1 00:00:00,000 --> 00:00:06,100 We're discussing personalization for e-commerce and Retail with 2 00:00:06,100 --> 00:00:11,000 Sean Milani, the chief technology officer of algo Lea. 3 00:00:11,400 --> 00:00:14,700 We power it search and browse and recommendation experiences 4 00:00:15,100 --> 00:00:18,200 across 17 thousand websites all over the internet. 5 00:00:18,200 --> 00:00:20,600 Where actually the second biggest search engine in the 6 00:00:20,600 --> 00:00:26,200 world behind Google, we power over 1.75 trillion requests 7 00:00:26,200 --> 00:00:29,300 every single year. So a very exciting powering a 8 00:00:29,308 --> 00:00:30,600 lot of the fun. Datian the internet. 9 00:00:30,700 --> 00:00:34,300 I have to say that at cxo, talk. We're actually one of your 10 00:00:34,300 --> 00:00:37,200 customers we use. I'll go Leah, on cxo talk.com. 11 00:00:37,200 --> 00:00:41,400 It's great to talk to one of our customers, and it's great to 12 00:00:41,408 --> 00:00:43,000 hear that. We can help our your experience 13 00:00:43,000 --> 00:00:46,900 as well. You're focused on e-commerce and 14 00:00:46,900 --> 00:00:48,900 Retail. So why don't you give us just 15 00:00:48,900 --> 00:00:52,200 the high-level overview of that? We actually serve customers 16 00:00:52,200 --> 00:00:55,100 across a whole range of Industries, but e-commerce is 17 00:00:55,100 --> 00:00:57,700 obviously the one of the ones that were best known for the 18 00:00:57,700 --> 00:01:00,100 reason is because search is so important as part of Whole 19 00:01:00,100 --> 00:01:03,800 Discovery journey in e-commerce, people love e-commerce stores. 20 00:01:03,800 --> 00:01:05,900 And one of the reasons is because they get such an 21 00:01:05,900 --> 00:01:09,700 incredibly wide selection of products that they can shop for. 22 00:01:10,000 --> 00:01:12,900 But it's also hugely overwhelming for a shopper. 23 00:01:13,000 --> 00:01:16,300 It would be like walking into a physical store and it's the size 24 00:01:16,300 --> 00:01:19,500 of like, a huge Stadium. You would just have million 25 00:01:19,500 --> 00:01:22,600 items to choose from but it's it's very overwhelming. 26 00:01:22,700 --> 00:01:25,600 They call this the Paradox of choice and more choice. 27 00:01:25,600 --> 00:01:29,100 You provide a customer often the less satisfied they are or the 28 00:01:29,100 --> 00:01:31,500 more overwhelmed They are and their experience can actually 29 00:01:31,500 --> 00:01:34,300 deteriorate with more choices. E-commerce brings a huge 30 00:01:34,300 --> 00:01:36,400 opportunity. But it's also means that we have 31 00:01:36,400 --> 00:01:38,400 to have new tools and Technologies to be able to help 32 00:01:38,400 --> 00:01:41,600 customers and guide them through this experience, showing how do 33 00:01:41,600 --> 00:01:47,800 advancements in Ai and machine learning such as chat GPT 34 00:01:48,400 --> 00:01:52,300 effect, this search landscape as you've been describing he hopes 35 00:01:52,300 --> 00:01:56,000 that around 20-30 years now but the experience to a shopper 36 00:01:56,000 --> 00:01:58,100 still feels very much. Like it could come out of the 37 00:01:58,100 --> 00:02:01,700 1990s during the.com. If a lot of e-commerce, sites 38 00:02:01,700 --> 00:02:05,900 are really just a product database or a product catalog 39 00:02:05,900 --> 00:02:07,800 that have a user experience on it. 40 00:02:07,800 --> 00:02:10,199 You still have to talk to the website and the way you would 41 00:02:10,199 --> 00:02:12,800 talk to a computer, right? You have to type in like 42 00:02:12,800 --> 00:02:16,400 specific words, you have to click on different filters to 43 00:02:16,400 --> 00:02:18,900 kind of narrow down the selection and this is going to 44 00:02:18,908 --> 00:02:22,300 change dramatically now. So a lot of these language 45 00:02:22,300 --> 00:02:26,600 models that are powering chat GPT of exploded in size and 46 00:02:26,600 --> 00:02:29,800 sophistication over the last few years and there. 47 00:02:29,900 --> 00:02:33,000 Now able to actually really understand humans in a way in 48 00:02:33,000 --> 00:02:36,000 which humans can use not natural language and understands the 49 00:02:36,000 --> 00:02:38,800 intent and the concepts behind what they're looking for. 50 00:02:39,200 --> 00:02:45,000 Sean can you elaborate on? How may I help some make this 51 00:02:45,000 --> 00:02:49,700 shopping experience, better for the user, you don't need to be 52 00:02:49,700 --> 00:02:51,600 able to talk to the computer anymore. 53 00:02:51,600 --> 00:02:53,200 The computer can actually understand you. 54 00:02:53,300 --> 00:02:56,000 So what we're seeing is people are using far more expressive 55 00:02:56,000 --> 00:02:58,300 language. Now, the number of words that 56 00:02:58,300 --> 00:03:01,000 they're typing into the search box, Is becoming significantly 57 00:03:01,000 --> 00:03:02,800 larger. It's like doubled in size in the 58 00:03:02,808 --> 00:03:06,200 last couple of years, but also people are expecting experiences 59 00:03:06,200 --> 00:03:09,100 that are far more sophisticated, and personalized. 60 00:03:09,200 --> 00:03:13,500 As I enter in searches or I click on filters or a view 61 00:03:13,500 --> 00:03:17,300 products, it expects the experience to adapt to me to 62 00:03:17,300 --> 00:03:20,600 really understand what I'm looking for to remember me, when 63 00:03:20,600 --> 00:03:25,200 I turn up to the site, Shawn you've described how natural 64 00:03:25,200 --> 00:03:29,500 language search just makes it easier for end users. 