1 00:00:00,320 --> 00:00:01,360 All right. Well, hello and thank you all 2 00:00:01,360 --> 00:00:04,280 for tuning into another episode of the Professional Pricing 3 00:00:04,280 --> 00:00:06,920 Society podcast. My name is Terence, and today 4 00:00:06,920 --> 00:00:10,960 we're going to be discussing B2B pricing in the era of AI and 5 00:00:10,960 --> 00:00:13,720 data science. Spearheading this conversation 6 00:00:13,720 --> 00:00:17,320 is a director at Gap named Vivek Anand. 7 00:00:17,720 --> 00:00:21,360 Vivek is an operations research professional with over a decade 8 00:00:21,360 --> 00:00:24,400 of expertise in leveraging machine learning and AI and 9 00:00:24,400 --> 00:00:26,200 pricing analytics and optimization. 10 00:00:26,920 --> 00:00:30,320 In his previous role as the Director of Science at a leading 11 00:00:30,320 --> 00:00:34,080 B2B pricing vendor, Vivet cultivated an impressive track 12 00:00:34,080 --> 00:00:37,240 record where he successfully developed analytically driven 13 00:00:37,240 --> 00:00:40,680 pricing solutions that led to revenue and profit growth for 14 00:00:40,680 --> 00:00:44,120 numerous Fortune 500 companies across various traditional B2B 15 00:00:44,120 --> 00:00:46,960 industry verticals. Additionally, his expertise 16 00:00:46,960 --> 00:00:50,880 extended to the niche industry segments such as oil and gas, 17 00:00:51,200 --> 00:00:55,280 legal advisories and subscriptions and HR solutions, 18 00:00:55,280 --> 00:00:57,920 where he innovative price optimization frameworks for 19 00:00:57,920 --> 00:01:00,920 these industries. Vivek, how are you doing today? 20 00:01:01,880 --> 00:01:04,599 I'm pretty good. It's great to be on this podcast 21 00:01:04,599 --> 00:01:07,120 with you. It's really a leading 22 00:01:07,120 --> 00:01:10,280 marketplace for pricing ideas. Yep, great to be with you. 23 00:01:10,680 --> 00:01:12,200 Good, good. We're glad to have you. 24 00:01:12,560 --> 00:01:15,120 Let's go ahead and just dive right into this conversation. 25 00:01:15,800 --> 00:01:20,080 B2B pricing in the era of AI and data science. 26 00:01:20,080 --> 00:01:23,240 So can you share with our audience a bit about your 27 00:01:23,240 --> 00:01:26,360 professional journey and delve into the specific ways your 28 00:01:26,360 --> 00:01:30,120 expertise and B2B pricing has kind of evolved over the years, 29 00:01:30,120 --> 00:01:32,640 just to kind of give everyone a better grasp of your background? 30 00:01:33,240 --> 00:01:35,840 Yeah, absolutely. So currently I'm Director of 31 00:01:35,840 --> 00:01:40,520 Advanced Analytics at a Fortune retailer Gap where my team's 32 00:01:40,520 --> 00:01:42,840 managed really is to optimize business outcomes. 33 00:01:42,880 --> 00:01:46,400 It could be profitability, productivity, actually it's both 34 00:01:47,240 --> 00:01:49,360 for use of analytics for the enterprise. 35 00:01:49,360 --> 00:01:53,280 So basically my team primarily deals with application of data 36 00:01:53,280 --> 00:01:57,720 science analytics in pricing analytics, inventory analytics 37 00:01:57,720 --> 00:02:01,400 and buying analytics. And like I said, prior to my 38 00:02:01,400 --> 00:02:04,640 current role, I was Director of Science at a leading pricing 39 00:02:04,640 --> 00:02:08,000 solution provider. And in that role I worked with a 40 00:02:08,000 --> 00:02:12,160 number of B to B clients across various verticals where I help 41 00:02:12,320 --> 00:02:16,480 kind of build or improve upon their existing pricing programs 42 00:02:17,240 --> 00:02:20,320 by developing machine learning driven practice optimization 43 00:02:20,320 --> 00:02:23,200 solution. I kind of also worked with like 44 00:02:23,200 --> 00:02:26,720 you said, niche industries like oil and gas and subscription 45 00:02:26,720 --> 00:02:30,120 legal solutions where like you did not find many academic or 46 00:02:30,120 --> 00:02:33,640 industrial scientific literature on how to use data science 47 00:02:33,640 --> 00:02:38,720 analytics to optimize prices. And to your second question 48 00:02:38,720 --> 00:02:42,040 about like how it has evolved, When I started in this industry 49 00:02:42,280 --> 00:02:45,840 like 5-7 ten years ago, it was like slightly different. 50 00:02:45,840 --> 00:02:48,640 The data science uses were like not prevalent. 51 00:02:49,640 --> 00:02:52,440 A typical price optimization solution like still constitute 52 00:02:52,440 --> 00:02:55,040 segmentation, computing, elasticity and optimization. 53 00:02:55,040 --> 00:02:57,760 But all of those 3 components have evolved over time, right? 54 00:02:57,760 --> 00:03:00,720 So previously when I started segmentation, attributes were 55 00:03:00,720 --> 00:03:04,960 kind of divided decided by the business stakeholders like they 56 00:03:04,960 --> 00:03:07,360 would feel like, hey these attributes are important. 