WEBVTT 1 00:00:02.940 --> 00:00:05.340 Things like brake fluid and chemical manufacturing 2 00:00:05.340 --> 00:00:08.200 may not seem like gee-whiz artificial intelligence, 3 00:00:08.200 --> 00:00:10.680 but we all may be benefiting from AI already 4 00:00:10.680 --> 00:00:11.733 and just not know it. 5 00:00:12.610 --> 00:00:14.450 Today, were talking with Chris Couch, 6 00:00:14.450 --> 00:00:16.750 senior vice president and chief technology officer 7 00:00:16.750 --> 00:00:17.930 of Cooper Standard, 8 00:00:17.930 --> 00:00:20.250 about how were all benefiting indirectly 9 00:00:20.250 --> 00:00:22.200 from artificial intelligence every day. 10 00:00:23.310 --> 00:00:25.320 Welcome to Me, Myself, and AI, 11 00:00:25.320 --> 00:00:28.480 a podcast on artificial intelligence in business. 12 00:00:28.480 --> 00:00:30.490 Each episode, we introduce you to someone 13 00:00:30.490 --> 00:00:32.360 innovating with AI. 14 00:00:32.360 --> 00:00:33.680 I'm Sam Ransbotham, 15 00:00:33.680 --> 00:00:36.587 professor of information systems at Boston College. 16 00:00:36.587 --> 00:00:37.979 I'm also the guest editor 17 00:00:37.979 --> 00:00:41.210 for the AI and Business Strategy Big Ideas program 18 00:00:41.210 --> 00:00:43.416 at MIT Sloan Management Review. 19 00:00:43.416 --> 00:00:46.150 And I'm Shervin Khodabandeh, 20 00:00:46.150 --> 00:00:47.860 senior partner with BCG, 21 00:00:47.860 --> 00:00:51.910 and I colead BCG's AI practice in North America. 22 00:00:51.910 --> 00:00:54.590 Together MIT, SMR, and BCG 23 00:00:54.590 --> 00:00:57.200 have been researching AI for five years, 24 00:00:57.200 --> 00:00:59.360 interviewing hundreds of practitioners 25 00:00:59.360 --> 00:01:01.400 and surveying thousands of companies 26 00:01:01.400 --> 00:01:02.738 on what it takes to build 27 00:01:02.738 --> 00:01:06.010 and to deploy and scale AI capabilities 28 00:01:06.010 --> 00:01:07.380 across the organization. 29 00:01:07.380 --> 00:01:09.930 And really transform the way organizations operate. 30 00:01:14.600 --> 00:01:16.290 Today were talking with Chris Couch. 31 00:01:16.290 --> 00:01:18.920 Chris is the SVP and the chief technology officer 32 00:01:18.920 --> 00:01:20.210 for Cooper Standard. 33 00:01:20.210 --> 00:01:21.880 Chris, thanks for taking the time to talk with us. 34 00:01:21.880 --> 00:01:22.713 Welcome. 35 00:01:22.713 --> 00:01:23.990 You bet. Thank you very much. 36 00:01:23.990 --> 00:01:24.890 Why don't we start by 37 00:01:24.890 --> 00:01:27.330 learning a little bit about your role at Cooper Standard. 38 00:01:27.330 --> 00:01:28.830 What do you do now? 39 00:01:28.830 --> 00:01:30.770 I am the CTO of Cooper Standard. 40 00:01:30.770 --> 00:01:33.400 We're a tier-one global automotive supplier. 41 00:01:33.400 --> 00:01:36.250 I'm also the founder and CEO of an AI startup 42 00:01:36.250 --> 00:01:38.300 called Liveline Technologies 43 00:01:38.300 --> 00:01:40.810 that has come out of some work that we did 44 00:01:40.810 --> 00:01:42.920 as R&D within Cooper Standard. 45 00:01:42.920 --> 00:01:46.560 We provide components in the spaces of 46 00:01:46.560 --> 00:01:48.280 vehicle sealing and enclosures, 47 00:01:48.280 --> 00:01:50.350 as well as fluid handling, 48 00:01:50.350 --> 00:01:52.540 whether it's brake fluid or coolant, 49 00:01:52.540 --> 00:01:54.840 all the fluid systems in the vehicle. 50 00:01:54.840 --> 00:01:58.200 We also invest in material science technologies 51 00:01:58.200 --> 00:01:59.680 that we believe can have an impact 52 00:01:59.680 --> 00:02:01.380 beyond automotive. 53 00:02:01.380 --> 00:02:03.460 Many of our products may be not visible 54 00:02:03.460 --> 00:02:04.550 to the average consumer. 55 00:02:04.550 --> 00:02:06.520 In fact, some of our products, 56 00:02:06.520 --> 00:02:08.700 hopefully nobody has to worry about them 57 00:02:08.700 --> 00:02:10.771 when were moving fuel around your vehicle, 58 00:02:10.771 --> 00:02:14.330 but they're critically important to the driving experience 59 00:02:14.330 --> 00:02:17.040 and having a safe and reliable vehicle. 60 00:02:17.040 --> 00:02:18.160 For example, 61 00:02:18.160 --> 00:02:20.870 we developed a brand-new 62 00:02:20.870 --> 00:02:22.800 category of polymer 63 00:02:22.800 --> 00:02:25.063 that we have called Fortrex. 64 00:02:26.500 --> 00:02:27.712 Fortrex provides a much 65 00:02:27.712 --> 00:02:29.230 better seal 66 00:02:29.230 --> 00:02:31.400 around the doors and windows in your vehicle. 67 00:02:31.400 --> 00:02:32.460 Why is that important? 68 00:02:32.460 --> 00:02:33.920 It's important especially 69 00:02:33.920 --> 00:02:37.500 as we move into an electrified-vehicle world. 70 00:02:37.500 --> 00:02:39.711 As engine and transmission noises 71 00:02:39.711 --> 00:02:43.360 decrease, because there's no more gasoline engine. 72 00:02:43.360 --> 00:02:45.750 Other sources of noise become more prevalent, 73 00:02:45.750 --> 00:02:46.760 and the largest one of those 74 00:02:46.760 --> 00:02:47.898 is the noise coming in 75 00:02:47.898 --> 00:02:50.971 due to wind around your doors and windows. 76 00:02:50.971 --> 00:02:54.260 By providing an enhanced sealing package for those, 77 00:02:54.260 --> 00:02:56.840 we believe we've got the right products 78 00:02:56.840 --> 00:02:59.650 to service an electrifying world. 79 00:02:59.650 --> 00:03:02.110 How is artificial intelligence involved in that 80 00:03:02.110 --> 00:03:03.850 development of the polymer? 81 00:03:03.850 --> 00:03:05.550 We spent a lot of time and money 82 00:03:05.550 --> 00:03:07.710 coming up with advanced polymer formulations. 83 00:03:07.710 --> 00:03:10.490 A lot of it historically has been trial and error. 84 00:03:10.490 --> 00:03:13.433 That's what industrial chemists often do. 85 00:03:14.328 --> 00:03:18.400 We used AI to develop a system 86 00:03:18.400 --> 00:03:20.870 that advises our chemists on 87 00:03:20.870 --> 00:03:23.680 the next set of recipes to try 88 00:03:23.680 --> 00:03:27.000 as they iterate toward a final solution. 89 00:03:27.000 --> 00:03:29.058 We've found dramatic reductions, 90 00:03:29.058 --> 00:03:31.746 in many cases, with that approach. 91 00:03:31.746 --> 00:03:33.250 Dramatic means reducing 92 00:03:33.250 --> 00:03:36.440 those R&D loops 70% or 80%. 