65 00:03:30,000 --> 00:03:32,500 Your Chief technology officer of algo. 66 00:03:32,500 --> 00:03:38,100 Leah, can you give us a glimpse behind the scenes of how this is 67 00:03:38,100 --> 00:03:40,500 possible? There's been a real Quantum Leap 68 00:03:40,500 --> 00:03:44,400 and what we're able to do over the last year or two, as Chachi 69 00:03:44,400 --> 00:03:48,500 beauty is really demonstrated to the world that you can interface 70 00:03:48,800 --> 00:03:50,600 with users using natural language. 71 00:03:50,600 --> 00:03:53,600 Both as the inputs to ask questions but also as an output 72 00:03:53,800 --> 00:03:55,200 to be able to answer their questions. 73 00:03:55,300 --> 00:03:58,300 But we're fundamentally changing the way in which we search and 74 00:03:58,300 --> 00:04:01,400 retrieve information with these Large language models in these 75 00:04:01,400 --> 00:04:04,400 new breakthroughs. So, instead of using the actual 76 00:04:04,400 --> 00:04:07,200 word, we're now actually taking words and we're turning them 77 00:04:07,200 --> 00:04:10,800 into something called a vector. And the vector captures, the 78 00:04:10,800 --> 00:04:15,200 concept, and the Nuance behind, either the word or the phrase of 79 00:04:15,200 --> 00:04:17,700 the sentence or your question. That means that we can then go 80 00:04:17,700 --> 00:04:21,500 and search through all the products are all the web pages 81 00:04:21,700 --> 00:04:24,500 for other vectors. We turn everything in the index 82 00:04:24,500 --> 00:04:27,500 into a vector and then it becomes this kind of exercise 83 00:04:27,500 --> 00:04:29,400 where we try to find similar vectors. 84 00:04:29,900 --> 00:04:32,600 Just like you use words and try to match them with similar words 85 00:04:32,600 --> 00:04:34,700 and web pages. But it means that you can do 86 00:04:34,700 --> 00:04:36,900 really, really interesting things to give you a really 87 00:04:36,900 --> 00:04:40,600 powerful example is if let's say you go to a website and you 88 00:04:40,608 --> 00:04:43,000 actually want to shop for a specific brand. 89 00:04:43,000 --> 00:04:47,000 So let's say I want to North Face jacket and if the site 90 00:04:47,000 --> 00:04:50,600 doesn't sell that brand, right? It's able to understand the 91 00:04:50,600 --> 00:04:53,700 concept behind the brand. So understands North Face sells 92 00:04:53,700 --> 00:04:56,800 outdoor jackets and it can Surface other outdoor jackets 93 00:04:56,800 --> 00:04:59,800 that are similar to North Face. We can actually find 94 00:04:59,900 --> 00:05:02,000 Specifically designed to solve those problems. 95 00:05:02,300 --> 00:05:05,400 Even if the words that you're using, when you describe the 96 00:05:05,400 --> 00:05:07,400 problem, don't match the words in the product. 97 00:05:07,700 --> 00:05:12,200 How is this different from traditional search, whether on 98 00:05:12,200 --> 00:05:17,200 e-commerce sites or just in just broadly, traditional search is 99 00:05:17,200 --> 00:05:20,000 literally just taking the literal words that you type in 100 00:05:20,400 --> 00:05:23,700 and trying to find products that use those literal words. 101 00:05:23,700 --> 00:05:26,400 You know, a good example is if you wanted to search for 102 00:05:26,500 --> 00:05:29,700 chocolate milk or milk chocolate, these are two terms 103 00:05:29,700 --> 00:05:32,900 of exactly the same words, but very different categories to 104 00:05:32,900 --> 00:05:35,200 search for. And so it can be very, very 105 00:05:35,200 --> 00:05:38,000 confusing sometimes for these search engines to be able to 106 00:05:38,000 --> 00:05:41,600 disambiguate the actual words. But when we can translate them 107 00:05:41,600 --> 00:05:45,300 into very specific Concepts and really understand the intent 108 00:05:45,300 --> 00:05:47,900 behind them. Then we're going to get far 109 00:05:47,900 --> 00:05:51,700 better more accurate results. And I have to say, some of our 110 00:05:51,900 --> 00:05:55,000 early customers are just seen incredible increases in the 111 00:05:55,000 --> 00:05:58,800 amount of conversions, and the amount of things that Shoppers 112 00:05:58,800 --> 00:06:01,400 are able to buy because They're able to find it. 113 00:06:01,600 --> 00:06:04,600 Sean you mentioned Vector search. 114 00:06:04,600 --> 00:06:08,800 It's a topic that has been around for a long time yet. 115 00:06:08,800 --> 00:06:12,300 It's not widely discussed. Why is that? 116 00:06:12,300 --> 00:06:15,100 And what are you doing with this? 117 00:06:15,300 --> 00:06:17,400 We've known that vectors are a better way to represent. 118 00:06:17,400 --> 00:06:21,200 Concepts Than Words, we've known vectors work from a scientific 119 00:06:21,200 --> 00:06:26,100 perspective and we can get great results in the lab but it turns 120 00:06:26,100 --> 00:06:28,000 out that they're really hard to scale. 121 00:06:28,200 --> 00:06:33,000 Like I mentioned we've Got we do about 1.75 trillion, search 122 00:06:33,000 --> 00:06:37,700 requests a year, across 17,000, different websites and to be 123 00:06:37,700 --> 00:06:42,100 able to, like, apply vectors on. Every single query is really, 124 00:06:42,100 --> 00:06:44,900 really challenging. Vectors are very, very 125 00:06:45,000 --> 00:06:48,100 computationally, heavy it. Take up a lot of memory both in 126 00:06:48,100 --> 00:06:50,800 the server but also when you're training and storing them and 127 00:06:50,800 --> 00:06:53,600 they're actually pretty slow and expensive to roll out. 128 00:06:53,600 --> 00:06:57,300 The Breakthrough that we've recently had is we figured out 129 00:06:57,300 --> 00:07:00,300 how to compress vectors, so we can actually compress Them into 130 00:07:00,300 --> 00:07:03,300 like a 10 times smaller format, keeping the same kind of 131 00:07:03,300 --> 00:07:08,000 relevancy and we can do it at extremely high speeds. 