1 00:00:00,000 --> 00:00:01,480 Welcome to the first episode 2 00:00:01,480 --> 00:00:03,040 of Technology Powered Growth, 3 00:00:03,040 --> 00:00:05,360 a podcast presented by Ping An. 4 00:00:05,360 --> 00:00:07,800 I’m your host, Treena Nairne. 5 00:00:07,800 --> 00:00:09,320 Today, I’m speaking with Michael Guo, 6 00:00:09,320 --> 00:00:11,480 Executive Director and Co-CEO of Ping An Group, 7 00:00:11,480 --> 00:00:14,480 and Dr. Xiao Jing, Group Chief Scientist. 8 00:00:14,480 --> 00:00:15,120 We’ll talk 9 00:00:15,120 --> 00:00:17,680 about how Ping An leverages technology to drive 10 00:00:17,680 --> 00:00:19,000 business growth. 11 00:00:19,000 --> 00:00:20,880 Ping An’s integrated finance 12 00:00:20,880 --> 00:00:21,680 plus health 13 00:00:21,680 --> 00:00:23,440 care and senior care strategy is unique 14 00:00:23,440 --> 00:00:24,800 for a financial services company. 15 00:00:24,800 --> 00:00:27,640 It's what stands out globally. Yes. 16 00:00:27,640 --> 00:00:28,160 So, Michael, 17 00:00:28,160 --> 00:00:29,360 how would you describe Ping An's 18 00:00:29,360 --> 00:00:31,800 competitive difference in the market? 19 00:00:31,800 --> 00:00:33,160 Yeah, well, that's a good question. 20 00:00:33,160 --> 00:00:35,600 Ping An is a big conglomerate 21 00:00:35,600 --> 00:00:37,000 of the financial services. 22 00:00:37,000 --> 00:00:38,480 It's one of the largest 23 00:00:38,480 --> 00:00:40,440 financial institutions in China 24 00:00:40,440 --> 00:00:42,040 and in the world. 25 00:00:42,040 --> 00:00:45,720 I believe a good strategy should be in line 26 00:00:45,720 --> 00:00:48,440 with the mega trend of the overall economy 27 00:00:48,440 --> 00:00:49,520 and should be centered 28 00:00:49,520 --> 00:00:50,920 around the customer needs. 29 00:00:50,920 --> 00:00:52,840 So that's exactly what we're doing. 30 00:00:52,840 --> 00:00:56,200 If you look at Ping An, we have different 31 00:00:56,200 --> 00:00:58,240 financial services in banking, 32 00:00:58,240 --> 00:01:00,280 insurance, investments, 33 00:01:00,280 --> 00:01:02,080 securities, et cetera. 34 00:01:02,080 --> 00:01:03,920 Then we have a second part of the arm, 35 00:01:03,920 --> 00:01:06,920 which is around health care and senior care. 36 00:01:07,160 --> 00:01:10,440 So if you look at the overall China's economy 37 00:01:10,440 --> 00:01:11,280 for the past 20, 38 00:01:11,280 --> 00:01:12,320 30 years, 39 00:01:12,320 --> 00:01:15,320 there has been rapid financial growth. 40 00:01:15,440 --> 00:01:18,160 Now through that process, Chinese families 41 00:01:18,160 --> 00:01:20,000 and individuals have accumulated 42 00:01:20,000 --> 00:01:21,840 a significant amount of wealth. 43 00:01:21,840 --> 00:01:25,360 So when we talk about middle class in China, 44 00:01:25,640 --> 00:01:28,440 there are now about one-third of the people, 45 00:01:28,440 --> 00:01:29,120 individuals 46 00:01:29,120 --> 00:01:31,960 or families, are called middle class. 47 00:01:31,960 --> 00:01:34,960 Now if you look at a typical Chinese family, 48 00:01:34,960 --> 00:01:36,920 for example, a couple with perhaps 49 00:01:36,920 --> 00:01:37,880 two children, 50 00:01:37,880 --> 00:01:38,800 maybe four parents 51 00:01:38,800 --> 00:01:40,440 to support, these days 52 00:01:40,440 --> 00:01:43,000 they have multiple credit cards, 53 00:01:43,000 --> 00:01:44,160 a number of insurance 54 00:01:44,160 --> 00:01:46,720 policies, a number of cars, perhaps a mortgage. 55 00:01:46,720 --> 00:01:48,520 Their financial needs are becoming 56 00:01:48,520 --> 00:01:50,720 more and more sophisticated and complicated. 57 00:01:50,720 --> 00:01:52,120 So one of the key 58 00:01:52,120 --> 00:01:54,440 competitive differences of Ping An 59 00:01:54,440 --> 00:01:56,840 is that we're able to integrate our side 60 00:01:56,840 --> 00:01:58,280 of the financial services 61 00:01:58,280 --> 00:01:59,600 and provide a one-stop 62 00:01:59,600 --> 00:02:01,280 solution for our customers 63 00:02:01,280 --> 00:02:02,800 to make them hassle-free 64 00:02:02,800 --> 00:02:04,040 without having to deal 65 00:02:04,040 --> 00:02:05,760 with all these different 66 00:02:05,760 --> 00:02:06,800 sophisticated, more 67 00:02:06,800 --> 00:02:08,520 and more sophisticated financial products. 68 00:02:08,520 --> 00:02:10,160 So that's one side of 69 00:02:11,600 --> 00:02:14,240 our strategy and differentiator. 70 00:02:14,240 --> 00:02:15,400 The other side is that 71 00:02:15,400 --> 00:02:16,920 we're looking at the overall 72 00:02:16,920 --> 00:02:18,840 long-term growth and demographic 73 00:02:18,840 --> 00:02:19,880 change of China's population. 74 00:02:19,880 --> 00:02:23,960 Now in five years’ time, roughly 75 00:02:24,280 --> 00:02:26,960 25 to 30 percent of the Chinese population 76 00:02:26,960 --> 00:02:29,760 is becoming more than 65 years old. 77 00:02:29,760 --> 00:02:30,600 With that, 78 00:02:30,600 --> 00:02:31,320 they have more and more 79 00:02:31,320 --> 00:02:33,280 diversified needs on the health care 80 00:02:33,280 --> 00:02:35,720 and the senior care part of the services. 81 00:02:35,720 --> 00:02:37,680 And for our customers, 82 00:02:37,680 --> 00:02:40,120 who are roughly middle-class 83 00:02:40,120 --> 00:02:41,360 families and individuals, 84 00:02:41,360 --> 00:02:44,280 they would prefer to have some high-quality, 85 00:02:44,280 --> 00:02:47,280 perhaps differentiated, tailored services 86 00:02:47,680 --> 00:02:48,360 in addition 87 00:02:48,360 --> 00:02:50,640 to the social welfare systems 88 00:02:50,640 --> 00:02:52,920 provided by the government. 89 00:02:52,920 --> 00:02:54,320 So from that perspective, 90 00:02:54,320 --> 00:02:57,080 we're building up our capabilities around 91 00:02:57,080 --> 00:02:57,720 health care 92 00:02:57,720 --> 00:02:58,920 and the senior care 93 00:02:58,920 --> 00:03:01,520 and provide a tailored solution to them 94 00:03:01,520 --> 00:03:04,520 based on our strong financial resources. 95 00:03:04,760 --> 00:03:06,320 So from that perspective, 96 00:03:06,320 --> 00:03:08,080 we are actually tailored to 97 00:03:08,080 --> 00:03:10,160 our customer needs to the future. 98 00:03:10,160 --> 00:03:12,200 So that's how we see ourselves. 99 00:03:12,200 --> 00:03:14,720 One side of the integrated finance solution 100 00:03:14,720 --> 00:03:15,440 to make sure 101 00:03:15,440 --> 00:03:17,080 we satisfy our needs, 102 00:03:17,080 --> 00:03:20,120 our customers' needs and the moment, but also 103 00:03:20,240 --> 00:03:21,280 we gear toward 104 00:03:21,280 --> 00:03:23,200 the needs for them for the future. 