93 00:03:36.440 --> 00:03:37.273 Got it, 94 00:03:37.273 --> 00:03:39.290 but before we talk more about Cooper Standard's 95 00:03:39.290 --> 00:03:40.490 success with AI, 96 00:03:40.490 --> 00:03:41.450 can you tell us a bit more about 97 00:03:41.450 --> 00:03:43.690 your own background and career path? 98 00:03:43.690 --> 00:03:44.900 Well, I think 99 00:03:44.900 --> 00:03:46.250 the best way to describe myself 100 00:03:46.250 --> 00:03:48.430 is a lifelong manufacturing addict, 101 00:03:48.430 --> 00:03:49.263 first of all. 102 00:03:49.263 --> 00:03:52.160 As a kid, I took apart everything in the house 103 00:03:52.160 --> 00:03:54.400 probably got hit with wall current more than once 104 00:03:54.400 --> 00:03:56.450 Explains a lot about me today, I suppose. 105 00:03:57.370 --> 00:04:00.250 I was the kid that built their first car out of a kit. 106 00:04:00.250 --> 00:04:02.490 I was a hard-core mechanical engineer 107 00:04:02.490 --> 00:04:05.990 with a focus on manufacturing and controls in school. 108 00:04:05.990 --> 00:04:07.360 My side projects include things 109 00:04:07.360 --> 00:04:10.110 like building autonomous drones that 110 00:04:10.110 --> 00:04:11.400 fly at high altitude. 111 00:04:11.400 --> 00:04:14.470 I'm just a manufacturing nerd. 112 00:04:14.470 --> 00:04:17.670 That has really served me well in my career. 113 00:04:17.670 --> 00:04:21.110 I spent the first third of my working life 114 00:04:21.110 --> 00:04:22.390 in a Japanese company. 115 00:04:22.390 --> 00:04:24.100 I worked for Toyota 116 00:04:24.100 --> 00:04:27.160 and went and joined them in Japan. 117 00:04:27.160 --> 00:04:29.200 Spent a dozen years with them 118 00:04:29.200 --> 00:04:31.150 designing and building and 119 00:04:31.150 --> 00:04:34.140 ultimately being involved in plant operations. 120 00:04:34.140 --> 00:04:36.340 I spent the next third of my career 121 00:04:36.340 --> 00:04:38.040 running a P&L 122 00:04:38.040 --> 00:04:39.690 for an automotive supplier 123 00:04:39.690 --> 00:04:41.290 on the business side, 124 00:04:41.290 --> 00:04:42.850 mostly based out of Asia, 125 00:04:42.850 --> 00:04:45.113 so I have a business bent as well, 126 00:04:45.113 --> 00:04:47.610 which may color a lot of what I say today. 127 00:04:47.610 --> 00:04:50.420 Then, the last third or so of my career has been 128 00:04:50.420 --> 00:04:51.510 in CTO gigs. 129 00:04:51.510 --> 00:04:53.200 I'm in my second one here. 130 00:04:53.200 --> 00:04:55.413 I'm again at an automotive supplier, 131 00:04:55.413 --> 00:04:58.980 but we get our fingers into all kinds of interesting 132 00:04:58.980 --> 00:05:01.570 tech domains, just given what's happening 133 00:05:01.570 --> 00:05:02.700 in the world today, 134 00:05:02.700 --> 00:05:05.580 whether it's material science or AI. 135 00:05:05.580 --> 00:05:07.169 So here I am. 136 00:05:07.169 --> 00:05:09.750 Didn't really expect to be doing 137 00:05:09.750 --> 00:05:10.830 my second job here, 138 00:05:10.830 --> 00:05:12.240 if you would have asked me two years ago, 139 00:05:12.240 --> 00:05:14.020 but it's certainly been 140 00:05:14.020 --> 00:05:14.853 a lot of fun, 141 00:05:14.853 --> 00:05:16.810 and were excited about delivering some impact 142 00:05:16.810 --> 00:05:18.317 with these technologies. 143 00:05:18.317 --> 00:05:21.330 Chris, tell us about the open innovation 144 00:05:21.330 --> 00:05:22.280 at Cooper Standard. 145 00:05:22.280 --> 00:05:24.100 What is that all about? 146 00:05:24.100 --> 00:05:24.933 You know, 147 00:05:24.933 --> 00:05:26.530 as we looked around at our tech portfolio 148 00:05:26.530 --> 00:05:29.760 a couple years ago when I joined the company. 149 00:05:29.760 --> 00:05:31.430 I was, first of all, overwhelmed 150 00:05:31.430 --> 00:05:33.620 by the different domains that we really 151 00:05:33.620 --> 00:05:35.390 had to compete in. 152 00:05:35.390 --> 00:05:37.470 I mentioned materials science earlier, 153 00:05:37.470 --> 00:05:38.890 but there's different aspects 154 00:05:38.890 --> 00:05:41.872 of manufacturing technology and product design. 155 00:05:41.872 --> 00:05:44.160 The whole topic of analytics and AI 156 00:05:44.160 --> 00:05:46.270 right, that were going to talk about today. 157 00:05:46.270 --> 00:05:48.460 And was very convinced 158 00:05:48.460 --> 00:05:50.560 that there was no way that we could do it all ourselves. 159 00:05:50.560 --> 00:05:52.840 Cooper Standard isn't a small company. 160 00:05:52.840 --> 00:05:55.690 We're just shy of three billion in revenue, 161 00:05:55.690 --> 00:05:57.903 but were not the biggest. 162 00:05:57.903 --> 00:05:59.410 And so open innovation 163 00:05:59.410 --> 00:06:01.030 was really an attempt to reach out 164 00:06:01.030 --> 00:06:02.440 and build a pipeline 165 00:06:02.440 --> 00:06:04.639 to draw ideas and technology, 166 00:06:04.639 --> 00:06:06.389 and maybe even talent, 167 00:06:06.389 --> 00:06:08.060 from the outside world. 168 00:06:08.060 --> 00:06:11.360 And so through that, we engage with 169 00:06:11.360 --> 00:06:13.790 universities, with consortia 170 00:06:13.790 --> 00:06:14.890 around the world. 171 00:06:14.890 --> 00:06:17.770 We engage heavily with startup companies 172 00:06:17.770 --> 00:06:21.090 and use that as a source of ideas. 173 00:06:21.090 --> 00:06:23.170 In fact, our first 174 00:06:23.170 --> 00:06:26.700 proper AI project, if you will, really came through that 175 00:06:26.700 --> 00:06:29.360 open-innovation pipeline. 176 00:06:29.360 --> 00:06:31.810 We partnered up with a brand-new startup 177 00:06:31.810 --> 00:06:33.620 that was called Uncountable, 178 00:06:33.620 --> 00:06:35.240 out of the Bay Area, 179 00:06:35.240 --> 00:06:37.720 and they helped us develop a system 180 00:06:37.720 --> 00:06:41.280 that would serve effectively as an adviser 181 00:06:41.280 --> 00:06:43.630 for our chemists that make new formulations 182 00:06:43.630 --> 00:06:45.220 for materials that we use all the time. 183 00:06:45.220 --> 00:06:49.300 That wound up being a great accelerator for our R&D process, 184 00:06:49.300 --> 00:06:51.090 cutting iterations out of those 185 00:06:51.090 --> 00:06:53.743 design-and-test loops, if you will. 186 00:06:53.743 --> 00:06:56.200 That was one of those big, aha! moments, right? 187 00:06:56.200 --> 00:06:58.820 That there is a huge 188 00:06:58.820 --> 00:07:00.490 potential to 189 00:07:00.490 --> 00:07:02.470 accelerate ourselves in many domains. 190 00:07:02.470 --> 00:07:04.144 We can't do it all ourselves, 191 00:07:04.144 --> 00:07:08.236 and so how do we really build that external pipeline? 192 00:07:08.236 --> 00:07:11.410 We now call it CS Open Innovation, 193 00:07:11.410 --> 00:07:12.988 but that was the impetus. 194 00:07:12.988 --> 00:07:16.380 Sounds like a very, sort of 195 00:07:16.380 --> 00:07:18.003 unique way of bringing 196 00:07:18.003 --> 00:07:20.240 folks with different backgrounds 197 00:07:20.240 --> 00:07:21.520 and different talents 198 00:07:21.520 --> 00:07:23.650 and getting them all work together. 