 --> 00:10:47.840 and the outcome of that was even more interesting. 259 00:10:47.840 --> 00:10:49.107 Because the algo was again, 260 00:10:49.107 --> 00:10:51.540 you know, a few percentage better than human, 261 00:10:51.540 --> 00:10:54.990 but the algo in combination with human, 262 00:10:54.990 --> 00:10:58.010 was twice as good as the algo itself. 263 00:10:58.010 --> 00:10:59.490 And actually it was, then we, 264 00:10:59.490 --> 00:11:01.530 we start talking about amplified intelligence 265 00:11:01.530 --> 00:11:04.040 because we realized the machine by itself won't help us. 266 00:11:04.040 --> 00:11:05.770 It's a combination of the human and the machine, 267 00:11:05.770 --> 00:11:07.197 the gut feeling and the data. 268 00:11:07.197 --> 00:11:08.450 May have to change your name 269 00:11:08.450 --> 00:11:09.694 from Arti to Amfi. 270 00:11:09.694 --> 00:11:12.194 (Arti laughs) 271 00:11:13.100 --> 00:11:15.748 Yeah, oh my mom won't be happy. 272 00:11:15.748 --> 00:11:18.550 Yeah, your mom might not be happy there. 273 00:11:18.550 --> 00:11:21.260 You described a process where you went further in depth 274 00:11:21.260 --> 00:11:23.490 into a pricing process, 275 00:11:23.490 --> 00:11:25.947 but earlier you were talking about a process of saying, 276 00:11:25.947 --> 00:11:28.049 "Well, we need to do lots of things in different areas." 277 00:11:28.049 --> 00:11:28.882 Yes. 278 00:11:28.882 --> 00:11:30.790 How do you balance that, 279 00:11:30.790 --> 00:11:32.610 tightening all the bolts 280 00:11:32.610 --> 00:11:34.648 versus tightening this one bolt harder? 281 00:11:34.648 --> 00:11:36.350 Because it sounded like in that example, 282 00:11:36.350 --> 00:11:38.681 you were doing some more tightening of one bolt. 283 00:11:38.681 --> 00:11:40.370 How do you balance those out? 284 00:11:40.370 --> 00:11:42.480 Well, you didn't hear the whole story. 285 00:11:42.480 --> 00:11:43.410 Oh, there's more, okay. 286 00:11:43.410 --> 00:11:44.537 Already from their first tighten up 287 00:11:44.537 --> 00:11:46.565 of the first bolt, so the first test that I got, 288 00:11:46.565 --> 00:11:49.340 which was good result, but it was not enough, 289 00:11:49.340 --> 00:11:51.500 already then I was happy about those results, 290 00:11:51.500 --> 00:11:53.698 and I took that result and went another part of the business 291 00:11:53.698 --> 00:11:56.410 and said, "Listen, we did this with these guys. 292 00:11:56.410 --> 00:11:58.100 They were really happy about the result. 293 00:11:58.100 --> 00:12:01.630 We saw that we were X amount percent better on net sales 294 00:12:01.630 --> 00:12:05.470 by using an algo which we created in four and a half weeks, 295 00:12:05.470 --> 00:12:08.350 and it has a huge impact on the business. 296 00:12:08.350 --> 00:12:10.390 Do you want to us to help you to look at this area? 297 00:12:10.390 --> 00:12:12.470 Because I know you have a problem here." 298 00:12:12.470 --> 00:12:14.300 And then we brought in the data scientists, 299 00:12:14.300 --> 00:12:16.020 and we put the experiments around that. 300 00:12:16.020 --> 00:12:17.760 And then when we started that discussion 301 00:12:17.760 --> 00:12:20.390 and that conversation for that specific project, 302 00:12:20.390 --> 00:12:22.450 it was connected to another part of their business, 303 00:12:22.450 --> 00:12:24.624 and they say, "Hey, if you're going to do change there, 304 00:12:24.624 --> 00:12:27.040 maybe we should do change here as well. 305 00:12:27.040 --> 00:12:28.667 Arti, do you want to help us in this case?" 306 00:12:28.667 --> 00:12:30.400 "Yes, please let me help." 307 00:12:30.400 --> 00:12:32.384 And then we started doing that, and then that follows, 308 00:12:32.384 --> 00:12:35.970 and by the end of the year, suddenly we had those eight, 309 00:12:35.970 --> 00:12:38.580 seven, whatever use cases that we had, 310 00:12:38.580 --> 00:12:41.570 and then we saw that we're actually applying this 311 00:12:41.570 --> 00:12:43.380 throughout the whole value chain. 