105 00:03:23,200 --> 00:03:24,000 So tell us a little bit 106 00:03:24,000 --> 00:03:25,720 about how technology feeds 107 00:03:25,720 --> 00:03:27,120 into that overall strategy. 108 00:03:27,120 --> 00:03:28,160 Because one of the key 109 00:03:28,160 --> 00:03:29,240 things of all 110 00:03:29,240 --> 00:03:31,800 these things has to be integrated, 111 00:03:31,800 --> 00:03:33,760 and thanks to our Ping An’s 112 00:03:33,760 --> 00:03:35,360 proprietary technology 113 00:03:35,360 --> 00:03:36,080 and the data 114 00:03:36,080 --> 00:03:37,160 we have accumulated 115 00:03:37,160 --> 00:03:38,120 for the past 10, 15 116 00:03:38,120 --> 00:03:41,120 years, we’re able to actually provide 117 00:03:41,200 --> 00:03:42,680 much more integrated 118 00:03:42,680 --> 00:03:43,680 and, on the other hand, 119 00:03:43,680 --> 00:03:45,080 tailored solutions 120 00:03:45,080 --> 00:03:47,640 to our individual families and customers. 121 00:03:47,640 --> 00:03:49,240 And through the past 10, 122 00:03:49,240 --> 00:03:51,200 15 years of investment, 123 00:03:51,200 --> 00:03:53,880 we have accumulated a massive amount 124 00:03:53,880 --> 00:03:55,120 of the customer data, 125 00:03:55,120 --> 00:03:56,960 not just on the financial side of it, 126 00:03:56,960 --> 00:03:59,240 but also end-to-end of the customer 127 00:03:59,240 --> 00:04:01,320 journey across different channels. 128 00:04:01,320 --> 00:04:03,560 These days, you’ve got internet, 129 00:04:03,560 --> 00:04:04,800 you’ve got apps, 130 00:04:04,800 --> 00:04:05,920 you’ve got offline, you’ve 131 00:04:05,920 --> 00:04:07,760 got online, you’ve got social media. 132 00:04:07,760 --> 00:04:09,480 Then you’ve got the behavior data 133 00:04:09,480 --> 00:04:11,440 accumulated from our ecosystem. 134 00:04:11,440 --> 00:04:14,120 And all this data has to be 135 00:04:14,120 --> 00:04:15,080 integrated together 136 00:04:15,080 --> 00:04:16,560 so we're able to actually 137 00:04:16,560 --> 00:04:17,360 use the data 138 00:04:17,360 --> 00:04:18,520 to provide more tailored 139 00:04:18,520 --> 00:04:20,600 solutions to our customers. 140 00:04:20,600 --> 00:04:23,600 And that becomes available and possible 141 00:04:23,760 --> 00:04:25,160 thanks to our technology 142 00:04:25,160 --> 00:04:26,120 and, these days, 143 00:04:26,120 --> 00:04:27,480 AI advancements. 144 00:04:27,480 --> 00:04:28,240 And, of course, 145 00:04:28,240 --> 00:04:31,120 AI is the hot topic for everybody this year. 146 00:04:31,120 --> 00:04:32,160 So tell us a little bit 147 00:04:32,160 --> 00:04:33,960 more about the specifics 148 00:04:33,960 --> 00:04:34,400 of how 149 00:04:34,400 --> 00:04:36,000 AI actually helps Ping An 150 00:04:36,000 --> 00:04:37,720 stand apart from competitors. 151 00:04:37,720 --> 00:04:40,200 Well, if you look at the AI, 152 00:04:40,200 --> 00:04:41,760 it's just part of the technology. 153 00:04:41,760 --> 00:04:42,920 It’s just these days 154 00:04:42,920 --> 00:04:43,880 the technology becomes 155 00:04:43,880 --> 00:04:45,280 more and more sophisticated. 156 00:04:45,280 --> 00:04:45,920 In Ping An, 157 00:04:45,920 --> 00:04:46,960 when we look at technology, 158 00:04:46,960 --> 00:04:49,960 our strategy is always around 159 00:04:50,000 --> 00:04:53,120 how do we use technology, the AI in particular, 160 00:04:53,160 --> 00:04:56,280 add value to our services, to our customers, 161 00:04:56,280 --> 00:04:58,640 to our business, to our management. 162 00:04:58,640 --> 00:04:59,720 So, in general, 163 00:04:59,720 --> 00:05:02,240 if we look at the technology and how we see it, 164 00:05:02,240 --> 00:05:02,840 there are perhaps 165 00:05:02,840 --> 00:05:03,640 a few areas 166 00:05:03,640 --> 00:05:06,640 the AI and technology is able to add value. 167 00:05:07,040 --> 00:05:09,800 It's around cost efficiency, it's 168 00:05:09,800 --> 00:05:11,480 around the productivity, 169 00:05:11,480 --> 00:05:13,760 it's around the risk management. 170 00:05:13,760 --> 00:05:14,840 So all these things 171 00:05:14,840 --> 00:05:15,560 come together 172 00:05:15,560 --> 00:05:18,120 to become one of the driving forces 173 00:05:18,120 --> 00:05:21,120 for our profitability of our business. 174 00:05:21,200 --> 00:05:21,600 That's why, 175 00:05:21,600 --> 00:05:22,480 if we look at Ping An, 176 00:05:22,480 --> 00:05:24,400 we are one of the most profitable 177 00:05:24,400 --> 00:05:25,680 financial institutions 178 00:05:25,680 --> 00:05:27,200 in the market in China 179 00:05:27,200 --> 00:05:28,880 and also according to 180 00:05:28,880 --> 00:05:30,000 global standards as well. 181 00:05:30,000 --> 00:05:30,720 Now we're going to come back 182 00:05:30,720 --> 00:05:32,440 and talk about some specific examples 183 00:05:32,440 --> 00:05:32,960 of what Ping An 184 00:05:32,960 --> 00:05:34,280 is actually doing with technology. 185 00:05:34,280 --> 00:05:35,720 But from what you're saying, 186 00:05:35,720 --> 00:05:36,240 Ping An's 187 00:05:36,240 --> 00:05:37,400 offering really comes 188 00:05:37,400 --> 00:05:39,280 through the scale of your business, 189 00:05:39,280 --> 00:05:41,080 the depth of customer data, 190 00:05:41,080 --> 00:05:42,680 and I understand 191 00:05:42,680 --> 00:05:44,360 from what you're saying that actually Ping 192 00:05:44,360 --> 00:05:45,720 An had a very early vision 193 00:05:45,720 --> 00:05:46,680 of digitalizing 194 00:05:46,680 --> 00:05:48,080 the business beyond that 195 00:05:48,080 --> 00:05:49,720 traditional financial services model. 196 00:05:49,720 --> 00:05:51,760 So what have been the challenges in delivering that? 197 00:05:51,760 --> 00:05:53,120 You mentioned the scale. 198 00:05:53,120 --> 00:05:55,320 This is actually one of the key challenges. 199 00:05:55,320 --> 00:05:57,880 On the one side is our strength. 200 00:05:57,880 --> 00:06:00,440 We got 240 million customers. 201 00:06:00,440 --> 00:06:01,240 We got one 202 00:06:01,240 --> 00:06:05,520 third of the market share, across the 2000. 203 00:06:05,520 --> 00:06:07,440 The more than 2000 cities in China. 204 00:06:07,440 --> 00:06:09,680 So that scale gives us very strong 205 00:06:09,680 --> 00:06:11,040 financial strength. 