199 00:07:23.650 --> 00:07:25.020 What did you find 200 00:07:25.020 --> 00:07:27.770 was the secret sauce of making that happen? 201 00:07:27.770 --> 00:07:28.603 I think 202 00:07:28.603 --> 00:07:30.640 whether it's AI, whether it's materials science, 203 00:07:30.640 --> 00:07:31.870 whether it's other domains, 204 00:07:31.870 --> 00:07:33.130 my answer is the same: 205 00:07:33.130 --> 00:07:36.398 It really is all about the ability to focus. 206 00:07:36.398 --> 00:07:39.980 The reason that we, as many other companies, 207 00:07:39.980 --> 00:07:41.380 have put in place 208 00:07:41.380 --> 00:07:43.620 innovation pipelines and processes 209 00:07:43.620 --> 00:07:45.687 and stage-gate processes 210 00:07:45.687 --> 00:07:49.030 that govern innovation is because of the focus. 211 00:07:49.030 --> 00:07:50.950 How do we quickly narrow down 212 00:07:50.950 --> 00:07:53.360 where were going to allocate our precious 213 00:07:53.360 --> 00:07:54.320 R&D dollars, 214 00:07:54.320 --> 00:07:56.370 and how do we govern 215 00:07:56.370 --> 00:07:57.340 those correctly? 216 00:07:57.340 --> 00:07:59.050 So we think like a startup. 217 00:07:59.050 --> 00:08:01.140 We're doing the minimal 218 00:08:01.140 --> 00:08:03.210 investment to sort of answer the next, 219 00:08:03.210 --> 00:08:05.100 most important question 220 00:08:05.100 --> 00:08:07.240 and either wind up killing things quickly 221 00:08:07.240 --> 00:08:09.620 or take them to fruition. 222 00:08:09.620 --> 00:08:12.180 And a fair amount of fail fast, and test 223 00:08:12.180 --> 00:08:13.013 and learn, 224 00:08:13.013 --> 00:08:16.330 and sort of go big behind things that are working 225 00:08:16.330 --> 00:08:18.430 and shut down things that aren't, right? 226 00:08:18.430 --> 00:08:19.609 Did I hear that correctly? 227 00:08:19.609 --> 00:08:20.442 Exactly, 228 00:08:20.442 --> 00:08:21.324 and that's not unique. 229 00:08:21.324 --> 00:08:22.950 I think that 230 00:08:22.950 --> 00:08:25.830 there's nothing special about AI-based projects, right? 231 00:08:25.830 --> 00:08:27.960 We sort of think in the same way 232 00:08:27.960 --> 00:08:30.540 and very quickly try to 233 00:08:30.540 --> 00:08:32.220 motivate those with 234 00:08:32.220 --> 00:08:34.123 a clear-eyed view of ROI. 235 00:08:35.420 --> 00:08:37.570 Frankly, one of the things 236 00:08:37.570 --> 00:08:39.560 that I think we've seen over the years 237 00:08:39.560 --> 00:08:42.790 when it comes to analytics, AI. 238 00:08:42.790 --> 00:08:45.940 Especially coupled with manufacturing and Industry 4.0, 239 00:08:45.940 --> 00:08:50.154 ROI has sometimes been hard to come by. 240 00:08:50.154 --> 00:08:52.060 A lot of creative ideas, 241 00:08:52.060 --> 00:08:54.587 a lot of interesting things to do with data, 242 00:08:54.587 --> 00:08:55.720 but the question is, 243 00:08:55.720 --> 00:08:57.500 how does it translate to the bottom line? 244 00:08:57.500 --> 00:08:58.610 And if that 245 00:08:58.610 --> 00:09:01.830 story can't be told, even as a hypothesis 246 00:09:01.830 --> 00:09:03.830 that were going to prove through the 247 00:09:03.830 --> 00:09:05.750 innovation project, then it's 248 00:09:05.750 --> 00:09:07.883 hard to justify working on it. 249 00:09:08.720 --> 00:09:09.990 It seems like the opposite, 250 00:09:09.990 --> 00:09:11.050 though, is that, 251 00:09:11.050 --> 00:09:12.910 just to push back a little bit, 252 00:09:12.910 --> 00:09:15.080 if you get too focused on ROI, 253 00:09:15.080 --> 00:09:16.210 where are you going to get something 254 00:09:16.210 --> 00:09:17.930 weird and big and unusual? 255 00:09:17.930 --> 00:09:19.030 Absolutely. 256 00:09:19.030 --> 00:09:21.160 How are you balancing that 257 00:09:21.160 --> 00:09:24.650 tension between focusing on ROI and also 258 00:09:24.650 --> 00:09:26.150 trying not to miss out on, 259 00:09:26.150 --> 00:09:28.230 or trying not to be too incremental? 260 00:09:28.230 --> 00:09:30.540 I think the stage-gate mentality is useful here. 261 00:09:30.540 --> 00:09:31.740 I think in the early state, 262 00:09:31.740 --> 00:09:33.370 we look at a lot of crazy stuff. 263 00:09:33.370 --> 00:09:36.430 We have crazy ideas that come in through open innovation. 264 00:09:36.430 --> 00:09:38.080 We have crazy ideas from our own 265 00:09:38.080 --> 00:09:40.220 teams, and that's fantastic. 266 00:09:40.220 --> 00:09:42.070 We don't hesitate to look at them 267 00:09:42.070 --> 00:09:44.300 and maybe even 268 00:09:44.300 --> 00:09:46.740 spend a little pocket money to chase them down 269 00:09:46.740 --> 00:09:47.573 to some degree. 270 00:09:47.573 --> 00:09:49.140 The question then is, 271 00:09:49.140 --> 00:09:52.260 what are we going to invest in to try to productize? 272 00:09:52.260 --> 00:09:54.660 That's really the next gate, if you will. 273 00:09:54.660 --> 00:09:55.560 So absolutely, 274 00:09:55.560 --> 00:09:57.520 the exploration is important. 275 00:09:57.520 --> 00:09:59.460 We certainly do some of that. 276 00:09:59.460 --> 00:10:00.780 I hesitate to say it almost, 277 00:10:00.780 --> 00:10:03.090 but it's having some space to play 278 00:10:03.090 --> 00:10:05.260 with ideas and technologies, 279 00:10:05.260 --> 00:10:07.430 but then, when it's time to go productize, 280 00:10:07.430 --> 00:10:09.620 right, you have to be clear-eyed on what 281 00:10:09.620 --> 00:10:11.360 you're going to get out of it. 282 00:10:11.360 --> 00:10:13.230 That seems like something that might differ for 283 00:10:13.230 --> 00:10:14.063 an AI approach. 284 00:10:14.063 --> 00:10:15.200 I mean, you said, Well, 285 00:10:15.200 --> 00:10:17.280 AI's no different, just a second ago, 286 00:10:17.280 --> 00:10:21.250 but it seems like, I guess I wonder if there is something 287 00:10:21.250 --> 00:10:22.230 different about 288 00:10:22.230 --> 00:10:24.970 these new technologies that may require 289 00:10:24.970 --> 00:10:26.660 a little more freedom up front 290 00:10:26.660 --> 00:10:28.260 to do something weird 291 00:10:28.260 --> 00:10:30.477 than perhaps some others. 292 00:10:30.477 --> 00:10:31.970 I think that's fair. 293 00:10:31.970 --> 00:10:32.803 In our experience, 294 00:10:32.803 --> 00:10:35.210 I think one of the differences with AI is that 295 00:10:35.210 --> 00:10:36.900 you probably have less 296 00:10:36.900 --> 00:10:41.290 familiarity with the tools and the applications among 297 00:10:41.290 --> 00:10:42.730 the general technical population. 