132 00:07:08,200 --> 00:07:11,800 We can take every single query across as one point seven five 133 00:07:11,800 --> 00:07:14,300 trillion and we can apply Vector search them. 134 00:07:14,300 --> 00:07:16,900 This is the big breakthrough. This Hashem technology that 135 00:07:16,900 --> 00:07:19,800 we've built means that for the real world production 136 00:07:19,800 --> 00:07:22,600 environments. When you need high speed, you 137 00:07:22,600 --> 00:07:25,600 need a reasonable cost and you need really big scale. 138 00:07:25,800 --> 00:07:29,600 So this is what makes it practical for you to. 139 00:07:29,800 --> 00:07:35,700 Use vectors in this, highly real-time shopping environment, 140 00:07:35,900 --> 00:07:39,500 every single query for our customers, we're able to use 141 00:07:39,500 --> 00:07:43,900 both factor and keyword in a single kind of API call. 142 00:07:43,900 --> 00:07:46,600 And by using both of these strategies, we actually get a 143 00:07:46,600 --> 00:07:50,200 lot more information back and we can use the extra information. 144 00:07:50,200 --> 00:07:52,700 Like how many keywords don't match but also how did the 145 00:07:52,700 --> 00:07:55,500 vectors match to do a much better ranking? 146 00:07:55,600 --> 00:07:56,800 And I think that's pretty unique. 147 00:07:56,800 --> 00:07:59,700 I no one else in the industry is able to do this at this size. 148 00:07:59,900 --> 00:08:03,000 Scale and speed that we're doing enough, because we're using this 149 00:08:03,000 --> 00:08:04,500 kind of technique called hashing. 150 00:08:04,900 --> 00:08:12,100 So, the technology is enabling a simpler easier user experience, 151 00:08:12,100 --> 00:08:13,800 that's at the same time, more effective. 152 00:08:14,200 --> 00:08:18,000 What we've seen is about 70% of all of the search queries are 153 00:08:18,000 --> 00:08:21,100 actually far more complicated one-off type queries, where 154 00:08:21,100 --> 00:08:23,300 people are asking for very specific things, they have a 155 00:08:23,300 --> 00:08:27,200 question and that kind of 70% of queers are going, very 156 00:08:27,200 --> 00:08:30,500 unanswered at the moment because they're very Uncle to answer 157 00:08:30,500 --> 00:08:33,799 with just matching words. So this kind of new AI Vector 158 00:08:33,799 --> 00:08:38,200 search is really monetizing and helping solve the customers 159 00:08:38,200 --> 00:08:42,200 needs in that 70% of a long tail case and it actually translates 160 00:08:42,200 --> 00:08:46,200 into a lot more sales. Sean we're does personalization 161 00:08:46,200 --> 00:08:51,100 fit into this landscape creating a loyal customers. 162 00:08:51,100 --> 00:08:53,600 One of the most important parts about creating a great business 163 00:08:54,000 --> 00:08:57,700 because you pay to acquire a customer the first time and if 164 00:08:57,700 --> 00:09:00,100 they either don't convert that first time because they Find 165 00:09:00,100 --> 00:09:03,100 something, or they end up, not coming back for a second or 166 00:09:03,100 --> 00:09:06,500 third shopping Journey. Then the economics of acquiring, 167 00:09:06,500 --> 00:09:09,300 the customer become pretty unattractive. 168 00:09:09,300 --> 00:09:12,500 How do I create an experience for a shopper where I recognize 169 00:09:12,500 --> 00:09:14,700 that there are a customer who's been to the store before? 170 00:09:15,600 --> 00:09:18,500 How do I make it feel? So that when a customer comes to 171 00:09:18,500 --> 00:09:21,700 my store, that they're going to get a superior experience, the 172 00:09:21,700 --> 00:09:25,700 second and third time versus going to a competitor store and 173 00:09:25,700 --> 00:09:27,100 that's all about personalization. 174 00:09:27,300 --> 00:09:31,200 Once I've seen what a customer prefers Purrs either through the 175 00:09:31,200 --> 00:09:34,800 searches that they make, on the website, or the products that 176 00:09:34,800 --> 00:09:37,500 they click on the products they buy, or the products that they 177 00:09:37,500 --> 00:09:39,600 click on a. Don't buy, you can observe the 178 00:09:39,600 --> 00:09:42,400 entire behavior of the customers, they're shopping and 179 00:09:42,400 --> 00:09:45,700 you can learn from it, and you can do that in two ways. 180 00:09:46,200 --> 00:09:49,300 The first way is when they actually appear for the first 181 00:09:49,300 --> 00:09:54,100 time, you can really hang on every single, click every single 182 00:09:54,200 --> 00:09:56,900 word that they type in and category that they look at. 183 00:09:57,300 --> 00:10:00,500 And in real time, even for a shopper that you've never Series 184 00:10:00,500 --> 00:10:03,500 never been to the store before. You can start to adapt. 185 00:10:03,500 --> 00:10:06,300 This is the called real-time personalization, and it's kind 186 00:10:06,300 --> 00:10:08,300 of like a cold start problem because you really don't know 187 00:10:08,300 --> 00:10:11,800 anything about the customer, but I can pretty quickly find out if 188 00:10:11,800 --> 00:10:14,200 their brands that they are clicking on more or if their 189 00:10:14,200 --> 00:10:16,400 price points that they're gravitating towards. 