206 00:06:11,040 --> 00:06:12,360 But on the other hand, 207 00:06:12,360 --> 00:06:14,000 when we talk about tailored 208 00:06:14,000 --> 00:06:15,680 personalized solutions, 209 00:06:15,680 --> 00:06:18,440 scalability versus personalization, 210 00:06:18,440 --> 00:06:21,520 standardization quality versus cost efficiency, 211 00:06:21,520 --> 00:06:23,360 those are the kind of competing 212 00:06:23,360 --> 00:06:24,680 and challenging factors 213 00:06:24,680 --> 00:06:26,760 that we have to be able to manage. 214 00:06:26,760 --> 00:06:29,400 So, again, coming back, the technology 215 00:06:29,400 --> 00:06:32,440 plays a key role in balancing all these kind of 216 00:06:32,440 --> 00:06:34,120 competing and conflicting factors 217 00:06:34,120 --> 00:06:36,560 through our service providing process. 218 00:06:36,560 --> 00:06:39,400 So one of the things we actually started early on 219 00:06:39,400 --> 00:06:42,840 is to standardize all the operational processes 220 00:06:43,160 --> 00:06:44,480 across all the services, 221 00:06:44,480 --> 00:06:46,440 financial services side of it, 222 00:06:46,440 --> 00:06:49,280 health care, senior care side of it. 223 00:06:49,280 --> 00:06:50,960 Now, through the standardization 224 00:06:50,960 --> 00:06:53,120 of the operational processes, 225 00:06:53,120 --> 00:06:54,640 we're able to generate, 226 00:06:54,640 --> 00:06:57,000 on the one hand, sufficient cost efficiency. 227 00:06:57,000 --> 00:06:57,760 On the other hand, 228 00:06:57,760 --> 00:06:58,160 we're able 229 00:06:58,160 --> 00:06:58,960 to actually 230 00:06:58,960 --> 00:07:02,160 make a very good standard of the qualities 231 00:07:02,160 --> 00:07:03,160 of the services. 232 00:07:04,480 --> 00:07:06,000 Now, with that being the backbone 233 00:07:06,000 --> 00:07:07,080 and the foundation, 234 00:07:07,080 --> 00:07:08,400 using all the data 235 00:07:08,400 --> 00:07:10,760 we have accumulated, using these days 236 00:07:10,760 --> 00:07:13,720 much more advanced tools and data analytics, 237 00:07:13,720 --> 00:07:14,240 we're able 238 00:07:14,240 --> 00:07:15,360 to actually tailor-made 239 00:07:15,360 --> 00:07:17,600 the solutions to our individual customers 240 00:07:17,600 --> 00:07:19,040 in different cities. 241 00:07:19,040 --> 00:07:22,040 So that’s why when we look at this balancing 242 00:07:22,200 --> 00:07:25,200 kind of act among all the competing factors, 243 00:07:25,520 --> 00:07:28,680 using the technology to standardize everything, 244 00:07:28,960 --> 00:07:32,000 but making sure we use the data and the tools 245 00:07:32,000 --> 00:07:34,360 to tailor-made some of the solutions 246 00:07:34,360 --> 00:07:35,800 to different customers, 247 00:07:35,800 --> 00:07:36,800 that's how we look at 248 00:07:36,800 --> 00:07:39,200 and how to actually make sure we satisfy 249 00:07:39,200 --> 00:07:40,160 the customer needs 250 00:07:40,160 --> 00:07:41,720 with different individuals, 251 00:07:41,720 --> 00:07:43,400 but with sufficient level 252 00:07:43,400 --> 00:07:44,080 for quality 253 00:07:44,080 --> 00:07:45,320 and cost efficiency. 254 00:07:45,320 --> 00:07:46,160 Now, Jing, 255 00:07:46,160 --> 00:07:47,840 as the Group Chief Scientist, 256 00:07:47,840 --> 00:07:49,120 could you give us some insight 257 00:07:49,120 --> 00:07:51,120 into Ping An's technology framework? 258 00:07:51,120 --> 00:07:52,000 So how do you prioritize? 259 00:07:52,000 --> 00:07:54,480 Where do you invest those technology efforts? 260 00:07:54,480 --> 00:07:56,560 We have been developing our own technology 261 00:07:56,560 --> 00:07:58,840 completely for well over a decade. 262 00:07:58,840 --> 00:08:02,000 So we have a tremendously rich financial 263 00:08:02,000 --> 00:08:03,240 and medical data sets, 264 00:08:03,240 --> 00:08:06,000 which we have checked throughout that time. 265 00:08:06,000 --> 00:08:08,680 Today we have a strong enhanced technology team 266 00:08:08,680 --> 00:08:13,680 of more than 3,000 scientists and 20,000 IT engineers 267 00:08:13,680 --> 00:08:18,160 whose work is organized in what we call our 953 framework. 268 00:08:18,160 --> 00:08:21,400 953 is a shorthand for our nine 269 00:08:21,440 --> 00:08:22,920 global leading databases, 270 00:08:22,920 --> 00:08:23,840 plus five 271 00:08:23,840 --> 00:08:27,440 laboratories and three technology companies. 272 00:08:27,920 --> 00:08:31,960 Currently we have accumulated over 55,000 273 00:08:32,200 --> 00:08:35,720 patent disclosures, leads globally in fintech 274 00:08:35,960 --> 00:08:39,400 and health tech, and also second in generative AI. 275 00:08:39,400 --> 00:08:40,920 We've just talked through a lot of numbers 276 00:08:40,920 --> 00:08:42,760 there, so let's break that down a little bit. 277 00:08:42,760 --> 00:08:46,320 You've talked about your 953 framework, so nine 278 00:08:46,320 --> 00:08:47,720 global leading databases, 279 00:08:47,720 --> 00:08:50,760 five laboratories, three technology companies. 280 00:08:51,160 --> 00:08:53,160 Can you walk us through each one of those? 281 00:08:53,160 --> 00:08:54,480 Yeah, of course. 282 00:08:54,480 --> 00:08:57,560 Basically, the nine databases 283 00:08:57,560 --> 00:09:00,680 we have consist of three different categories. 284 00:09:00,800 --> 00:09:02,680 One is financial databases, 285 00:09:02,680 --> 00:09:04,680 another is medical and health care, 286 00:09:04,680 --> 00:09:06,480 and the third one contains the data 287 00:09:06,480 --> 00:09:08,120 from the internal business 288 00:09:08,120 --> 00:09:09,920 operation and management 289 00:09:09,920 --> 00:09:10,720 in Ping An. 290 00:09:10,720 --> 00:09:12,200 For the financial databases, 291 00:09:12,200 --> 00:09:15,160 we have three different categories of data, 292 00:09:15,160 --> 00:09:17,800 including individual customers, corporate 293 00:09:17,800 --> 00:09:19,880 customers, and financial products. 294 00:09:19,880 --> 00:09:20,760 We process 295 00:09:20,760 --> 00:09:24,080 more than one billion records per day, serve 296 00:09:24,400 --> 00:09:28,360 for more than 240 million individual customers 297 00:09:28,640 --> 00:09:29,440 to generate 298 00:09:29,440 --> 00:09:30,200 deep insights 299 00:09:30,200 --> 00:09:31,240 into customers’ 300 00:09:31,240 --> 00:09:34,280 needs to improve the user experience. 