298 00:10:42.730 --> 00:10:44.650 Right, so if you're talking to design engineers, 299 00:10:44.650 --> 00:10:47.300 or talking to manufacturing process engineers, 300 00:10:47.300 --> 00:10:48.750 they may have read some things 301 00:10:48.750 --> 00:10:51.380 and maybe seen an interesting demo somewhere 302 00:10:51.380 --> 00:10:52.810 but may not be so versed 303 00:10:52.810 --> 00:10:55.560 in the nuts and bolts of how it works, 304 00:10:55.560 --> 00:10:56.940 much less the nuts and bolts 305 00:10:56.940 --> 00:10:58.640 of what would it take to scale that 306 00:10:58.640 --> 00:11:00.280 at an enterprise level. 307 00:11:00.280 --> 00:11:02.640 Because getting models running in a Jupyter Notebook 308 00:11:02.640 --> 00:11:03.994 off of a 309 00:11:03.994 --> 00:11:05.450 CSV file on your hard drive 310 00:11:05.450 --> 00:11:07.230 is a whole different story from 311 00:11:07.230 --> 00:11:08.676 production on a global scale. 312 00:11:08.676 --> 00:11:09.509 Exactly. 313 00:11:09.509 --> 00:11:11.510 And so I think just that lack 314 00:11:11.510 --> 00:11:14.690 of exposure to the technologies makes it a bit different. 315 00:11:14.690 --> 00:11:16.160 If were talking about 316 00:11:16.160 --> 00:11:18.120 traditional robotics, or 317 00:11:18.120 --> 00:11:21.250 maybe simpler types of IOT concepts, 318 00:11:21.250 --> 00:11:22.930 plenty of engineers have a good clue 319 00:11:22.930 --> 00:11:25.110 and maybe have used some things in their career, 320 00:11:25.110 --> 00:11:27.730 but much less so when it comes to AI. 321 00:11:27.730 --> 00:11:29.260 That is a difference, I would agree. 322 00:11:29.260 --> 00:11:31.010 The good news is, 323 00:11:31.010 --> 00:11:33.630 I am very convinced that 324 00:11:33.630 --> 00:11:35.860 one of the wonderful things about AI 325 00:11:35.860 --> 00:11:37.210 is that it is 326 00:11:37.210 --> 00:11:39.040 cheap to pilot. 327 00:11:39.040 --> 00:11:40.750 I was just sort of 328 00:11:40.750 --> 00:11:44.530 making up a silly example about Jupyter Notebooks and 329 00:11:44.530 --> 00:11:45.440 CSV files, 330 00:11:45.440 --> 00:11:47.810 but that's a great way to explore some concepts, 331 00:11:47.810 --> 00:11:49.460 and the cost of that 332 00:11:49.460 --> 00:11:51.480 is very close to zero, 333 00:11:51.480 --> 00:11:53.844 other than acquiring the knowledge 334 00:11:53.844 --> 00:11:54.677 to do it. 335 00:11:54.677 --> 00:11:58.200 And even then, I think that we've proven 336 00:11:58.200 --> 00:11:59.930 over and over in our internal teams 337 00:11:59.930 --> 00:12:02.750 that even the knowledge acquisition 338 00:12:02.750 --> 00:12:05.710 is reasonably priced, if you will. 339 00:12:05.710 --> 00:12:08.070 Chris, I want to build on that point you said, 340 00:12:08.070 --> 00:12:09.990 that AI is 341 00:12:09.990 --> 00:12:12.150 relatively inexpensive to pilot, 342 00:12:12.150 --> 00:12:14.170 and I agree with that because 343 00:12:14.170 --> 00:12:16.220 we see, of course, 344 00:12:16.220 --> 00:12:17.800 a proliferation of 345 00:12:17.800 --> 00:12:19.800 proofs of concept and 346 00:12:19.800 --> 00:12:21.500 different teams trying different approaches, 347 00:12:21.500 --> 00:12:22.870 different ideas. 348 00:12:22.870 --> 00:12:25.260 It also seems there is a fact 349 00:12:25.260 --> 00:12:28.020 that AI is also quite hard to scale. 350 00:12:28.020 --> 00:12:28.853 Right. 351 00:12:28.853 --> 00:12:31.380 And so why not sort of get your 352 00:12:31.380 --> 00:12:33.210 reactions to, 353 00:12:33.210 --> 00:12:36.130 something's easy to pilot, 354 00:12:36.130 --> 00:12:38.120 quite hard to scale. 355 00:12:38.120 --> 00:12:40.790 the real meaningful ROI will come 356 00:12:40.790 --> 00:12:42.470 after you scale it. 357 00:12:42.470 --> 00:12:44.910 So how do you make that transition? 358 00:12:44.910 --> 00:12:47.560 And how do you sort of make 359 00:12:47.560 --> 00:12:49.500 things that are really easy to pilot 360 00:12:49.500 --> 00:12:51.220 and get excitement around 361 00:12:51.220 --> 00:12:54.290 but then harder down the line to actually 362 00:12:54.290 --> 00:12:57.640 embed into business processes and ways of working? 363 00:12:57.640 --> 00:12:59.933 How do you envision that transition working? 364 00:13:00.770 --> 00:13:02.630 Right. Yeah, it's a great question, 365 00:13:02.630 --> 00:13:04.530 and it's definitely not 366 00:13:04.530 --> 00:13:07.080 easy and maybe not for the faint of heart, right? 367 00:13:07.080 --> 00:13:08.850 Because sometimes it does 368 00:13:08.850 --> 00:13:10.740 take a leap of faith 369 00:13:10.740 --> 00:13:13.290 in the ability to scale, ultimately. 370 00:13:13.290 --> 00:13:16.290 The best I can say from our experience with Liveline, 371 00:13:16.290 --> 00:13:19.030 We did some very early prototyping, 372 00:13:19.030 --> 00:13:20.660 we thought we understood 373 00:13:20.660 --> 00:13:22.300 the data science aspect, 374 00:13:22.300 --> 00:13:24.030 but that was only the beginning. 375 00:13:24.030 --> 00:13:26.113 That was nearly two years ago. 376 00:13:27.480 --> 00:13:30.260 Only in the past months have we begun 377 00:13:30.260 --> 00:13:32.510 to go to a global scale-out. 378 00:13:32.510 --> 00:13:34.710 The only insight I have there is, 379 00:13:34.710 --> 00:13:37.040 as you prototype, as you pilot, 380 00:13:37.040 --> 00:13:39.620 you've just got to try to be as judicious 381 00:13:39.620 --> 00:13:41.080 as you can 382 00:13:41.080 --> 00:13:45.050 about selecting use cases that are realistic 383 00:13:45.050 --> 00:13:47.140 and that everybody can get their 384 00:13:47.140 --> 00:13:49.890 heads around and connect the dots to the ROI 385 00:13:49.890 --> 00:13:51.730 at the end of the day. 386 00:13:51.730 --> 00:13:53.560 How are you getting people, 387 00:13:53.560 --> 00:13:55.520 you know once you've got these solutions in place, 388 00:13:55.520 --> 00:13:57.620 what about the adoption within the organization? 389 00:13:57.620 --> 00:14:00.520 How are you getting people to work on teams that 390 00:14:00.520 --> 00:14:02.050 used to have human partners 391 00:14:02.050 --> 00:14:04.260 and now have machine partners? 392 00:14:04.260 --> 00:14:07.140 With Liveline, the basic concept of Liveline 393 00:14:07.140 --> 00:14:10.110 is to automate the creation of automation 394 00:14:10.110 --> 00:14:12.710 for complex manufacturing environments. 395 00:14:12.710 --> 00:14:14.990 We're using machine learning techniques to design 396 00:14:14.990 --> 00:14:16.580 the control policies 397 00:14:16.580 --> 00:14:18.170 that we deploy 398 00:14:18.170 --> 00:14:20.750 onto the lines to control machine parameters 399 00:14:20.750 --> 00:14:21.753 in real time. 400 00:14:22.790 --> 00:14:24.320 We think this is very useful for 401 00:14:24.320 --> 00:14:27.760 attacking a diverse range of processes that have been 402 00:14:27.760 --> 00:14:30.690 too complex or too costly to automate. 