190 00:10:16,700 --> 00:10:21,100 And then I can start to show and put more of these types of 191 00:10:21,108 --> 00:10:25,100 products around brands or prices or categories in front of them 192 00:10:25,100 --> 00:10:27,700 as they click and discover around the site. 193 00:10:27,900 --> 00:10:29,600 The second really important thing is once they've actually 194 00:10:29,800 --> 00:10:32,300 Bought from you, the next time they come back, you acknowledge 195 00:10:32,300 --> 00:10:33,700 that. They are a customer that you've 196 00:10:33,700 --> 00:10:36,900 seen before and you're able to offer them again. 197 00:10:36,900 --> 00:10:40,000 Similar personalized experience and we've seen that these types 198 00:10:40,000 --> 00:10:44,200 of personalization algorithms and personalized, experiences 199 00:10:44,500 --> 00:10:47,100 can drive substantial increases in conversions. 200 00:10:47,100 --> 00:10:50,800 On the second and third visit, as well as this kind of 201 00:10:50,800 --> 00:10:52,600 real-time when they're shopping the first time. 202 00:10:52,700 --> 00:10:55,900 So I think it's a very important part of the whole customer 203 00:10:55,900 --> 00:10:58,900 lifecycle from acquisition first shopping. 204 00:10:59,800 --> 00:11:02,000 Third, and creating a long-term loyal customer. 205 00:11:02,300 --> 00:11:06,000 So you're talking about two kinds of personalization one. 206 00:11:06,000 --> 00:11:09,400 Is the cold start? As you describe when a visitor, 207 00:11:09,400 --> 00:11:14,900 First shows up at your site, but then remembering that visitor, 208 00:11:14,900 --> 00:11:18,500 when they come back the second time, the third time and 209 00:11:18,600 --> 00:11:21,900 creating an even more tailored and personalized customer 210 00:11:21,900 --> 00:11:23,500 experience for them. What is this? 211 00:11:23,500 --> 00:11:29,200 Do to the overall customer lifetime value? 212 00:11:29,500 --> 00:11:33,000 You A lot of people when they first do their online experience 213 00:11:33,000 --> 00:11:37,300 and move online, they start to focus really on that very first 214 00:11:37,400 --> 00:11:40,000 experience of my put some advertising out. 215 00:11:40,000 --> 00:11:43,000 It cost me X, those Shoppers turned up to the site and they 216 00:11:43,000 --> 00:11:46,200 bought why, you know, here's my return on investment. 217 00:11:46,700 --> 00:11:49,700 But as you start to build a more loyal following and you start to 218 00:11:49,700 --> 00:11:51,900 become a little bit more sophisticated in the way, you're 219 00:11:51,900 --> 00:11:55,100 thinking about your online business, you start to look at 220 00:11:55,100 --> 00:11:58,700 the return over a much longer Lifetime with a customer. 221 00:11:58,900 --> 00:12:01,800 You take the whole Time value. And then you set that against 222 00:12:01,800 --> 00:12:05,600 the acquisition cost and what it means is, if you create loyal 223 00:12:05,600 --> 00:12:08,600 customers, you can spend a higher dollar to acquire a 224 00:12:08,600 --> 00:12:10,700 customer. And so this is really like how 225 00:12:10,700 --> 00:12:14,100 to create a scalable customer, acquisition business model 226 00:12:14,100 --> 00:12:18,300 online is trying to get your cost per customer acquisition 227 00:12:18,300 --> 00:12:19,600 and your customer lifetime value. 228 00:12:19,600 --> 00:12:23,500 And the ratio of these two numbers, you want to get that 229 00:12:23,500 --> 00:12:25,500 very high. So high customer lifetime value 230 00:12:25,500 --> 00:12:29,200 low customer acquisition, how is this different from the 231 00:12:29,200 --> 00:12:32,500 traditional? Additional concept of customer 232 00:12:32,500 --> 00:12:36,300 Journeys and looking at the total lifetime value of a 233 00:12:36,308 --> 00:12:39,200 customer. Typically people will look in 234 00:12:39,200 --> 00:12:42,300 session. So the only look at the specific 235 00:12:42,300 --> 00:12:45,900 shopping session that the customer is in Manila. 236 00:12:45,900 --> 00:12:49,700 Tribute basically all of the value against the acquisition 237 00:12:49,700 --> 00:12:53,700 cost for the customer but as you start to create much more a, I 238 00:12:53,700 --> 00:12:56,900 powered experiences and personalization, you can start 239 00:12:56,900 --> 00:13:01,300 to account for a much larger amount of It's been created when 240 00:13:01,300 --> 00:13:03,500 the customer is acquired, over more and more sessions. 241 00:13:03,900 --> 00:13:11,000 So, the technology enables this broader Horizon and more 242 00:13:11,000 --> 00:13:15,400 accurate understanding of the customer Behavior. 243 00:13:15,600 --> 00:13:20,100 So, that in effect, you can amortize your initial customer 244 00:13:20,100 --> 00:13:25,600 acquisition cost over this longer broader time Horizon, a 245 00:13:25,600 --> 00:13:28,400 great e-commerce stores one where you spend, you know, 246 00:13:28,600 --> 00:13:31,500 potentially even more money acquiring customers because 247 00:13:31,500 --> 00:13:34,500 those customers are going to spend over a series of months 248 00:13:34,500 --> 00:13:37,400 and years with you and they're going to become loyal and 249 00:13:37,400 --> 00:13:41,300 generate long-term profitability on an individual customer 250 00:13:41,300 --> 00:13:44,700 perspective. So think of it like a PL for a 251 00:13:44,700 --> 00:13:48,400 specific customer, right where your profits are the long-term 252 00:13:48,400 --> 00:13:50,400 value that they're going to generate through loyalty? 