312 00:12:43.380 --> 00:12:46.950 So, and then, meanwhile you start finding each of the bolts 313 00:12:46.950 --> 00:12:47.930 little bit more and more. 314 00:12:47.930 --> 00:12:48.870 It's like a pitch crew. 315 00:12:48.870 --> 00:12:50.300 Yeah (chuckles) exactly. 316 00:12:50.300 --> 00:12:52.300 And it's the whole line of being agile, right? 317 00:12:52.300 --> 00:12:54.810 You start small, you dream big, 318 00:12:54.810 --> 00:12:57.110 you start small and you scale fast. 319 00:12:57.110 --> 00:12:58.464 So, you start small with something here, 320 00:12:58.464 --> 00:13:00.396 and then you start the next wave and the next wave, 321 00:13:00.396 --> 00:13:02.340 the next wave, and all these waves 322 00:13:02.340 --> 00:13:04.610 have a cycle of starting small, 323 00:13:04.610 --> 00:13:07.150 testing a little bit larger, failing, pivoting, 324 00:13:07.150 --> 00:13:08.916 and testing more, failing, pivoting, learning, 325 00:13:08.916 --> 00:13:11.140 and then goes on and goes on, and you know, 326 00:13:11.140 --> 00:13:14.180 organizations such as ours, we are huge like 5,000 stores, 327 00:13:14.180 --> 00:13:17.530 70 plus countries, 180,000 people, you know. 328 00:13:17.530 --> 00:13:19.980 So there's huge amount of things when you start 329 00:13:19.980 --> 00:13:21.890 to industrialize that. 330 00:13:21.890 --> 00:13:25.780 And AI doesn't mean anything if you don't industrialize it. 331 00:13:25.780 --> 00:13:27.710 And Arti, as you were talking about 332 00:13:27.710 --> 00:13:31.970 industrializing AI and getting real scale out of it, 333 00:13:31.970 --> 00:13:34.390 in the context of amplified intelligence, 334 00:13:34.390 --> 00:13:35.223 you know. Yes. 335 00:13:35.223 --> 00:13:38.780 Machine is easy to get them to play with human 336 00:13:38.780 --> 00:13:40.520 because they don't have choice. 337 00:13:40.520 --> 00:13:45.130 How do you get the human to play nice with the machine 338 00:13:45.130 --> 00:13:48.500 and what's the kind of pushback 339 00:13:48.500 --> 00:13:49.976 that you would expect or you get, 340 00:13:49.976 --> 00:13:52.199 and how do you deal with that? 341 00:13:52.199 --> 00:13:55.850 So, it's different from, you know, 342 00:13:55.850 --> 00:13:58.918 different levels of people in the organization. 343 00:13:58.918 --> 00:13:59.940 Mm-hmm. 344 00:13:59.940 --> 00:14:01.380 One thing I found out is that, 345 00:14:01.380 --> 00:14:03.663 maybe it's also about the culture in the organization as 346 00:14:03.663 --> 00:14:06.580 well, right? So we come from a company with a culture of, 347 00:14:06.580 --> 00:14:09.264 always being entrepreneurial, always seeking for, 348 00:14:09.264 --> 00:14:11.440 you know, improving ourselves. 349 00:14:11.440 --> 00:14:13.470 And that has been part of the organization, 350 00:14:13.470 --> 00:14:16.534 it's in our DNA somehow to always try to be better. 351 00:14:16.534 --> 00:14:18.820 And that's why also we talked about, 352 00:14:18.820 --> 00:14:20.870 you know, the combination that we're just taking 353 00:14:20.870 --> 00:14:22.230 a small part of the work, 354 00:14:22.230 --> 00:14:24.750 and we're actually internal consultants. 355 00:14:24.750 --> 00:14:26.700 So, when they own that, 356 00:14:26.700 --> 00:14:29.130 it also makes them feel pride about what they're doing. 