403 00:14:30.690 --> 00:14:32.840 otherwise, and our early successes 404 00:14:32.840 --> 00:14:34.900 have been in continuous-flow 405 00:14:34.900 --> 00:14:36.870 manufacturing processes, 406 00:14:36.870 --> 00:14:39.300 chemical conversion, polymer extrusion, 407 00:14:39.300 --> 00:14:41.780 and we think there's a broad applicability to this 408 00:14:41.780 --> 00:14:45.460 to areas like oil and gas, wire and cable, etc. 409 00:14:45.460 --> 00:14:47.360 One of my fears when we 410 00:14:47.360 --> 00:14:51.440 first got into plants to do live production trials 411 00:14:51.440 --> 00:14:54.130 is that the plant personnel might view this as 412 00:14:54.130 --> 00:14:56.730 sort of a threat, right were automating. 413 00:14:56.730 --> 00:14:58.680 that has some negative connotations 414 00:14:58.680 --> 00:15:02.040 sometimes in terms of impacts on people's jobs 415 00:15:02.040 --> 00:15:03.460 and so forth. 416 00:15:03.460 --> 00:15:06.580 But there's a couple of things I think really 417 00:15:06.580 --> 00:15:07.890 gained us some traction, 418 00:15:07.890 --> 00:15:10.280 and the reception has been quite warm. 419 00:15:10.280 --> 00:15:12.920 In fact, the plants are pulling very hard now 420 00:15:12.920 --> 00:15:14.310 to roll this out. 421 00:15:14.310 --> 00:15:18.220 My attitude is to really democratize the information 422 00:15:18.220 --> 00:15:20.253 and what's happening with the tool. 423 00:15:21.678 --> 00:15:23.700 For example, we spent 424 00:15:23.700 --> 00:15:25.640 quite some effort to make sure that 425 00:15:25.640 --> 00:15:27.640 operators in the plant environment 426 00:15:27.640 --> 00:15:29.620 had screens where they can 427 00:15:29.620 --> 00:15:32.320 see data streams in real time, 428 00:15:32.320 --> 00:15:34.340 that they couldn't before. 429 00:15:34.340 --> 00:15:36.900 Sometimes they were data streams that we had created 430 00:15:36.900 --> 00:15:39.460 for the sake of the machine learning. 431 00:15:39.460 --> 00:15:41.510 We give them visibility into it. 432 00:15:41.510 --> 00:15:43.680 We give them visibility, if they want, 433 00:15:43.680 --> 00:15:47.120 into the decisions that the system is making. 434 00:15:47.120 --> 00:15:49.390 We also give them the ability to turn it off, 435 00:15:49.390 --> 00:15:51.610 haha, the big red button, right 436 00:15:51.610 --> 00:15:53.097 if they're not comfortable with what the 437 00:15:53.097 --> 00:15:56.380 HAL 9000's doing on their production line. 438 00:15:56.380 --> 00:15:59.920 Also, we give them the ability to bias it. 439 00:15:59.920 --> 00:16:01.410 If they feel, 440 00:16:01.410 --> 00:16:02.420 based on their experience, 441 00:16:02.420 --> 00:16:04.390 that the system is making parts 442 00:16:04.390 --> 00:16:05.343 that are a little, 443 00:16:06.500 --> 00:16:08.140 let's just say too thin or too thick, 444 00:16:08.140 --> 00:16:10.210 they can bias it down a little bit. 445 00:16:10.210 --> 00:16:12.040 I think that sort of exposure 446 00:16:12.040 --> 00:16:14.040 and opening up the black box, 447 00:16:14.040 --> 00:16:15.550 at least in the plant environment, 448 00:16:15.550 --> 00:16:17.270 is very critical to 449 00:16:17.270 --> 00:16:20.960 people buying in and believing in what's going on. 450 00:16:20.960 --> 00:16:23.570 One of our learnings at Liveline with that 451 00:16:23.570 --> 00:16:25.900 was the enhanced feedback that we get. 452 00:16:25.900 --> 00:16:29.290 We have received several very 453 00:16:29.290 --> 00:16:31.700 influential and useful ideas 454 00:16:31.700 --> 00:16:34.340 from people that were really doing nothing more 455 00:16:34.340 --> 00:16:37.040 than looking at data streams and watching 456 00:16:37.040 --> 00:16:39.360 the system making decisions. 457 00:16:39.360 --> 00:16:41.580 They asked good questions back to us 458 00:16:41.580 --> 00:16:43.440 and gave us good insights. 459 00:16:43.440 --> 00:16:44.870 Suggested 460 00:16:44.870 --> 00:16:46.300 new types of data that 461 00:16:46.300 --> 00:16:48.610 we could be tagging that might be useful 462 00:16:48.610 --> 00:16:50.880 once they really began to get a little intuition 463 00:16:50.880 --> 00:16:53.770 about what we were trying to do with the data science. 464 00:16:53.770 --> 00:16:56.360 I think that sort of democratization, 465 00:16:56.360 --> 00:16:58.140 if you will, of the system and 466 00:16:58.140 --> 00:17:00.930 opening up and exposing the guts as it 467 00:17:00.930 --> 00:17:03.590 does its thing has been, at least in this case, 468 00:17:03.590 --> 00:17:05.700 one of the success factors. 469 00:17:05.700 --> 00:17:08.600 That's a great example. It covers, 470 00:17:08.600 --> 00:17:09.433 Exactly. Haha 471 00:17:09.433 --> 00:17:10.460 Sam, I feel like it covers 472 00:17:10.460 --> 00:17:11.460 a lot of what we've talked 473 00:17:11.460 --> 00:17:13.940 about in our report, 474 00:17:13.940 --> 00:17:18.770 in terms of different modes of human-AI interaction. 475 00:17:18.770 --> 00:17:19.770 No black box. 476 00:17:19.770 --> 00:17:23.370 Allowing the human to override or bias, 477 00:17:23.370 --> 00:17:25.470 but also - I was going to ask you, 478 00:17:25.470 --> 00:17:27.180 Chris, you hit the point before 479 00:17:27.180 --> 00:17:28.570 I got a chance to ask you, 480 00:17:28.570 --> 00:17:30.070 which is the feedback loop. 481 00:17:30.070 --> 00:17:32.600 I guess my follow-on question is, 482 00:17:32.600 --> 00:17:34.890 how has that feedback loop been working 483 00:17:34.890 --> 00:17:35.920 in terms of 484 00:17:35.920 --> 00:17:37.290 maybe skeptics 485 00:17:37.290 --> 00:17:40.450 having become more sort of AI-friendly, 486 00:17:40.450 --> 00:17:42.360 or more trust having been formed 487 00:17:42.360 --> 00:17:45.540 between humans and AI? 488 00:17:45.540 --> 00:17:48.490 Any anecdotes you can comment on that? 489 00:17:48.490 --> 00:17:50.030 Absolutely. 490 00:17:50.030 --> 00:17:51.630 I'll give you a great anecdote from 491 00:17:51.630 --> 00:17:54.040 one of our plants in the southern U.S. 492 00:17:54.040 --> 00:17:56.890 In fact, this was the plant where we did our final 493 00:17:56.890 --> 00:17:58.360 pilot for Liveline 494 00:17:58.360 --> 00:18:01.220 before we made the decision as a company to go 495 00:18:01.220 --> 00:18:02.743 do a global rollout. 