253 00:13:51,000 --> 00:13:55,000 And the initial outlay the costs are how much you could spend to 254 00:13:55,000 --> 00:13:57,700 acquire them you've touched on this. 255 00:13:58,400 --> 00:14:02,100 But what is the impact? Act on the customer or the 256 00:14:02,100 --> 00:14:07,900 benefit that the customer receives to encourage them to 257 00:14:08,100 --> 00:14:12,800 come back and therefore provide that loyalty to the retailer 258 00:14:13,100 --> 00:14:15,700 from a Shoppers perspective. The benefit is really clear, 259 00:14:15,700 --> 00:14:19,200 which is that they're able to find like better products and 260 00:14:19,200 --> 00:14:21,600 they're able to do it in a much shorter time period. 261 00:14:21,700 --> 00:14:24,900 And it's as simple as like when I turn up to the site, do I even 262 00:14:24,900 --> 00:14:28,300 feel like this is a shock, that is kind of tailored towards my 263 00:14:28,300 --> 00:14:32,300 needs and to my Tastes so for Shoppers, they're able to find 264 00:14:32,300 --> 00:14:35,000 the things they want. But they also have a sense of 265 00:14:35,000 --> 00:14:38,200 feeling of being welcomed and a feeling that the place that 266 00:14:38,200 --> 00:14:40,800 they're shopping is the type of place that they want to continue 267 00:14:40,800 --> 00:14:43,600 doing business with, because it's it reflects their tastes 268 00:14:43,600 --> 00:14:47,600 their style, their values. I think the benefit of this type 269 00:14:47,600 --> 00:14:55,700 of Lifetime customer value. Analysis is obviously crucial 270 00:14:55,700 --> 00:14:59,600 for retailers, but it leads to the next question. 271 00:14:59,800 --> 00:15:03,900 In which is how do we implement this. 272 00:15:04,200 --> 00:15:06,600 You need to have a partner who can help guide you through the 273 00:15:06,600 --> 00:15:08,500 process, their companies like out go. 274 00:15:08,500 --> 00:15:13,500 Lea and other people who have solved these problems and who 275 00:15:13,500 --> 00:15:17,400 have really focused on making the integration and the data 276 00:15:17,400 --> 00:15:19,600 collection pieces that are needed to power, these AI 277 00:15:19,600 --> 00:15:22,500 algorithms, extremely simple with, a few lines of code, 278 00:15:22,500 --> 00:15:25,300 you're able to integrate with. Like I'll go Lea, unlock our 279 00:15:25,900 --> 00:15:29,300 search API working with vendors and working with folks who have 280 00:15:29,300 --> 00:15:33,100 dedicated Dated large amounts of resources, time and expertise to 281 00:15:33,100 --> 00:15:35,500 solving these problems tends to be the first starting point. 282 00:15:35,600 --> 00:15:37,500 And often the best solution, the long run. 283 00:15:37,900 --> 00:15:41,500 What are some of the best practices for implementing 284 00:15:41,500 --> 00:15:45,600 personalization? Across all of the touch points 285 00:15:45,600 --> 00:15:48,500 as customers interact with your site? 286 00:15:48,800 --> 00:15:52,400 The more touch points in the more data that you're able to 287 00:15:52,400 --> 00:15:56,800 capture about how your users are using a product, the better, all 288 00:15:56,800 --> 00:15:59,600 of these algorithms and reading need to be able to understand 289 00:16:00,300 --> 00:16:03,900 Views conversions, but also the product data customers are 290 00:16:03,900 --> 00:16:07,900 looking at and having something that can indicate which 291 00:16:07,900 --> 00:16:10,000 customers which between sessions. 292 00:16:10,100 --> 00:16:13,300 So if they can log into your website, that's the best, but if 293 00:16:13,300 --> 00:16:15,600 you can add cookies to your website so you can track them 294 00:16:15,600 --> 00:16:18,400 between visits. That's even that's good. 295 00:16:18,400 --> 00:16:21,100 As well. What about the buying funnel? 296 00:16:21,200 --> 00:16:25,400 We are customers are in their journey in their relationship to 297 00:16:25,400 --> 00:16:28,300 you as a retailer. Customers will turn up on the 298 00:16:28,308 --> 00:16:32,100 homepage for example and And you'll have very, very little 299 00:16:32,100 --> 00:16:35,700 intent about what they want in the shopping Journey that the 300 00:16:35,700 --> 00:16:38,100 kind of high level of the funnel where you're not sure what 301 00:16:38,100 --> 00:16:41,900 they're looking for. You want to showcase a breath to 302 00:16:41,900 --> 00:16:44,500 them of things that they can find on the store. 303 00:16:44,800 --> 00:16:47,000 And if there are returning customer, you want to have a 304 00:16:47,000 --> 00:16:50,900 strong personalization bias on this home page, so that they can 305 00:16:50,900 --> 00:16:54,400 see things that are at their taste or style or price points. 306 00:16:54,700 --> 00:16:58,000 But as they go and they searched and they start giving you more 307 00:16:58,000 --> 00:17:01,300 information about their intent. Really want to focus on the 308 00:17:01,300 --> 00:17:05,300 relevance, and on the accuracy of the results, you're showing 309 00:17:05,300 --> 00:17:09,000 them because they're giving you a very specific request. 310 00:17:09,099 --> 00:17:12,000 And then, once they land on a product detail page, for 311 00:17:12,000 --> 00:17:15,700 example, you have an even more specific piece of information 312 00:17:15,800 --> 00:17:17,500 they're interested in this product. 313 00:17:17,900 --> 00:17:20,900 So again, you want to become even more tightly focused on 314 00:17:20,908 --> 00:17:25,000 your recommendations, you maybe might not want to use as much 315 00:17:25,000 --> 00:17:27,800 personalization when you're on a product landing page. 316 00:17:28,400 --> 00:17:32,000 Because actually the The Shoppers told you specifically 317 00:17:32,000 --> 00:17:33,300 something that they're looking for. 318 00:17:33,600 --> 00:17:36,300 And so as a customer goes in through the Journey, you've got 319 00:17:36,300 --> 00:17:39,200 to think about how much intent you really have and either 320 00:17:39,200 --> 00:17:41,400 dial-up personalization. When you don't have as much 321 00:17:41,400 --> 00:17:45,000 intent or dial-up relevance accuracy. 