357 00:14:29.130 --> 00:14:32.080 We really went for this DevOps model 358 00:14:32.080 --> 00:14:34.480 where we put people into teams 359 00:14:34.480 --> 00:14:35.548 where you have a use case lead, 360 00:14:35.548 --> 00:14:37.101 and then you have business experts, 361 00:14:37.101 --> 00:14:39.950 and you have machinery engineers, data engineers, 362 00:14:39.950 --> 00:14:42.120 software engineers, UX designers, 363 00:14:42.120 --> 00:14:46.230 all working together on a daily basis in a very agile way 364 00:14:46.230 --> 00:14:48.930 by sitting together in the same office, 365 00:14:48.930 --> 00:14:50.460 having morning stand-ups. 366 00:14:50.460 --> 00:14:52.310 So they were part the whole development team, 367 00:14:52.310 --> 00:14:53.856 and my job to make sure that, 368 00:14:53.856 --> 00:14:56.570 my other colleagues also understand the impact, 369 00:14:56.570 --> 00:14:59.570 and then utilize this to, you know, realize their value 370 00:14:59.570 --> 00:15:00.850 in their part of the company, 371 00:15:00.850 --> 00:15:01.870 they're part of our organization, 372 00:15:01.870 --> 00:15:03.640 making our company continue to thrive 373 00:15:03.640 --> 00:15:05.310 and become even better. 374 00:15:05.310 --> 00:15:07.150 I think there's an important lesson in there 375 00:15:07.150 --> 00:15:09.764 for everyone as well, including for CEO's 376 00:15:09.764 --> 00:15:11.317 and heads of businesses. 377 00:15:11.317 --> 00:15:15.920 We are underscoring the importance of focus on value versus 378 00:15:15.920 --> 00:15:19.700 a single sort of minded, tech centric 379 00:15:19.700 --> 00:15:21.265 focus on building capabilities, 380 00:15:21.265 --> 00:15:25.520 and building more and more capabilities, and the need for 381 00:15:25.520 --> 00:15:28.280 the business and the users and the people 382 00:15:28.280 --> 00:15:31.530 to come in from the beginning to design that. 383 00:15:31.530 --> 00:15:33.443 Arti, thank you so much for making time. 384 00:15:33.443 --> 00:15:35.480 Yes, thanks for taking the time today. 385 00:15:35.480 --> 00:15:36.769 Thank you guys. 386 00:15:36.769 --> 00:15:39.519 (ethereal music) 387 00:15:40.880 --> 00:15:42.680 Sam, I thought that was such a stimulating 388 00:15:42.680 --> 00:15:45.282 conversation with Arti, let's quickly recap. 389 00:15:45.282 --> 00:15:48.230 Sounds good. Shervin, it's interesting, you know, 390 00:15:48.230 --> 00:15:49.270 we talked with Porsche, 391 00:15:49.270 --> 00:15:52.501 we never even talked about tires and hubcaps and tightening, 392 00:15:52.501 --> 00:15:55.730 (chuckles) tightening bolts, but when we turn to fashion, 393 00:15:55.730 --> 00:15:58.340 he talked about the importance of doing that, 394 00:15:58.340 --> 00:16:00.750 and his analogy was very ultra-related. 395 00:16:00.750 --> 00:16:03.130 That's right, and when we talked with Porsche, 396 00:16:03.130 --> 00:16:04.140 if you're talking about coffee, 397 00:16:04.140 --> 00:16:05.330 (chuckles) Exactly. 398 00:16:05.330 --> 00:16:08.580 Everyone is, you know, joking aside, 399 00:16:08.580 --> 00:16:10.550 I think the common thing we're hearing, 400 00:16:10.550 --> 00:16:14.260 is the importance of archetypal problems 401 00:16:14.260 --> 00:16:16.760 and translating or transferring these learnings 402 00:16:16.760 --> 00:16:20.490 across problems, which interestingly enough is a 403 00:16:20.490 --> 00:16:22.833 topic in AI itself like transfer learning. 404 00:16:22.833 --> 00:16:26.350 This is not exactly mean what we're talking about here. 405 00:16:26.350 --> 00:16:29.120 I also thought it was quite interesting 406 00:16:29.120 --> 00:16:31.527 how from the beginning Arti said, 407 00:16:31.527 --> 00:16:35.758 "Look, it's about changing the mindset of the people, 408 00:16:35.758 --> 00:16:38.897 and it's about the organization, and it's about." 409 00:16:38.897 --> 00:16:42.030 You know, he talked about it as amplified intelligence, 410 00:16:42.030 --> 00:16:45.260 you know, bringing human and AI together 411 00:16:45.260 --> 00:16:47.866 rather than all one or the other. 412 00:16:47.866 --> 00:16:49.540 I think if we pulled out the words, 413 00:16:49.540 --> 00:16:51.321 he said learning more than anything else. 414 00:16:51.321 --> 00:16:53.020 That's right, that's right. 415 00:16:53.020 --> 00:16:55.000 The other important point he made, 416 00:16:55.000 --> 00:16:57.610 which I think might be lost on many, 417 00:16:57.610 --> 00:16:59.810 is that you can't just start with technology 418 00:16:59.810 --> 00:17:02.780 and capabilities and, you know, 419 00:17:02.780 --> 00:17:03.898 reminds me of "Field of Dreams," 420 00:17:03.898 --> 00:17:06.250 you know, "If you build it, they will come." 421 00:17:06.250 --> 00:17:07.610 I think this is-- He said the opposite. 