301 00:09:34,280 --> 00:09:36,400 For the medical and health care databases, 302 00:09:36,400 --> 00:09:39,400 we have the information from five categories, 303 00:09:39,680 --> 00:09:42,160 respectively regarding disease 304 00:09:42,160 --> 00:09:44,600 and the symptoms, prescriptions, medicines 305 00:09:44,600 --> 00:09:47,600 and medical equipments, doctors and hospitals, 306 00:09:47,600 --> 00:09:49,080 and customer health records. 307 00:09:49,080 --> 00:09:51,280 The third is our internal business 308 00:09:51,280 --> 00:09:53,200 operation and management database. 309 00:09:53,200 --> 00:09:54,800 So hold on there for a second. 310 00:09:54,800 --> 00:09:56,240 Let's go back. 311 00:09:56,240 --> 00:09:59,040 One billion records per day. 312 00:09:59,040 --> 00:10:00,880 That is impressive. YUh-huh. Yeah. 313 00:10:00,880 --> 00:10:01,640 Thank you. Yeah, that’s pretty big. 314 00:10:01,640 --> 00:10:03,680 But that's not all. 315 00:10:03,680 --> 00:10:06,560 We have five laboratories that focus 316 00:10:06,560 --> 00:10:08,600 on developing our capabilities 317 00:10:08,600 --> 00:10:10,600 in different AI areas, 318 00:10:10,600 --> 00:10:12,480 including computer vision, speech, 319 00:10:12,480 --> 00:10:13,840 and natural language 320 00:10:13,840 --> 00:10:15,360 understanding, data 321 00:10:15,360 --> 00:10:17,560 analytics, micro-expressions, 322 00:10:17,560 --> 00:10:18,680 and the Silicon Valley Lab, 323 00:10:18,680 --> 00:10:22,560 which is targeted at cutting-edge technologies development. 324 00:10:23,120 --> 00:10:26,840 By the way, we are ranked number one 325 00:10:27,080 --> 00:10:28,760 in terms of fintech, 326 00:10:28,760 --> 00:10:31,280 health tech patent disclosures. 327 00:10:31,280 --> 00:10:33,080 Now, I think most people have an idea 328 00:10:33,080 --> 00:10:34,720 of what data analytics is about, 329 00:10:34,720 --> 00:10:36,680 but computer vision research 330 00:10:36,680 --> 00:10:38,920 is about using AI to interpret images, right? 331 00:10:38,920 --> 00:10:39,680 Yeah. 332 00:10:39,680 --> 00:10:41,640 Speech and NLP or natural language 333 00:10:41,640 --> 00:10:43,000 processing is about computers 334 00:10:43,000 --> 00:10:43,320 being able 335 00:10:43,320 --> 00:10:45,800 to interpret and generate human language, 336 00:10:45,800 --> 00:10:48,520 and micro-expressions That’s about computer 337 00:10:48,520 --> 00:10:51,600 interpretation of tiny facial muscle movements 338 00:10:51,600 --> 00:10:52,680 that reveal human emotions, 339 00:10:52,680 --> 00:10:55,680 which I think is absolutely fascinating. 340 00:10:56,120 --> 00:10:57,680 So those, along with research 341 00:10:57,680 --> 00:10:59,200 on cutting-edge technologies, 342 00:10:59,200 --> 00:11:01,520 make up your five labs, right? Yeah. 343 00:11:01,520 --> 00:11:03,240 So what does the three represent? 344 00:11:03,240 --> 00:11:05,720 The three represents our three tech companies, 345 00:11:05,720 --> 00:11:08,720 Ping An Technology, Ping An Health, 346 00:11:08,840 --> 00:11:11,000 and One Connect Financial Technology. 347 00:11:11,000 --> 00:11:13,400 These are the companies that develop 348 00:11:13,400 --> 00:11:14,720 digital solutions 349 00:11:14,720 --> 00:11:17,600 based on the AI research from our labs 350 00:11:17,600 --> 00:11:18,440 and our data, 351 00:11:18,440 --> 00:11:20,240 covering thousands of diverse 352 00:11:20,240 --> 00:11:21,600 business scenarios 353 00:11:21,600 --> 00:11:24,560 with hundreds, millions of large language 354 00:11:24,560 --> 00:11:25,880 model calls per day. 355 00:11:25,880 --> 00:11:27,640 These have given us strong 356 00:11:27,640 --> 00:11:30,040 AI technologies foundations 357 00:11:30,040 --> 00:11:31,760 for unique attributes, 358 00:11:31,760 --> 00:11:34,280 especially with our rich data sets, 359 00:11:34,280 --> 00:11:37,280 which we collected from real cases 360 00:11:37,760 --> 00:11:39,440 among our more than 240 361 00:11:39,440 --> 00:11:42,440 million retail customers in China. 362 00:11:42,440 --> 00:11:43,800 Just to give you an idea 363 00:11:43,800 --> 00:11:45,800 how much of data we have, 364 00:11:45,800 --> 00:11:46,520 our public 365 00:11:46,520 --> 00:11:50,480 domain text is made up of approximately 366 00:11:50,480 --> 00:11:54,240 3.2 trillion tokens, and approximately 367 00:11:54,360 --> 00:11:57,600 310,000 hours of labelled speech, 368 00:11:58,280 --> 00:12:00,920 and more than 7.5 billion images. 369 00:12:00,920 --> 00:12:03,080 We use this immense data 370 00:12:03,080 --> 00:12:06,080 to create our unique ecosystem. 371 00:12:06,240 --> 00:12:08,640 Yeah, all these are big numbers. 372 00:12:08,640 --> 00:12:11,400 So you can see we have, we always say 373 00:12:11,400 --> 00:12:14,400 we're sitting on a gold mine of data, 374 00:12:14,560 --> 00:12:16,880 but that didn't really come overnight. 375 00:12:16,880 --> 00:12:17,400 Ping An started 376 00:12:17,400 --> 00:12:19,040 building our data starting around 377 00:12:19,040 --> 00:12:20,000 15 years ago. 378 00:12:20,000 --> 00:12:20,600 One of the key 379 00:12:20,600 --> 00:12:22,640 things we've done is to standardize 380 00:12:22,640 --> 00:12:24,640 all the operational processes 381 00:12:24,640 --> 00:12:25,800 so that we’re able 382 00:12:25,800 --> 00:12:27,840 to capture the customer's data 383 00:12:27,840 --> 00:12:30,520 from channel to channel, end to end. 384 00:12:30,520 --> 00:12:32,880 So it's not just about their transaction 385 00:12:32,880 --> 00:12:33,560 data. 386 00:12:33,560 --> 00:12:35,000 It's also about the behavior 387 00:12:35,000 --> 00:12:37,640 data, how they move from app to internet 388 00:12:37,640 --> 00:12:40,120 banking to offline sales channel. 389 00:12:40,120 --> 00:12:42,200 Then coming back to the call center. 390 00:12:42,200 --> 00:12:44,840 So one of the key things we're able to actually 391 00:12:44,840 --> 00:12:48,760 leverage is the different customer 392 00:12:48,760 --> 00:12:49,760 end-to-end data 393 00:12:49,760 --> 00:12:52,160 across the different channels, behavior data, 394 00:12:52,160 --> 00:12:54,240 but also with advancement 395 00:12:54,240 --> 00:12:56,240 of the analytics engine 396 00:12:56,240 --> 00:12:58,080 and technology and algorithm, 397 00:12:58,080 --> 00:12:59,800 we're able to generate the tools 398 00:12:59,800 --> 00:13:02,680 and analytical engines to be able to interpret 399 00:13:02,680 --> 00:13:03,560 those data 400 00:13:03,560 --> 00:13:06,000 and generate good insights for our customers 401 00:13:06,000 --> 00:13:07,280 and for our sales agents. 