322 00:17:45,800 --> 00:17:49,100 When you have high intent Sean what is the role of real-time 323 00:17:49,100 --> 00:17:52,600 data in all of this process. You got to think that a large 324 00:17:52,600 --> 00:17:54,200 number of customers coming to your website. 325 00:17:54,200 --> 00:17:56,200 It may be the first time that you've seen. 326 00:17:56,400 --> 00:17:59,200 It's their first experience. You want that first experience 327 00:17:59,200 --> 00:18:01,700 to be great. So as soon as they click on a 328 00:18:01,708 --> 00:18:04,800 single item or they type in a search request, you want that 329 00:18:04,800 --> 00:18:08,600 data to flow back into your systems and adapt, your 330 00:18:08,600 --> 00:18:11,100 algorithms and real-time you have to have systems that are 331 00:18:11,100 --> 00:18:15,200 extremely fast at both ingesting and getting the data in but also 332 00:18:15,200 --> 00:18:17,900 like retraining and adapting the algorithm. 333 00:18:18,000 --> 00:18:20,800 So I think real-time data is very important for that first 334 00:18:20,800 --> 00:18:24,900 experience but also what you sometimes find is customers come 335 00:18:24,900 --> 00:18:28,400 back and they shop slightly different way to their taste. 336 00:18:28,500 --> 00:18:30,800 So you also want to be able to adapt In real time, even for 337 00:18:30,800 --> 00:18:34,700 customers, that you know, a lot about, then you use this data to 338 00:18:34,700 --> 00:18:41,000 make the decision around what to show next to that user. 339 00:18:41,000 --> 00:18:44,300 And of course, it's happening. Essentially instantaneously from 340 00:18:44,300 --> 00:18:46,900 the user perspective. Instantaneous, we're talking 341 00:18:46,900 --> 00:18:49,400 about milliseconds here that matter in the shopping 342 00:18:49,400 --> 00:18:51,400 experience. The faster, the experience is, 343 00:18:52,200 --> 00:18:54,300 the more likely customers are going to continue. 344 00:18:55,300 --> 00:18:58,500 We've seen 100 200 milliseconds worth of delay. 345 00:18:58,500 --> 00:19:01,300 Actually reduces the conversion. And customers end up, leaving 346 00:19:01,300 --> 00:19:04,000 the website. So speed is really really 347 00:19:04,000 --> 00:19:06,500 important in the Commerce world are there. 348 00:19:06,500 --> 00:19:11,900 Common implementation challenges that retailers face when they're 349 00:19:11,900 --> 00:19:15,600 trying to implement this kind of personalization. 350 00:19:16,000 --> 00:19:19,300 One of the hard things is making sure that you are capturing this 351 00:19:19,300 --> 00:19:23,600 type of 121 data around. This is the user that's logged 352 00:19:23,600 --> 00:19:26,600 in. This is the cookie ID that they 353 00:19:26,600 --> 00:19:29,900 accepted, this is the session ID for the shopping session. 354 00:19:30,000 --> 00:19:33,300 Ian and making that information available across every click 355 00:19:33,400 --> 00:19:36,800 every product view in every single time they come to the 356 00:19:36,800 --> 00:19:40,300 site so that the algorithms can do their work, the algorithms 357 00:19:40,300 --> 00:19:44,000 can only do so much, they need to be able to identify users. 358 00:19:44,000 --> 00:19:48,400 So that's that's really one of the bigger areas where Ecommerce 359 00:19:48,400 --> 00:19:50,600 companies need to make sure they've implemented Google 360 00:19:50,600 --> 00:19:53,700 analytics pretty well or whatever their analytics 361 00:19:53,700 --> 00:19:57,100 packages. What's the solution to this very 362 00:19:57,100 --> 00:19:59,500 common problem using off-the-shelf analytics 363 00:19:59,500 --> 00:20:01,900 products? But really focusing on doing a 364 00:20:01,900 --> 00:20:03,700 comprehensive job and implementing them. 365 00:20:03,800 --> 00:20:07,300 Also, making sure that when you choose a vendor that you really 366 00:20:07,300 --> 00:20:10,900 take the extra time when you do the integration to send all the 367 00:20:10,900 --> 00:20:13,500 events, send all the clicks and conversions with the right 368 00:20:13,500 --> 00:20:17,100 fields and everything. How do you address the ethical 369 00:20:17,100 --> 00:20:21,600 or the privacy considerations that people think of, when we 370 00:20:21,600 --> 00:20:25,400 talk about this kind of hyper personalization, one of the 371 00:20:25,400 --> 00:20:29,400 things that we have to understand first is that a lot 372 00:20:29,400 --> 00:20:32,600 of customers Workers and a lot of Shoppers are very happy to 373 00:20:32,600 --> 00:20:36,100 make the trade-off of getting a better experience by allowing 374 00:20:36,100 --> 00:20:40,100 the merchant or the site that they're shopping on to be able 375 00:20:40,100 --> 00:20:43,700 to collect this type of data. And when you think about it, 376 00:20:43,700 --> 00:20:46,500 that way, you want to make a transparent to them what you're 377 00:20:46,500 --> 00:20:48,600 doing. And you want to allow them to 378 00:20:48,600 --> 00:20:51,200 make that trade-off for customers who don't feel 379 00:20:51,200 --> 00:20:53,600 comfortable with that. You want to enable them to have 380 00:20:53,600 --> 00:20:56,700 a very clear trade-off as well and provide a great default 381 00:20:56,700 --> 00:20:58,500 experience. I think it's just important 382 00:20:58,500 --> 00:21:02,100 about transparency But also telling customers with the value 383 00:21:02,100 --> 00:21:05,600 of it is so they know ahead of time when they click the kind of 384 00:21:05,600 --> 00:21:08,400 accept cookies button or decide to log in and create an account 385 00:21:08,400 --> 00:21:12,000 for example that by doing. So they're really going to get a 386 00:21:12,700 --> 00:21:15,600 far better experience throughout the journey Shawn. 