496 00:18:04.010 --> 00:18:06.460 We first had the line running in 497 00:18:06.460 --> 00:18:08.260 what we call automatic mode. 498 00:18:08.260 --> 00:18:11.203 Gosh, I think it was about Q3 of last year. 499 00:18:13.130 --> 00:18:15.360 One of the criteria for the pilot 500 00:18:15.360 --> 00:18:18.340 was that we would do some A-versus-B runs. 501 00:18:18.340 --> 00:18:19.920 Concept's very simple. 502 00:18:19.920 --> 00:18:21.690 For these four hours, were going to run 503 00:18:21.690 --> 00:18:24.300 with the system engaged in automatic mode. 504 00:18:24.300 --> 00:18:25.990 For these four hours, were going to turn it off, 505 00:18:25.990 --> 00:18:29.950 and you all can run the plant like you always do. 506 00:18:29.950 --> 00:18:31.780 Then, over a series of days and weeks, 507 00:18:31.780 --> 00:18:34.090 we'll add up the statistics about 508 00:18:34.090 --> 00:18:36.080 scrap rates and quality 509 00:18:36.080 --> 00:18:38.520 and unplanned line stops, 510 00:18:38.520 --> 00:18:41.923 and we will quantify exactly what the value was. 511 00:18:42.920 --> 00:18:44.500 We came to the first review point 512 00:18:44.500 --> 00:18:46.270 a few weeks into that, 513 00:18:46.270 --> 00:18:48.010 and as I sat with the team, 514 00:18:48.010 --> 00:18:49.280 they sort of 515 00:18:49.280 --> 00:18:50.890 pulled up chairs and looked at their shoes 516 00:18:50.890 --> 00:18:52.340 and said, Hey, we have an issue. 517 00:18:52.340 --> 00:18:55.103 We don't have the B data with the system off. 518 00:18:56.060 --> 00:18:57.070 I said, Why is that? 519 00:18:57.070 --> 00:18:58.680 They said, Because once the plant turned on, 520 00:18:58.680 --> 00:19:00.380 they refused to turn it off again. 521 00:19:02.520 --> 00:19:04.770 They don't want to run with the system disengaged anymore 522 00:19:04.770 --> 00:19:07.600 because the impact was so significant to them 523 00:19:07.600 --> 00:19:10.830 and helped them to operate the lines better 524 00:19:10.830 --> 00:19:12.340 that they don't want to run with it off anymore. 525 00:19:12.340 --> 00:19:14.960 That was very consistent with the type of 526 00:19:14.960 --> 00:19:16.440 reaction we saw in other 527 00:19:16.440 --> 00:19:19.800 pilots in Canada and our tech center in Michigan. 528 00:19:19.800 --> 00:19:20.640 That is great. 529 00:19:20.640 --> 00:19:23.410 Yeah, that sort of feedback is very reassuring, 530 00:19:23.410 --> 00:19:25.090 but again, I think that 531 00:19:25.090 --> 00:19:27.480 from the get-go, having a philosophy of really 532 00:19:27.480 --> 00:19:30.930 just opening up and showing people what's going on, 533 00:19:30.930 --> 00:19:32.200 letting them look at data, 534 00:19:32.200 --> 00:19:33.860 be participants in 535 00:19:33.860 --> 00:19:36.820 problem-solving and tuning and enhancement, 536 00:19:36.820 --> 00:19:39.440 really sets the stage for that emotional 537 00:19:39.440 --> 00:19:41.990 connection and commitment to the project. 538 00:19:41.990 --> 00:19:43.300 Those seem like some very different 539 00:19:43.300 --> 00:19:45.930 ways of getting feedback to a system, 540 00:19:45.930 --> 00:19:47.400 and then the other one you mentioned 541 00:19:47.400 --> 00:19:48.445 was the idea of 542 00:19:48.445 --> 00:19:52.030 suggesting new tags or new data to come back in. 543 00:19:52.030 --> 00:19:52.863 Right. 544 00:19:52.863 --> 00:19:53.880 I can see, for example, 545 00:19:53.880 --> 00:19:56.080 adjusting the bias being a real-time 546 00:19:56.080 --> 00:19:57.360 sort of feedback, and clearly, 547 00:19:57.360 --> 00:19:59.460 pressing the red button would happen immediately, 548 00:19:59.460 --> 00:20:02.073 I hope. That's the point of a red button. 549 00:20:03.070 --> 00:20:06.300 How do the processes work for some of these 550 00:20:06.300 --> 00:20:07.860 non-immediate feedback? 551 00:20:07.860 --> 00:20:09.237 What do you do with the suggestions 552 00:20:09.237 --> 00:20:11.090 for new data and tags? 553 00:20:11.090 --> 00:20:12.800 Is there a process around those? 554 00:20:12.800 --> 00:20:13.760 This is, by the way, 555 00:20:13.760 --> 00:20:16.400 sorry to interrupt your response, 556 00:20:16.400 --> 00:20:19.370 this is Sam's and to some extent, 557 00:20:19.370 --> 00:20:22.369 my chemical engineering background coming out. 558 00:20:22.369 --> 00:20:23.710 And you can think of, 559 00:20:23.710 --> 00:20:24.910 at least for Cooper Standard, 560 00:20:24.910 --> 00:20:27.450 the majority of our lines are chemical processing lines. 561 00:20:27.450 --> 00:20:28.730 We're taking 562 00:20:28.730 --> 00:20:30.060 different types of compounds, 563 00:20:30.060 --> 00:20:32.030 and were extruding them and, 564 00:20:32.030 --> 00:20:33.630 in the case of thermosets, 565 00:20:33.630 --> 00:20:35.790 putting them through oven stages, 566 00:20:35.790 --> 00:20:37.357 200 meters of processes, 567 00:20:37.357 --> 00:20:39.200 and a lot of it is chemistry, as it goes. 568 00:20:39.200 --> 00:20:42.043 Yeah, so you guys are in your sweet spot. 569 00:20:43.379 --> 00:20:45.750 What's the process for new data tags? 570 00:20:45.750 --> 00:20:47.620 How do you formalize that process 571 00:20:47.620 --> 00:20:50.268 in something that's less real time? 572 00:20:50.268 --> 00:20:52.380 I'll give you a real example. 573 00:20:52.380 --> 00:20:55.240 We were doing a pilot maybe a year ago, 574 00:20:55.240 --> 00:20:57.980 and we had a process engineer who's not 575 00:20:57.980 --> 00:21:00.350 a machine learning expert, 576 00:21:00.350 --> 00:21:02.780 was watching the system run, 577 00:21:02.780 --> 00:21:04.040 was looking at the data, 578 00:21:04.040 --> 00:21:07.610 was looking at the analysis 579 00:21:07.610 --> 00:21:09.730 that the machine learning has generated 580 00:21:09.730 --> 00:21:12.570 and how predictable the outcomes 581 00:21:12.570 --> 00:21:14.663 from the line were. 582 00:21:16.431 --> 00:21:17.400 At that stage, 583 00:21:17.400 --> 00:21:19.790 we weren't getting the results that we wanted, 584 00:21:19.790 --> 00:21:22.720 and we were seeing variation in output in the real world 585 00:21:22.720 --> 00:21:24.610 that we weren't picking up and predicting 586 00:21:24.610 --> 00:21:26.313 in the silicone world. 587 00:21:27.580 --> 00:21:28.940 As he was watching the line 588 00:21:28.940 --> 00:21:30.180 he said, Look, I have a theory. 589 00:21:30.180 --> 00:21:33.100 I have a theory that there's something going on 590 00:21:33.100 --> 00:21:36.110 with one of the raw materials were feeding the line. 591 00:21:36.110 --> 00:21:40.720 My theory is that material is more susceptible 592 00:21:40.720 --> 00:21:44.290 to the history of temperature and humidity that 593 00:21:44.290 --> 00:21:46.970 it's experienced as it was shipped to the plant. 