402 00:13:07,280 --> 00:13:08,840 So that's actually very important 403 00:13:08,840 --> 00:13:10,160 to add value to our business. 404 00:13:10,160 --> 00:13:12,800 So every day as you're adding that data, 405 00:13:12,800 --> 00:13:15,080 it becomes harder and harder to replicate 406 00:13:15,080 --> 00:13:15,600 isn't it. 407 00:13:15,600 --> 00:13:16,160 It's not like 408 00:13:16,160 --> 00:13:17,240 someone could just walk in 409 00:13:17,240 --> 00:13:18,680 and copy what you do. 410 00:13:18,680 --> 00:13:20,280 Yes, absolutely. 411 00:13:20,280 --> 00:13:20,760 So Michael, 412 00:13:20,760 --> 00:13:21,120 could you 413 00:13:21,120 --> 00:13:22,040 give us an example 414 00:13:22,040 --> 00:13:25,040 of how AI actually impacts the customer today? 415 00:13:25,080 --> 00:13:25,600 Sure. 416 00:13:25,600 --> 00:13:27,760 If you look at our customer's interactions with us, 417 00:13:28,720 --> 00:13:31,200 the one of the key channels is call center. 418 00:13:31,200 --> 00:13:33,000 Now we have 240 419 00:13:33,000 --> 00:13:34,640 million retail customers 420 00:13:34,640 --> 00:13:37,280 these days across different businesses, 421 00:13:37,280 --> 00:13:38,520 and every day 422 00:13:38,520 --> 00:13:40,000 there are more than 2 million 423 00:13:40,000 --> 00:13:42,560 incoming calls into our call center, 424 00:13:42,560 --> 00:13:44,120 and our call center is integrated. 425 00:13:44,120 --> 00:13:45,560 It doesn't really matter 426 00:13:45,560 --> 00:13:46,520 if you’re a 427 00:13:46,520 --> 00:13:48,000 credit card customer or you’re 428 00:13:48,000 --> 00:13:49,440 an insurance customer. 429 00:13:49,440 --> 00:13:51,320 They all go to the same call center, 430 00:13:51,320 --> 00:13:53,200 and the calls are answered 431 00:13:53,200 --> 00:13:55,280 by the same group of agents. 432 00:13:55,280 --> 00:13:56,400 So that's a challenge 433 00:13:56,400 --> 00:13:57,880 because if you look at the incoming 434 00:13:57,880 --> 00:13:58,760 insight, 435 00:13:58,760 --> 00:14:01,760 people can make calls from all sorts of places, 436 00:14:01,920 --> 00:14:03,800 on the car, in the train station, 437 00:14:03,800 --> 00:14:05,600 maybe in the farm. 438 00:14:05,600 --> 00:14:08,240 So we have to be able to actually filter out 439 00:14:08,240 --> 00:14:09,840 all the background noises 440 00:14:09,840 --> 00:14:12,960 for the machine these days to understand 441 00:14:12,960 --> 00:14:15,960 and interpret people's voice and their calls. 442 00:14:16,560 --> 00:14:20,600 Also, China is a diversified country. 443 00:14:20,600 --> 00:14:22,920 We have people from different regions 444 00:14:22,920 --> 00:14:24,600 speaking different dialects, 445 00:14:24,600 --> 00:14:26,520 and that actually, for the machine, 446 00:14:26,520 --> 00:14:29,160 is very difficult to tell, okay, 447 00:14:29,160 --> 00:14:31,360 this person is talking Cantonese, 448 00:14:31,360 --> 00:14:33,040 and the other person is talking 449 00:14:33,040 --> 00:14:34,520 Sichuan dialect. 450 00:14:34,520 --> 00:14:36,160 It's very different. 451 00:14:36,160 --> 00:14:38,720 So that requires very strong capability, 452 00:14:38,720 --> 00:14:39,800 voice recognition, 453 00:14:39,800 --> 00:14:41,200 natural language processing, 454 00:14:41,200 --> 00:14:43,040 all this advanced technology 455 00:14:43,040 --> 00:14:45,320 to be able to actually understand 456 00:14:45,320 --> 00:14:46,760 the customer's voice 457 00:14:46,760 --> 00:14:48,920 and what they're trying to talk about. 458 00:14:48,920 --> 00:14:51,160 So that's only one side of the challenge. 459 00:14:51,160 --> 00:14:51,880 The other side of the 460 00:14:51,880 --> 00:14:54,440 challenge is with this multimedia, 461 00:14:54,440 --> 00:14:56,680 with multi-channel kind of service capability 462 00:14:56,680 --> 00:14:57,280 we have, 463 00:14:57,280 --> 00:14:59,680 and not just about call center coming 464 00:14:59,680 --> 00:15:01,080 in, people could say, oh, 465 00:15:01,080 --> 00:15:03,440 I just put some information on my app 466 00:15:03,440 --> 00:15:04,720 or internet banking, 467 00:15:04,720 --> 00:15:06,800 and now I want to check on the process. 468 00:15:06,800 --> 00:15:08,840 So our call center agents 469 00:15:08,840 --> 00:15:10,920 will have to be able to capture 470 00:15:10,920 --> 00:15:13,640 the information straightaway, almost real time, 471 00:15:13,640 --> 00:15:15,000 from the information 472 00:15:15,000 --> 00:15:15,760 through internet 473 00:15:15,760 --> 00:15:17,000 banking or app 474 00:15:17,000 --> 00:15:19,240 and integrate it into their answers 475 00:15:19,240 --> 00:15:20,160 to the customers 476 00:15:20,160 --> 00:15:20,840 and say, okay, 477 00:15:20,840 --> 00:15:22,560 that's what you have asked for, 478 00:15:22,560 --> 00:15:24,600 and this is what we are going to do for you. 479 00:15:24,600 --> 00:15:27,480 And all this requires analytical 480 00:15:27,480 --> 00:15:28,560 AI technology 481 00:15:28,560 --> 00:15:30,600 to make all these things possible. 482 00:15:30,600 --> 00:15:31,760 So before, 483 00:15:31,760 --> 00:15:32,880 if you think about it, 484 00:15:32,880 --> 00:15:33,800 all these things 485 00:15:33,800 --> 00:15:35,840 would have to be handled by humans. 486 00:15:35,840 --> 00:15:39,280 But these days, with advanced technology of AI, 487 00:15:39,480 --> 00:15:42,440 we're able to actually reduce our call center 488 00:15:42,440 --> 00:15:43,520 agents’ capacity 489 00:15:43,520 --> 00:15:46,880 by 50%, number of headcount by 50%. 490 00:15:47,080 --> 00:15:48,120 But at the same time, 491 00:15:48,120 --> 00:15:50,240 we're still able to answer the customer calls 492 00:15:51,560 --> 00:15:52,120 with increasing volume, with 493 00:15:52,120 --> 00:15:53,400 standard high-level quality. 494 00:15:53,400 --> 00:15:56,000 These days, every call we require 495 00:15:56,000 --> 00:15:57,320 would have to be picked up 496 00:15:57,320 --> 00:15:59,440 until three beeps, within three beeps. 497 00:15:59,440 --> 00:16:01,760 And that has always been the case. 498 00:16:01,760 --> 00:16:02,720 We're able to maintain 499 00:16:02,720 --> 00:16:05,120 the standard of our quality 500 00:16:05,120 --> 00:16:06,680 and improve the efficiency 501 00:16:06,680 --> 00:16:08,800 with fewer and fewer people. 502 00:16:08,800 --> 00:16:09,680 Actually, it sounds like with fewer people 503 00:16:09,680 --> 00:16:10,800 you're actually able 504 00:16:10,800 --> 00:16:12,920 to increase the standard of quality. 