422 00:17:07.610 --> 00:17:09.130 Exactly the opposite. 423 00:17:09.130 --> 00:17:12.090 You can't build it and wait for them to come, 424 00:17:12.090 --> 00:17:14.110 you actually have to build it together. 425 00:17:14.110 --> 00:17:16.945 You need to get them first and you got to build it 426 00:17:16.945 --> 00:17:20.084 together. I think it's an important lesson here. 427 00:17:20.084 --> 00:17:21.910 Yeah, he emphasized the value first 428 00:17:21.910 --> 00:17:23.593 and then structure and that was important. 429 00:17:23.593 --> 00:17:25.610 There was an element too of, 430 00:17:25.610 --> 00:17:28.570 of a weakest link thinking that came through that, 431 00:17:28.570 --> 00:17:30.990 he talked about lots of different places 432 00:17:30.990 --> 00:17:32.770 that they were using artificial intelligence, 433 00:17:32.770 --> 00:17:34.620 and he didn't actually use this phrase, 434 00:17:34.620 --> 00:17:37.070 but part of the idea was that 435 00:17:37.070 --> 00:17:39.258 it doesn't do any good to strengthen one area extensively 436 00:17:39.258 --> 00:17:40.980 but not another one. 437 00:17:40.980 --> 00:17:43.660 And so, you know, that kind of speaks to his, 438 00:17:43.660 --> 00:17:46.870 the tension with value versus structure, 439 00:17:46.870 --> 00:17:49.420 he wanted to kind of the same time progressing. 440 00:17:49.420 --> 00:17:51.005 He wanted different parts of the business 441 00:17:51.005 --> 00:17:53.130 to be progressing at the same time as well, 442 00:17:53.130 --> 00:17:55.540 not too deep in one area, 443 00:17:55.540 --> 00:17:57.600 not just completely shallow everywhere either 444 00:17:57.600 --> 00:18:00.057 and so, almost everything seemed to be about a balance. 445 00:18:00.057 --> 00:18:01.787 You don't start by saying, 446 00:18:01.787 --> 00:18:03.240 "I've got to move everybody 447 00:18:03.240 --> 00:18:05.660 all into one center of excellence. 448 00:18:05.660 --> 00:18:08.280 Then I'm going to go build this baseball field, 449 00:18:08.280 --> 00:18:11.738 and then everybody will come and play," you know. 450 00:18:11.738 --> 00:18:13.950 Right, the other part of that was that he said, 451 00:18:13.950 --> 00:18:15.860 they didn't have AI use cases. 452 00:18:15.860 --> 00:18:17.440 And that sort of structure first 453 00:18:17.440 --> 00:18:21.510 would lend you towards a thinking of AI use cases first, 454 00:18:21.510 --> 00:18:22.950 he said, "Not AI use cases." 455 00:18:22.950 --> 00:18:26.410 He said, "Business challenges that we sometimes 456 00:18:26.410 --> 00:18:27.514 and often solve with AI." 457 00:18:27.514 --> 00:18:28.988 That's right, that's right. 458 00:18:28.988 --> 00:18:30.930 Well, that's all the time we have for today. 459 00:18:30.930 --> 00:18:34.007 Join us next time with our last episode for this season. 460 00:18:34.007 --> 00:18:36.182 We'll be talking with Kay Firth-Butterfield, 461 00:18:36.182 --> 00:18:38.130 from the World Economic Forum. 462 00:18:38.130 --> 00:18:39.252 See you next time Shervin. 463 00:18:39.252 --> 00:18:40.567 You too. 464 00:18:40.567 --> 00:18:42.985 (ethereal music) 465 00:18:42.985 --> 00:18:46.120 Thanks for listening to Me, Myself, and AI. 466 00:18:46.120 --> 00:18:47.730 If you're enjoying the show, 467 00:18:47.730 --> 00:18:49.870 take a minute to write us a review. 468 00:18:49.870 --> 00:18:51.470 If you send us a screenshot, 469 00:18:51.470 --> 00:18:53.980 we'll send you a collection of MIT SMR's 470 00:18:53.980 --> 00:18:56.420 best articles on artificial intelligence, 471 00:18:56.420 --> 00:18:58.310 free for a limited time. 472 00:18:58.310 --> 00:19:02.823 Send your review screenshot to smrfeedback@mit.edu. 473 00:19:02.823 --> 00:19:05.573 (ethereal music)