594 00:21:46.970 --> 00:21:49.320 Why don't we throw some data-logging 595 00:21:49.320 --> 00:21:51.140 devices on those palettes 596 00:21:51.140 --> 00:21:53.250 as we ship it around the country 597 00:21:53.250 --> 00:21:55.328 and be able to look at that 598 00:21:55.328 --> 00:21:57.600 time and temperature history 599 00:21:57.600 --> 00:21:59.130 and integrate that into the analytics and 600 00:21:59.130 --> 00:22:02.520 see if it would help us be more predictive? 601 00:22:02.520 --> 00:22:04.360 Lo and behold, that was actually helpful. 602 00:22:04.360 --> 00:22:09.360 That's a real-life example of a non-AI expert 603 00:22:09.520 --> 00:22:10.880 interacting with the system 604 00:22:10.880 --> 00:22:13.600 and using their human judgment to suggest 605 00:22:13.600 --> 00:22:14.710 ways to improve it, 606 00:22:14.710 --> 00:22:18.023 even though they can't write AI code. 607 00:22:19.237 --> 00:22:21.260 Once we had exposed enough 608 00:22:21.260 --> 00:22:22.400 of them to what's going on that they were 609 00:22:22.400 --> 00:22:24.930 able to get some human intuition about 610 00:22:24.930 --> 00:22:25.763 what's happening here, 611 00:22:25.763 --> 00:22:27.600 then they were able to participate in the process. 612 00:22:27.600 --> 00:22:29.450 That's a very powerful thing. 613 00:22:29.450 --> 00:22:32.040 Chris, I want to ask you about talent. 614 00:22:32.040 --> 00:22:33.770 You've been talking about a lot of innovation, 615 00:22:33.770 --> 00:22:35.840 a lot of cool ideas. 616 00:22:35.840 --> 00:22:38.670 Different groups, internal and external, 617 00:22:38.670 --> 00:22:40.530 coming together to really 618 00:22:40.530 --> 00:22:41.870 experiment new things, 619 00:22:41.870 --> 00:22:43.310 try new things, 620 00:22:43.310 --> 00:22:46.100 make a really game-changing impact. 621 00:22:46.100 --> 00:22:49.850 What do you think it takes to get the right talent, 622 00:22:49.850 --> 00:22:50.760 motivate them, 623 00:22:50.760 --> 00:22:52.580 keep them excited, 624 00:22:52.580 --> 00:22:55.970 and sort of get that virtuous cycle of 625 00:22:55.970 --> 00:22:59.513 excitement and energy and innovation going? 626 00:23:00.410 --> 00:23:01.790 That's a great question. 627 00:23:01.790 --> 00:23:04.650 I think the answer may be a little different depending on 628 00:23:04.650 --> 00:23:07.313 what sort of technical talent you're talking about. 629 00:23:08.320 --> 00:23:10.620 The way that we would think about a 630 00:23:10.620 --> 00:23:13.330 manufacturing process engineer or controls engineer 631 00:23:13.330 --> 00:23:15.540 may be a bit different how we think about 632 00:23:15.540 --> 00:23:16.590 folks with different 633 00:23:16.590 --> 00:23:18.890 skills in the world of AI, 634 00:23:18.890 --> 00:23:20.510 and sometimes the talent is 635 00:23:20.510 --> 00:23:22.260 in different places in the country. 636 00:23:23.550 --> 00:23:25.720 I'm not sure there's a one-size-fits-all answer. 637 00:23:25.720 --> 00:23:26.940 I think, 638 00:23:26.940 --> 00:23:29.030 in general, when 639 00:23:29.030 --> 00:23:31.650 we find people that we would like to 640 00:23:31.650 --> 00:23:32.860 bring into the company, 641 00:23:32.860 --> 00:23:35.730 I think if we can show them 642 00:23:35.730 --> 00:23:39.320 that the sustained commitment to innovation 643 00:23:39.320 --> 00:23:41.110 and doing cool stuff 644 00:23:41.110 --> 00:23:42.060 is real, 645 00:23:42.060 --> 00:23:43.890 that helps a lot. 646 00:23:43.890 --> 00:23:46.050 I think being able to prove to people 647 00:23:46.050 --> 00:23:47.270 that you're willing to 648 00:23:47.270 --> 00:23:50.700 stay the course in what you're investing in 649 00:23:50.700 --> 00:23:52.223 is part of the story. 650 00:23:53.343 --> 00:23:55.080 Then the second thing that I think is important 651 00:23:55.080 --> 00:23:56.660 is just the culture. 652 00:23:56.660 --> 00:23:58.810 Having people believe that, 653 00:23:58.810 --> 00:24:00.720 in addition to investments 654 00:24:00.720 --> 00:24:03.020 in resource availability, 655 00:24:03.020 --> 00:24:04.920 were just serious about being innovative. 656 00:24:04.920 --> 00:24:06.300 We're just serious 657 00:24:06.300 --> 00:24:08.150 about doing things better. 658 00:24:08.150 --> 00:24:10.970 We're serious about winning through technology, 659 00:24:10.970 --> 00:24:13.800 from the boardroom to the shop floor. 660 00:24:13.800 --> 00:24:16.360 If that culture is real, people know it, 661 00:24:16.360 --> 00:24:18.080 and if it's not real and you're faking it, 662 00:24:18.080 --> 00:24:19.890 I think people know it. 663 00:24:19.890 --> 00:24:22.240 You can't earn that in a quarter; 664 00:24:22.240 --> 00:24:25.597 you've got to earn it over a couple or several years. 665 00:24:25.597 --> 00:24:27.740 I like to think that we've done 666 00:24:27.740 --> 00:24:30.520 a pretty good job with that, but that's really key 667 00:24:30.520 --> 00:24:31.353 in my mind. 668 00:24:32.250 --> 00:24:33.550 Chris, many thanks for taking 669 00:24:33.550 --> 00:24:34.640 the time to talk with us today. 670 00:24:34.640 --> 00:24:37.240 You brought out some quite interesting points. 671 00:24:37.240 --> 00:24:38.490 Thanks for taking the time. 672 00:24:38.490 --> 00:24:40.130 Yeah, Chris, thank you so much. 673 00:24:40.130 --> 00:24:41.390 You're more than welcome. 674 00:24:41.390 --> 00:24:43.240 Hopefully, you can tell I'm excited about 675 00:24:43.240 --> 00:24:45.150 AI. I'm excited about what it can do 676 00:24:45.150 --> 00:24:47.920 in manufacturing as well as other industries. 677 00:24:47.920 --> 00:24:50.080 I think it's going to be a fun future, 678 00:24:50.080 --> 00:24:51.685 I'm looking forward to help build it. 679 00:24:51.685 --> 00:24:54.352 (upbeat melody) 680 00:24:57.970 --> 00:25:00.120 Shervin, Chris covered quite a few key points. 681 00:25:00.120 --> 00:25:02.386 What struck you as particularly noteworthy? 682 00:25:02.386 --> 00:25:05.190 I thought that was really, really insightful. 683 00:25:05.190 --> 00:25:07.030 Obviously, they've done a ton with AI 684 00:25:07.030 --> 00:25:08.830 and a lot of innovation 685 00:25:08.830 --> 00:25:10.430 and cool ideas, and they've put 686 00:25:10.430 --> 00:25:12.400 many of those into production. 687 00:25:12.400 --> 00:25:16.120 I felt a lot of the key sort of 688 00:25:16.120 --> 00:25:18.180 steps in getting value from AI 689 00:25:18.180 --> 00:25:19.920 that we've been talking about 690 00:25:19.920 --> 00:25:22.040 was echoed in what he talked about. 