505 00:16:12,920 --> 00:16:15,240 Yes, that's true, that's true. 506 00:16:15,240 --> 00:16:17,400 Jing, how else is Ping An using AI? 507 00:16:17,400 --> 00:16:19,080 Besides financial services, 508 00:16:19,080 --> 00:16:19,960 medical and health 509 00:16:19,960 --> 00:16:22,480 care are another strategic direction 510 00:16:22,480 --> 00:16:23,600 of Ping An Group. 511 00:16:23,600 --> 00:16:24,320 We launched 512 00:16:24,320 --> 00:16:28,920 a self-developed AI-assisted diagnosis system, 513 00:16:28,920 --> 00:16:33,160 AI-Doctor, as far back as 2018, 514 00:16:33,600 --> 00:16:35,760 and has undergone 515 00:16:35,760 --> 00:16:38,600 continuous updates and enhancements 516 00:16:38,600 --> 00:16:39,240 to develop 517 00:16:39,240 --> 00:16:42,440 into a comprehensive end-to-end solution 518 00:16:42,720 --> 00:16:44,160 from pre-diagnosis, 519 00:16:44,160 --> 00:16:47,960 treatment, and post-recovery phases. 520 00:16:47,960 --> 00:16:49,720 Today, this ecosystem 521 00:16:49,720 --> 00:16:53,320 encompasses collaborations with nearly 4,000 522 00:16:53,680 --> 00:16:57,520 hospitals, over 50,000 medical professionals, 523 00:16:57,760 --> 00:17:00,960 and more than 150 senior care services 524 00:17:00,960 --> 00:17:02,360 providers nationwide. 525 00:17:02,360 --> 00:17:05,640 The system now covers diagnostic expertise 526 00:17:05,640 --> 00:17:08,640 for over 2,000 most common diseases, 527 00:17:08,840 --> 00:17:10,520 achieving the navigation 528 00:17:10,520 --> 00:17:13,520 accuracy of existing 99% 529 00:17:13,520 --> 00:17:17,000 and the diagnosis precision of above 95%. 530 00:17:17,000 --> 00:17:19,280 It is absolutely incredible 531 00:17:19,280 --> 00:17:21,560 how many AI models you're working on. 532 00:17:21,560 --> 00:17:24,320 You must need immense data inputs to feed them. 533 00:17:24,320 --> 00:17:25,960 So how does Ping An actually manage 534 00:17:25,960 --> 00:17:27,400 all of these different models? 535 00:17:27,400 --> 00:17:28,640 Our core technology 536 00:17:28,640 --> 00:17:30,960 foundation is called Ping’an Brain, 537 00:17:30,960 --> 00:17:33,080 which is the fundamental engine 538 00:17:33,080 --> 00:17:35,880 efficiently and effectively empowering 539 00:17:35,880 --> 00:17:39,840 the group's various business units to advance 540 00:17:39,840 --> 00:17:41,120 and scale up their 541 00:17:41,120 --> 00:17:42,880 AI business solutions 542 00:17:42,880 --> 00:17:45,120 across enormous scenarios. 543 00:17:45,120 --> 00:17:46,160 The Ping An Brain 544 00:17:46,160 --> 00:17:47,440 engine consists 545 00:17:47,440 --> 00:17:49,880 of three layers of platforms. 546 00:17:49,880 --> 00:17:52,600 The first layer is computing power platform, 547 00:17:52,600 --> 00:17:54,840 which manages and optimizes 548 00:17:54,840 --> 00:17:58,440 the massive GPU and CPU servers clusters 549 00:17:58,800 --> 00:18:01,640 to support the computing power requirements 550 00:18:01,640 --> 00:18:03,280 of AI model training 551 00:18:03,280 --> 00:18:05,640 and inference applications. 552 00:18:05,640 --> 00:18:08,200 The second layer is data platform, 553 00:18:08,200 --> 00:18:10,560 incorporating key processes 554 00:18:10,560 --> 00:18:12,440 such as data collection, 555 00:18:12,440 --> 00:18:14,000 merging, cleansing, 556 00:18:14,000 --> 00:18:16,520 labeling, storing, quality control, and so on. 557 00:18:16,520 --> 00:18:18,760 The thousands of data clusters 558 00:18:18,760 --> 00:18:20,360 in our data platform 559 00:18:20,360 --> 00:18:21,880 have accumulated 560 00:18:21,880 --> 00:18:24,760 tremendous amount of high-quality data 561 00:18:24,760 --> 00:18:26,200 in the domains of financial 562 00:18:26,200 --> 00:18:29,280 and medical services, both offline and online. 563 00:18:29,400 --> 00:18:29,960 This data 564 00:18:29,960 --> 00:18:31,920 not only help on business, 565 00:18:31,920 --> 00:18:34,920 such as better strategic decision making, 566 00:18:34,920 --> 00:18:36,400 but also create 567 00:18:36,400 --> 00:18:37,280 the data free 568 00:18:37,280 --> 00:18:40,320 will to continuously produce high-quality 569 00:18:40,320 --> 00:18:41,960 AI models. 570 00:18:41,960 --> 00:18:44,320 At the meantime, we always take data 571 00:18:44,320 --> 00:18:46,640 security and privacy protection 572 00:18:46,640 --> 00:18:48,240 as the highest priority. 573 00:18:48,240 --> 00:18:51,560 It is the red line that cannot be violated. 574 00:18:51,920 --> 00:18:54,920 Thus, we have developed comprehensive policies 575 00:18:54,960 --> 00:18:57,960 and systems for extremely rigorous 576 00:18:58,160 --> 00:19:01,160 privacy protection and data security, 577 00:19:01,520 --> 00:19:03,720 covering the finest activities 578 00:19:03,720 --> 00:19:05,360 in the data platform. 579 00:19:05,360 --> 00:19:08,360 In addition to the monitoring, inspection 580 00:19:08,640 --> 00:19:09,560 and protection systems, 581 00:19:09,560 --> 00:19:11,200 we specifically developed 582 00:19:11,200 --> 00:19:14,840 the 4-stack Privacy Computing Systems, BEEHIVE, 583 00:19:15,320 --> 00:19:17,400 which allows different organizations 584 00:19:17,400 --> 00:19:20,720 to share their data efficiently and effectively 585 00:19:21,000 --> 00:19:24,000 in a compliant and secure manner. 586 00:19:24,000 --> 00:19:24,400 he third 587 00:19:24,400 --> 00:19:25,880 layer is AI platform, 588 00:19:25,880 --> 00:19:28,360 which consists of three sub-layers. 589 00:19:28,360 --> 00:19:30,040 The first is the model layer, 590 00:19:30,040 --> 00:19:33,080 which stores and well-manages all the models 591 00:19:33,240 --> 00:19:34,160 we have trained 592 00:19:34,160 --> 00:19:35,600 and tested in the past, 593 00:19:35,600 --> 00:19:38,200 covering a wide range of AI areas, 594 00:19:38,200 --> 00:19:39,240 such as large-range 595 00:19:39,240 --> 00:19:40,240 models, computer 596 00:19:40,240 --> 00:19:41,160 vision, speech 597 00:19:41,160 --> 00:19:43,280 recognition and synthesis, data 598 00:19:43,280 --> 00:19:46,280 analytics, micro-expression, and so on. 599 00:19:46,400 --> 00:19:48,240 The model number is innumerable. 600 00:19:48,240 --> 00:19:50,360 Specifically for large-range models, 601 00:19:50,360 --> 00:19:51,360 we constructed 602 00:19:51,360 --> 00:19:52,760 three-tier architecture, 603 00:19:52,760 --> 00:19:54,360 general-purpose large-range 604 00:19:54,360 --> 00:19:57,280 models, domain-specific large-range models, 605 00:19:57,280 --> 00:20:00,040 and specialized scenario large-range models. 