387 00:21:15,600 --> 00:21:22,000 Let's talk about measurement and kpis what are the best 388 00:21:22,000 --> 00:21:26,900 mechanisms or approaches to measuring the results of these 389 00:21:26,900 --> 00:21:29,200 kinds of personalization efforts? 390 00:21:29,300 --> 00:21:32,200 We use a A/B Testing where we will make a change to the 391 00:21:32,200 --> 00:21:36,300 website, like, we will update our algorithm and we will send 392 00:21:36,300 --> 00:21:38,500 some of the customers down the, a channel, which is the old 393 00:21:38,500 --> 00:21:40,900 experience. And some customers down, the B 394 00:21:40,900 --> 00:21:42,400 Channel, which is the new experience. 395 00:21:42,900 --> 00:21:46,000 And using statistics, we can find out whether or not the 396 00:21:46,300 --> 00:21:49,200 customers who are getting the new product or new experience 397 00:21:49,700 --> 00:21:53,500 convert at a higher rate, or spend more money per purchase or 398 00:21:53,500 --> 00:21:58,200 have a higher revenue and total and we can figure out the 399 00:21:58,300 --> 00:22:01,700 statistical significance of So that when we actually decide to 400 00:22:01,700 --> 00:22:05,100 roll out the change, we're confident that it's actually 401 00:22:05,100 --> 00:22:07,500 going to improve the experience for customers. 402 00:22:08,200 --> 00:22:10,400 So it's exciting. It's a bit like scientific 403 00:22:10,400 --> 00:22:14,400 experimentation in this experimentation, how much is 404 00:22:14,400 --> 00:22:18,900 managed by the retail shop owner versus? 405 00:22:18,900 --> 00:22:23,800 How much is managed by algo Lea behind the scenes, we have A/B 406 00:22:23,800 --> 00:22:26,500 Testing built into everything that we do on the platform 407 00:22:26,900 --> 00:22:29,600 because we think it's just like one of the highest velocity ways 408 00:22:29,600 --> 00:22:31,600 of improving. Proving your experience. 409 00:22:32,000 --> 00:22:36,200 So it's definitely built into Al kolya, we can run manage all the 410 00:22:36,200 --> 00:22:38,500 A/B experiments for you, show you the data. 411 00:22:38,800 --> 00:22:43,000 So you feel confidence in making changes Are there specific 412 00:22:43,000 --> 00:22:47,500 metrics that reflect customer Revenue but also customer 413 00:22:47,500 --> 00:22:50,200 loyalty and customer lifetime value. 414 00:22:50,600 --> 00:22:52,500 There are a lot of really important metrics that you need 415 00:22:52,500 --> 00:22:56,500 to keep track of when you're running an online business and 416 00:22:56,600 --> 00:22:59,200 it really depends on kind of where in the buying funnel, I 417 00:22:59,208 --> 00:23:02,200 think you are. So, for example, when a customer 418 00:23:02,200 --> 00:23:04,800 first arrives in the site, it's important to look at the bounce 419 00:23:04,800 --> 00:23:06,600 rate. How many clicks do they get into 420 00:23:06,600 --> 00:23:08,500 the product before they? It will experience it. 421 00:23:09,000 --> 00:23:11,200 Whereas, if someone's made it all the way to a landing page, 422 00:23:11,200 --> 00:23:13,600 for example, you probably want to be more focused on the 423 00:23:13,608 --> 00:23:16,000 conversion rate and whether they're actually going to check 424 00:23:16,000 --> 00:23:19,700 out and buy something and then after they've actually bought 425 00:23:19,700 --> 00:23:22,000 something, you really want to be looking at that lifetime value 426 00:23:22,000 --> 00:23:23,600 number. How often do they come back to 427 00:23:23,600 --> 00:23:25,900 the website? Visit again, one of the things 428 00:23:25,900 --> 00:23:27,900 that I think is extremely important, particularly for 429 00:23:27,900 --> 00:23:32,600 search is the Mission it, which someone has clicked on a result 430 00:23:32,600 --> 00:23:34,200 when given a set of search results. 431 00:23:34,300 --> 00:23:39,300 If you have, let's say, 10 rows worth of search results, the 432 00:23:39,300 --> 00:23:42,300 results that appear on the very first row will always get 433 00:23:42,300 --> 00:23:44,800 clicked on a far higher rate than the ones that appear on the 434 00:23:44,800 --> 00:23:48,000 tenth row. So I really think that making 435 00:23:48,000 --> 00:23:50,300 sure that the products that you're getting into that first 436 00:23:50,300 --> 00:23:53,400 row. We're second row are extremely 437 00:23:53,400 --> 00:23:55,100 relevant. We see it makes a big 438 00:23:55,100 --> 00:23:59,100 difference. All of these very granular. 