691 00:25:22.040 --> 00:25:23.220 The notion of 692 00:25:23.220 --> 00:25:25.715 experimentation and test and learn. 693 00:25:25.715 --> 00:25:30.190 The culture and importance of allowing folks to try ideas 694 00:25:30.190 --> 00:25:32.540 and fail fast and then moving on. 695 00:25:32.540 --> 00:25:35.570 The notion of focusing on 696 00:25:35.570 --> 00:25:37.490 a few things to scale, 697 00:25:37.490 --> 00:25:40.740 focusing on a lot to sort of test and prototype, 698 00:25:40.740 --> 00:25:43.830 but a few to scale and invest behind. 699 00:25:43.830 --> 00:25:45.460 I thought that was really interesting. 700 00:25:45.460 --> 00:25:47.370 I thought Chris also had a nice blend 701 00:25:47.370 --> 00:25:49.563 of both excitement and patience. 702 00:25:50.636 --> 00:25:51.469 I mean, clearly excited 703 00:25:51.469 --> 00:25:52.750 about some of the things they're doing, 704 00:25:52.750 --> 00:25:54.470 but at the same time, 705 00:25:54.470 --> 00:25:55.670 some of the initiatives were taking 706 00:25:55.670 --> 00:25:57.930 two years or so to come to fruition. 707 00:25:57.930 --> 00:25:59.120 That has to be hard to balance. 708 00:25:59.120 --> 00:26:00.220 Being excited about something 709 00:26:00.220 --> 00:26:02.740 and then also waiting two years for it to come out. 710 00:26:02.740 --> 00:26:04.430 I thought that was a nice blend. 711 00:26:04.430 --> 00:26:06.480 Yeah, and also, to that point, 712 00:26:06.480 --> 00:26:08.060 the importance of focus, right? 713 00:26:08.060 --> 00:26:09.960 I mean, once you've picked it 714 00:26:09.960 --> 00:26:12.920 and you've decided that this is the right thing to do, 715 00:26:12.920 --> 00:26:14.820 and you're sort of seeing it 716 00:26:14.820 --> 00:26:16.760 progressing toward that, 717 00:26:16.760 --> 00:26:19.680 realizing that it's not time to give up now, 718 00:26:19.680 --> 00:26:22.790 you just have to mobilize and double down on it. 719 00:26:22.790 --> 00:26:24.830 The one thing that really struck me a lot was 720 00:26:24.830 --> 00:26:27.180 the importance of culture 721 00:26:27.180 --> 00:26:29.260 and how he said, 722 00:26:29.260 --> 00:26:30.810 from boardroom 723 00:26:30.810 --> 00:26:33.330 all the way to middle management, 724 00:26:33.330 --> 00:26:34.820 they have to believe 725 00:26:36.150 --> 00:26:37.530 that were behind it, 726 00:26:37.530 --> 00:26:40.540 and were investing, and this is not just a fad. 727 00:26:40.540 --> 00:26:42.250 That has to be sort of 728 00:26:42.250 --> 00:26:44.590 permeating across the entire organization 729 00:26:44.590 --> 00:26:47.410 to keep talent really excited and interested. 730 00:26:47.410 --> 00:26:48.950 And it went down even into the 731 00:26:48.950 --> 00:26:50.370 people who were using the systems. 732 00:26:50.370 --> 00:26:52.490 I thought that was a beautiful example of 733 00:26:52.490 --> 00:26:54.700 people who may have been so busy 734 00:26:54.700 --> 00:26:56.980 trying to just get their job done 735 00:26:56.980 --> 00:26:58.330 that they couldn't 736 00:26:58.330 --> 00:26:59.730 step back and think a little bit. 737 00:26:59.730 --> 00:27:02.060 He gave a great example of how 738 00:27:02.060 --> 00:27:05.200 that freedom of having the machine do some of the work 739 00:27:05.200 --> 00:27:07.730 let the human do things that humans are good at. 740 00:27:07.730 --> 00:27:09.960 He covered almost all of the steps in our prior report 741 00:27:09.960 --> 00:27:11.780 about the different ways of people 742 00:27:11.780 --> 00:27:13.120 working with machines. 743 00:27:13.120 --> 00:27:14.700 We didn't prompt him for that. 744 00:27:14.700 --> 00:27:16.290 The other thing I really liked 745 00:27:16.290 --> 00:27:18.700 was his view on talent. 746 00:27:18.700 --> 00:27:22.180 I asked him, what does it take to recruit and 747 00:27:22.180 --> 00:27:24.120 cultivate and retain good talent? 748 00:27:24.120 --> 00:27:26.430 He said, it's not a one-size-fits-all. 749 00:27:26.430 --> 00:27:28.120 That recognition that 750 00:27:28.120 --> 00:27:30.630 not all talent is 751 00:27:30.630 --> 00:27:32.010 of the same cloth, 752 00:27:32.010 --> 00:27:33.950 and different people, different skill sets, 753 00:27:33.950 --> 00:27:34.783 have different 754 00:27:34.783 --> 00:27:35.750 sensibilities, and they're 755 00:27:35.750 --> 00:27:37.210 looking for different things, 756 00:27:37.210 --> 00:27:39.020 but the common theme of 757 00:27:40.090 --> 00:27:43.130 people that go there want a continuous 758 00:27:43.130 --> 00:27:46.603 focus and commitment to innovation, 759 00:27:47.723 --> 00:27:48.830 and they want to see that, 760 00:27:48.830 --> 00:27:50.490 and maybe that's the common thing. 761 00:27:50.490 --> 00:27:53.210 Then, data scientists and technologists 762 00:27:53.210 --> 00:27:55.959 and chemists and engineers might have different 763 00:27:55.959 --> 00:27:58.750 career paths and career aspirations, 764 00:27:58.750 --> 00:28:01.610 but they all share in that common 765 00:28:01.610 --> 00:28:02.900 striving for innovation. 766 00:28:02.900 --> 00:28:04.490 I don't think he mentioned it, 767 00:28:04.490 --> 00:28:06.910 but Chris is a Techstars mentor, 768 00:28:06.910 --> 00:28:08.440 and I'm sure that some of that background 769 00:28:08.440 --> 00:28:10.840 also influences the way he thinks about 770 00:28:10.840 --> 00:28:12.240 different people and different ideas 771 00:28:12.240 --> 00:28:14.420 and how that talent can come together. 772 00:28:14.420 --> 00:28:15.910 Yup, that's right. 773 00:28:15.910 --> 00:28:18.245 He didn't mention it, but that's true. 774 00:28:18.245 --> 00:28:19.078 (upbeat melody) 775 00:28:19.078 --> 00:28:20.640 Thanks for joining us today. 776 00:28:20.640 --> 00:28:22.460 Next time, we'll talk with Huiming Qu 777 00:28:22.460 --> 00:28:24.460 about how The Home Depot continues to build 778 00:28:24.460 --> 00:28:26.100 its AI capabilities. 779 00:28:26.100 --> 00:28:26.953 Please join us. 780 00:28:29.570 --> 00:28:32.320 Thanks for listening to Me, Myself, and AI. 781 00:28:32.320 --> 00:28:33.950 If you're enjoying the show, 782 00:28:33.950 --> 00:28:35.757 take a minute to write us a review. 783 00:28:35.757 --> 00:28:37.680 If you send us a screenshot, 784 00:28:37.680 --> 00:28:39.030 we'll send you a collection of 785 00:28:39.030 --> 00:28:42.580 MIT SMR's best articles on artificial intelligence, 786 00:28:42.580 --> 00:28:44.510 free for a limited time. 787 00:28:44.510 --> 00:28:46.130 Send your review screenshot to 788 00:28:46.130 --> 00:28:49.253 smrfeedback@mit.edu.