606 00:20:00,040 --> 00:20:01,640 The second layer provides 607 00:20:01,640 --> 00:20:02,320 all the model 608 00:20:02,320 --> 00:20:03,080 training, 609 00:20:03,080 --> 00:20:06,080 inference, operation and maintenance tools 610 00:20:06,080 --> 00:20:07,120 and systems. 611 00:20:07,120 --> 00:20:09,920 Only with the strong support of these tools, 612 00:20:09,920 --> 00:20:11,360 all the models in the model 613 00:20:11,360 --> 00:20:14,360 layer can be produced, utilized and maintained 614 00:20:14,360 --> 00:20:17,360 at such a high standard quality and efficiency 615 00:20:17,360 --> 00:20:18,320 at a large scale. 616 00:20:18,320 --> 00:20:19,960 The third sub-layer is a platform 617 00:20:19,960 --> 00:20:22,600 providing agentic capabilities, 618 00:20:22,600 --> 00:20:24,560 which allows every member of the Ping 619 00:20:24,560 --> 00:20:24,840 An group 620 00:20:24,840 --> 00:20:28,000 to build their own agents freely, efficiently 621 00:20:28,000 --> 00:20:29,000 and effectively, 622 00:20:29,000 --> 00:20:31,200 without the need of a technical background. 623 00:20:31,200 --> 00:20:33,120 Wow, I mean, Ping An Brain, 624 00:20:33,120 --> 00:20:36,400 Beehive, Robot Assistants, it sounds very, 625 00:20:36,400 --> 00:20:37,640 very exciting, 626 00:20:37,640 --> 00:20:39,440 but this isn't science fiction, right? 627 00:20:39,440 --> 00:20:40,640 This is how Ping An's business 628 00:20:40,640 --> 00:20:42,880 is transforming right now. 629 00:20:42,880 --> 00:20:44,880 So, how about some specific examples 630 00:20:44,880 --> 00:20:46,840 about how Gen AI applications 631 00:20:46,840 --> 00:20:48,320 are supporting Ping An's businesses? 632 00:20:48,320 --> 00:20:51,280 Well, let me just give you an example, 633 00:20:51,280 --> 00:20:54,200 an application that we all use every day. 634 00:20:54,200 --> 00:20:57,440 We have this app called Ask Bob, 635 00:20:57,720 --> 00:21:00,560 which is essentially an AI-powered tool. 636 00:21:00,560 --> 00:21:01,880 So the interaction model 637 00:21:01,880 --> 00:21:03,200 is very much like these days 638 00:21:03,200 --> 00:21:05,640 you have with chat GPT, for example. 639 00:21:05,640 --> 00:21:08,160 You ask Bob a question, 640 00:21:08,160 --> 00:21:11,120 then Bob will come back with some answers, 641 00:21:11,120 --> 00:21:13,400 and those answers are generated 642 00:21:13,400 --> 00:21:15,400 based on the massive data 643 00:21:15,400 --> 00:21:16,880 that we have accumulated 644 00:21:16,880 --> 00:21:18,280 and also the knowledge base 645 00:21:18,280 --> 00:21:19,640 that we have built up 646 00:21:19,640 --> 00:21:22,640 internally based on our proprietary data 647 00:21:22,640 --> 00:21:25,240 in the financial services area, in the health 648 00:21:25,240 --> 00:21:25,920 and the senior care 649 00:21:25,920 --> 00:21:29,760 area, around medical, hospital prescriptions, 650 00:21:30,000 --> 00:21:31,200 et cetera, et cetera. 651 00:21:31,200 --> 00:21:33,120 So that, powered by the Gen 652 00:21:33,120 --> 00:21:34,160 AI technology, 653 00:21:34,160 --> 00:21:36,560 we're able to make Bob more like a human-like. 654 00:21:36,560 --> 00:21:39,040 It’s almost like an AI assistant. 655 00:21:39,040 --> 00:21:40,560 This is actually very interesting 656 00:21:40,560 --> 00:21:42,480 and especially helpful 657 00:21:42,480 --> 00:21:44,400 for our Salesforce people. 658 00:21:44,400 --> 00:21:46,280 Now, when they interact with the customers, 659 00:21:46,280 --> 00:21:48,080 Bob is able to come up 660 00:21:48,080 --> 00:21:49,160 with the first round 661 00:21:49,160 --> 00:21:51,000 of proposals, for example, 662 00:21:51,000 --> 00:21:52,200 based on different customer 663 00:21:52,200 --> 00:21:53,000 needs 664 00:21:53,000 --> 00:21:54,760 and their existing current 665 00:21:54,760 --> 00:21:56,040 financial situation. 666 00:21:56,040 --> 00:21:58,680 So that significantly improves 667 00:21:58,680 --> 00:22:01,680 our sales capabilities and their expertise 668 00:22:01,760 --> 00:22:03,440 and also the efficiency. 669 00:22:03,440 --> 00:22:05,480 So now we've seen very good 670 00:22:05,480 --> 00:22:08,240 uplift of our sales agents’ productivity, 671 00:22:08,240 --> 00:22:09,480 thanks to this Bob. 672 00:22:09,480 --> 00:22:11,200 And AI and Gen 673 00:22:11,200 --> 00:22:12,560 AI is really starting 674 00:22:12,560 --> 00:22:15,520 to come into everybody's daily life now. 675 00:22:15,520 --> 00:22:17,080 We've heard so much about open source 676 00:22:17,080 --> 00:22:19,000 gen AI in recent months. 677 00:22:19,000 --> 00:22:20,360 Things such as Deep Seek. 678 00:22:20,360 --> 00:22:21,480 So is Ping An using 679 00:22:21,480 --> 00:22:22,640 those kinds of technologies 680 00:22:22,640 --> 00:22:24,080 in its products and services, Jing? 681 00:22:24,080 --> 00:22:24,400 Yeah. 682 00:22:24,400 --> 00:22:27,680 As we discussed earlier, our large-language 683 00:22:27,680 --> 00:22:30,680 model platform has a three-tier architecture, 684 00:22:30,840 --> 00:22:32,720 general-purpose large-language 685 00:22:32,720 --> 00:22:35,360 models, domain-specific large-language models, 686 00:22:35,360 --> 00:22:37,400 and a specialized scenario 687 00:22:37,400 --> 00:22:38,640 large-language models. 688 00:22:38,640 --> 00:22:39,200 And yes, 689 00:22:39,200 --> 00:22:41,280 we do have incorporated 690 00:22:41,280 --> 00:22:43,960 DeepSeek models into the first tier 691 00:22:43,960 --> 00:22:45,400 as one category 692 00:22:45,400 --> 00:22:47,840 of the general-purpose large-language models. 693 00:22:47,840 --> 00:22:49,960 Of course, the DeepSeek-backed 694 00:22:49,960 --> 00:22:51,440 large-language models 695 00:22:51,440 --> 00:22:54,440 do not behave better in all the scenarios. 696 00:22:54,640 --> 00:22:56,520 For example, in some scenarios, 697 00:22:56,520 --> 00:22:58,520 the tasks are relatively simple, 698 00:22:58,520 --> 00:23:01,520 where slow thinking is not necessary 699 00:23:01,760 --> 00:23:03,920 and fast thinking is good enough. 700 00:23:03,920 --> 00:23:05,000 Then DeepSeek-type 701 00:23:05,000 --> 00:23:07,080 models are not the right choice 702 00:23:07,080 --> 00:23:10,440 due to the higher cost and possibly worse result. 