439 00:24:00,000 --> 00:24:05,200 Sets of reference data, become the building blocks, or the 440 00:24:05,200 --> 00:24:11,300 components of understanding the consumer, and then you can 441 00:24:11,300 --> 00:24:14,900 respond, you need to understand the products really well that 442 00:24:14,900 --> 00:24:17,600 you're selling. And the kind of options you have 443 00:24:17,600 --> 00:24:20,800 to present. Secondly, you have to understand 444 00:24:20,800 --> 00:24:24,600 the Shopper who turns up through personalization through profile 445 00:24:24,600 --> 00:24:27,500 building through, looking at all, their past interactions, 446 00:24:28,100 --> 00:24:31,400 and then thirdly, you need to understand and Real-time the 447 00:24:31,400 --> 00:24:33,800 things that they're telling you that they're looking for the 448 00:24:33,800 --> 00:24:36,400 intent. How do you manage to different 449 00:24:36,400 --> 00:24:40,200 kinds of Shoppers you have folks who show up and they leisurely 450 00:24:40,200 --> 00:24:44,800 want to browse your catalog and then you have other people who 451 00:24:44,800 --> 00:24:48,400 just, you know, I don't have any time and I just need to buy this 452 00:24:48,400 --> 00:24:52,200 product and if you don't have this product right now, I'm 453 00:24:52,200 --> 00:24:54,700 going away. And if you do have it, I'm going 454 00:24:54,700 --> 00:24:56,400 to buy. There are a couple of shopping 455 00:24:56,400 --> 00:24:59,000 modes. The first one which obviously 456 00:24:59,000 --> 00:25:02,600 you need to be. Incredibly good at capturing a 457 00:25:02,600 --> 00:25:07,000 high intent, Shopper is come to buy specific product and you 458 00:25:07,008 --> 00:25:09,900 need to do that with Incredible search results and really great 459 00:25:09,900 --> 00:25:12,400 ranking of those results. Then you have a second set of 460 00:25:12,408 --> 00:25:16,900 Shoppers who really enjoy the experience of coming to a site 461 00:25:16,900 --> 00:25:20,600 and being inspired and discovering this site has to 462 00:25:20,600 --> 00:25:22,700 offer. And these are Shoppers where you 463 00:25:22,700 --> 00:25:28,200 really want to create a browse experience which is far more 464 00:25:28,200 --> 00:25:31,400 interactive and far more. Or inspirational. 465 00:25:32,100 --> 00:25:35,400 So I'll tell you at the moment. Most sites have a browse 466 00:25:35,400 --> 00:25:38,800 experience which is something along the lines of, okay, a 467 00:25:38,800 --> 00:25:41,700 customer clicks on. The dress category, we have 468 00:25:41,700 --> 00:25:46,000 30,000 dresses, you know, give them 100 Pages worth of dresses 469 00:25:46,000 --> 00:25:50,600 and order that by the most popular and it's just like 470 00:25:50,600 --> 00:25:53,700 entirely overwhelming. And often the most popular 471 00:25:53,700 --> 00:25:58,100 products are some of the least inspirational Shawn, where is 472 00:25:58,100 --> 00:26:01,500 search going over the next few? Here's one of the big things is 473 00:26:01,500 --> 00:26:04,000 going to happen to search. We talked about how vectors and 474 00:26:04,000 --> 00:26:07,100 large language models are really transforming the way that we 475 00:26:07,100 --> 00:26:09,200 searched. We're not just matching keywords 476 00:26:09,200 --> 00:26:11,100 anymore. We're actually understanding 477 00:26:11,100 --> 00:26:14,300 customers understanding the concepts and coming up with much 478 00:26:14,300 --> 00:26:19,600 better matching algorithms, but I think one other major change 479 00:26:19,600 --> 00:26:24,300 is going to be the way that we have more conversational 480 00:26:24,300 --> 00:26:27,800 approach to shopping. So when you think about the 481 00:26:27,800 --> 00:26:31,500 offline shopping experience off, When people go into a physical 482 00:26:31,500 --> 00:26:35,800 retail store because they want to get some assistance and they 483 00:26:35,800 --> 00:26:39,500 want to talk to someone who has some expert knowledge that used 484 00:26:39,500 --> 00:26:42,800 to be an experience. That wasn't very good online, 485 00:26:42,900 --> 00:26:44,700 it's very static. It's like here just read the 486 00:26:44,700 --> 00:26:46,500 information. You figure it out for yourself 487 00:26:46,700 --> 00:26:50,200 but with the power of chap GPT and these large language models, 488 00:26:50,600 --> 00:26:53,700 we're going to be able to have much more expert, personal 489 00:26:53,700 --> 00:26:56,100 assistant like conversations at scale. 490 00:26:56,400 --> 00:26:59,700 Sean what should eCommerce retailers? 491 00:26:59,900 --> 00:27:03,600 Do now to take advantage of all of these capabilities you've 492 00:27:03,600 --> 00:27:05,700 been describing. So I think there are three main 493 00:27:05,700 --> 00:27:08,700 things that eCommerce operator should be thinking about the 494 00:27:08,700 --> 00:27:11,900 first is the vendors, they work with, they should be working 495 00:27:11,900 --> 00:27:15,200 with their existing vendor today to take advantage of some of the 496 00:27:15,200 --> 00:27:18,900 AI capabilities that are coming or to you know select or add 497 00:27:18,900 --> 00:27:21,000 vendors. Who have these AI capabilities. 498 00:27:22,100 --> 00:27:24,700 The second thing is that I think they really need to focus on 499 00:27:24,700 --> 00:27:28,800 their own analytics. So really capturing all the data 500 00:27:28,800 --> 00:27:31,700 correctly about their Customers whether that's clicks and 501 00:27:31,700 --> 00:27:34,400 conversion data. But also like making sure the 502 00:27:34,400 --> 00:27:37,200 capture the cookie IDs and who's logged in ETC. 503 00:27:37,200 --> 00:27:40,100 So that you can personalize the experience. 504 00:27:40,500 --> 00:27:43,300 And then thirdly, I think they should be thinking about the 505 00:27:43,300 --> 00:27:45,800 metrics that matter to them as a business. 506 00:27:45,900 --> 00:27:49,200 So they can understand how the AI algorithms are actually 507 00:27:49,200 --> 00:27:52,600 creating a more sustainable profitable and long-term loyal. 508 00:27:52,600 --> 00:27:55,300 Customer base. We've been talking with Sean 509 00:27:55,300 --> 00:28:00,200 Milani, Chief technology officer of algo, Leah, For more 510 00:28:00,200 --> 00:28:04,400 information, check out www.google.com.