703 00:23:10,440 --> 00:23:10,960 In fact, 704 00:23:10,960 --> 00:23:11,720 we found that 705 00:23:11,720 --> 00:23:15,680 only less than 40% of our existing AI solutions 706 00:23:15,880 --> 00:23:16,880 could be improved 707 00:23:16,880 --> 00:23:19,880 by using the DeepSeek-backed models. 708 00:23:19,880 --> 00:23:21,160 On the other hand, 709 00:23:21,160 --> 00:23:24,000 language models such as DeepSeek is only one 710 00:23:24,000 --> 00:23:26,040 kind of generative AI models 711 00:23:26,040 --> 00:23:27,920 which targeted at text data. 712 00:23:27,920 --> 00:23:28,920 In practice, 713 00:23:28,920 --> 00:23:31,280 there are other modalities of data 714 00:23:31,280 --> 00:23:31,800 that need 715 00:23:31,800 --> 00:23:33,920 to be understood or analyzed, 716 00:23:33,920 --> 00:23:36,360 such as images, audios, and videos. 717 00:23:36,360 --> 00:23:38,800 Therefore, we are also developing generative 718 00:23:38,800 --> 00:23:39,800 AI models 719 00:23:39,800 --> 00:23:42,160 targeting at other modalities 720 00:23:42,160 --> 00:23:44,840 and multiple modalities together. 721 00:23:44,840 --> 00:23:46,680 Ping An Brain Engine organizes 722 00:23:46,680 --> 00:23:47,400 and operates 723 00:23:47,400 --> 00:23:48,920 these generative models 724 00:23:48,920 --> 00:23:51,000 systematically to help achieve 725 00:23:51,000 --> 00:23:52,160 better business values. 726 00:23:52,160 --> 00:23:53,360 It sounds like 727 00:23:53,360 --> 00:23:55,160 there is an incredible amount of work 728 00:23:55,160 --> 00:23:56,720 happening behind the scenes. 729 00:23:56,720 --> 00:23:58,920 But if we can zoom out a little bit, Michael, 730 00:23:58,920 --> 00:23:59,880 can you tell us a little bit 731 00:23:59,880 --> 00:24:01,640 about how Ping An actually assesses 732 00:24:01,640 --> 00:24:03,440 ROI on technology investments? 733 00:24:03,440 --> 00:24:05,920 Yeah, well, that's a good question 734 00:24:05,920 --> 00:24:08,720 because technology can be very costly 735 00:24:08,720 --> 00:24:10,800 as these days we all heard about, 736 00:24:10,800 --> 00:24:11,000 you know, 737 00:24:11,000 --> 00:24:12,160 big companies pouring 738 00:24:12,160 --> 00:24:15,160 billions of dollars onto technology, onto chips 739 00:24:15,440 --> 00:24:17,600 and onto a computing power. So are we. 740 00:24:17,600 --> 00:24:19,600 As one of the largest financial institutions 741 00:24:19,600 --> 00:24:20,120 in the world, 742 00:24:20,120 --> 00:24:21,440 we invest in technology 743 00:24:21,440 --> 00:24:23,600 with billions of dollars every year. 744 00:24:23,600 --> 00:24:25,520 But being a financial institution, 745 00:24:25,520 --> 00:24:26,640 we are very much focused 746 00:24:26,640 --> 00:24:28,520 on getting value out of our investment. 747 00:24:28,520 --> 00:24:31,200 So we usually take a very disciplined 748 00:24:31,200 --> 00:24:32,440 and value-driven approach 749 00:24:32,440 --> 00:24:34,400 looking at our technology investment. 750 00:24:34,400 --> 00:24:35,600 In general, 751 00:24:35,600 --> 00:24:37,160 we put our technology investment 752 00:24:37,160 --> 00:24:38,480 into three buckets. 753 00:24:38,480 --> 00:24:41,480 The first bucket is what we call must-have. 754 00:24:41,720 --> 00:24:43,000 So those are the investments 755 00:24:43,000 --> 00:24:46,040 that go into, say, technology infrastructure, 756 00:24:46,040 --> 00:24:47,040 the data centers, 757 00:24:47,040 --> 00:24:49,240 the architecture of the new, 758 00:24:49,240 --> 00:24:50,440 next-generation architecture 759 00:24:50,440 --> 00:24:51,240 for data, et cetera. 760 00:24:51,240 --> 00:24:53,800 So those are the things that we believe 761 00:24:53,800 --> 00:24:55,040 are fundamental 762 00:24:55,040 --> 00:24:57,560 that would give us competitive advantage 763 00:24:57,560 --> 00:24:58,880 for the long term. 764 00:24:58,880 --> 00:24:59,960 And those things, we’re 765 00:24:59,960 --> 00:25:01,240 just putting money 766 00:25:01,240 --> 00:25:04,240 without looking at the return in general. 767 00:25:04,240 --> 00:25:07,880 Now, the second bucket is specific investment 768 00:25:07,880 --> 00:25:11,600 into applications, into use cases, et cetera. 769 00:25:11,600 --> 00:25:12,960 And there, we're very much 770 00:25:12,960 --> 00:25:14,760 focused on the ROI. 771 00:25:14,760 --> 00:25:15,920 We have to look at 772 00:25:15,920 --> 00:25:16,920 what kind of a return 773 00:25:16,920 --> 00:25:17,480 on the business 774 00:25:17,480 --> 00:25:19,080 impact or value creation 775 00:25:19,080 --> 00:25:20,840 from that particular project 776 00:25:20,840 --> 00:25:23,640 and what kind of ROI that we could qualitative 777 00:25:23,640 --> 00:25:25,280 and quantitatively measure. 778 00:25:25,280 --> 00:25:27,040 And then we have a rigorous process 779 00:25:27,040 --> 00:25:28,840 to check and making sure 780 00:25:28,840 --> 00:25:32,480 the expected return are able to be coming back. 781 00:25:33,080 --> 00:25:35,240 The last bucket is what we call 782 00:25:35,240 --> 00:25:36,600 innovation per se. 783 00:25:36,600 --> 00:25:39,560 Those are the kind of trial and error exercises 784 00:25:39,560 --> 00:25:40,880 where we say every year 785 00:25:40,880 --> 00:25:43,080 we allocate a fixed amount of money 786 00:25:43,080 --> 00:25:44,600 into those experiments, 787 00:25:44,600 --> 00:25:47,000 we're willing to put in money and we're willing 788 00:25:47,000 --> 00:25:48,440 to, you know, quote unquote, 789 00:25:48,440 --> 00:25:50,440 waste those money if we didn't succeed. 790 00:25:50,440 --> 00:25:52,000 But because of that, here 791 00:25:52,000 --> 00:25:52,720 and there, now 792 00:25:52,720 --> 00:25:53,800 and then, we're able to actually 793 00:25:53,800 --> 00:25:55,400 get breakthrough into some of the new 794 00:25:55,400 --> 00:25:57,040 cutting edge technologies. 795 00:25:57,040 --> 00:25:58,640 So that's how we measure and manage 796 00:25:58,640 --> 00:26:00,200 our technology investments. 797 00:26:00,200 --> 00:26:01,120 Michael, Jing, 798 00:26:01,120 --> 00:26:02,720 thank you so much for your time today. 799 00:26:02,720 --> 00:26:04,040 It's been great to gain 800 00:26:04,040 --> 00:26:04,960 all of these insights 801 00:26:04,960 --> 00:26:06,280 about what's happening with Ping An’s, 802 00:26:06,280 --> 00:26:09,120 technology foundations and their use of AI. 803 00:26:10,160 --> 00:26:10,960 Thank you so much. 804 00:26:10,960 --> 00:26:13,080 Thank you for having us. Been a pleasure.