57 00:03:07,360 --> 00:03:10,240 They will have like a short list of attributes and you use 58 00:03:10,240 --> 00:03:14,240 primarily, primarily use like basic or simple statistical 59 00:03:14,240 --> 00:03:16,680 techniques to identify like which of those attributes made 60 00:03:16,680 --> 00:03:19,440 sense and which one of those did not. 61 00:03:20,320 --> 00:03:25,320 And similarly when you go to elasticity again it used to be 62 00:03:25,320 --> 00:03:28,240 like a regression and still is a regression can be log log 63 00:03:28,240 --> 00:03:32,600 regression, linear regression. But those methods are not always 64 00:03:32,600 --> 00:03:34,920 appropriate all the time. So for example there have been 65 00:03:34,920 --> 00:03:38,120 cases where you do not have enough data and in that case 66 00:03:38,120 --> 00:03:40,800 you're trying to create a regression. 67 00:03:40,800 --> 00:03:43,320 Then like a few months in, you have new sets of data come in 68 00:03:43,320 --> 00:03:46,880 and all of a sudden your results are different because like 1 69 00:03:46,880 --> 00:03:49,880 outlier can change it, right. So it's still a robust method, 70 00:03:49,880 --> 00:03:53,400 but the way it was used traditionally was not 71 00:03:53,400 --> 00:03:55,160 appropriate, it's not optimal if you will. 72 00:03:55,160 --> 00:03:58,520 And similarly like price optimization, like there is this 73 00:03:58,520 --> 00:04:01,680 challenge industry, but there are many, many places where this 74 00:04:01,680 --> 00:04:04,800 price optimization is basically a rules based price setting, 75 00:04:04,800 --> 00:04:06,240 right? Like OK, when this happens to 76 00:04:06,240 --> 00:04:11,280 this kind of thing, right. So when I started, what I did 77 00:04:11,280 --> 00:04:14,080 was I started experimenting with data science algorithm. 78 00:04:14,120 --> 00:04:20,600 So basically to get these things done faster and better, so I 79 00:04:20,600 --> 00:04:24,000 developed a new machine learning framework that would quickly 80 00:04:24,000 --> 00:04:25,920 assess a battery of segmentation attributes. 81 00:04:25,920 --> 00:04:28,720 And then we are not just limited by what the business stakeholder 82 00:04:28,720 --> 00:04:31,560 thought, they were important, so we could run data analysis EDA 83 00:04:31,680 --> 00:04:34,320 quickly, faster to understand like OK, these are the 84 00:04:34,320 --> 00:04:38,960 dimensions on which you can slice the data and then also 85 00:04:38,960 --> 00:04:42,480 assess their attributes. Sometimes like the business, the 86 00:04:42,480 --> 00:04:44,840 hypothesis that we get from the business is not visible in the 87 00:04:44,840 --> 00:04:46,880 data and we are coming up with better attributes. 88 00:04:47,120 --> 00:04:49,320 So this machine learning framework can assess a battery 89 00:04:49,320 --> 00:04:51,400 of attributes in a relatively short amount of time. 90 00:04:52,560 --> 00:04:56,200 Similarly for elasticities, I brought in like newer and faster 91 00:04:56,200 --> 00:04:58,640 techniques. I need to talk more about these 92 00:04:58,640 --> 00:05:03,120 like later, like faster sampling algorithms that could converge 93 00:05:03,120 --> 00:05:05,240 and reduce the run times. And scale is a big problem, 94 00:05:05,240 --> 00:05:07,400 right? Like 5-10 years ago, you are 95 00:05:07,400 --> 00:05:10,240 primarily dealing with a small amount of data on transactions, 96 00:05:10,240 --> 00:05:12,000 customers, products. That's about it, right? 97 00:05:12,440 --> 00:05:15,840 And now we have a lot of data, right. 98 00:05:15,960 --> 00:05:19,960 So how do you go from having a method that was suited for a 99 00:05:19,960 --> 00:05:21,840 small amount of data to a large amount of data? 100 00:05:22,280 --> 00:05:28,040 So this is like net net the cultural shift where the 101 00:05:28,040 --> 00:05:31,200 clients, the solution providers and the end users are kind of 102 00:05:31,360 --> 00:05:36,360 not shying away from using data science and analytics and 103 00:05:36,360 --> 00:05:39,200 leaning on machine learning, data science to make faster and 104 00:05:39,200 --> 00:05:42,200 smarter decisions. That's like overall tectonic 105 00:05:42,200 --> 00:05:43,760 change that has happened in the landscape. 106 00:05:44,120 --> 00:05:45,440 OK, awesome. That's cool. 107 00:05:45,800 --> 00:05:48,120 Now, you do have a very extensive background in this, 108 00:05:48,120 --> 00:05:51,600 even though things have changed in the past five to 10 years and 109 00:05:51,600 --> 00:05:54,360 you've been at the forefront of applying innovative data science 110 00:05:54,360 --> 00:05:59,040 and AI techniques in pricing. So, starting with segmentation, 111 00:05:59,040 --> 00:06:02,480 do you mind walking us through some fascinating instances where 112 00:06:02,760 --> 00:06:06,240 you have kind of harnessed AI and data science to create a 113 00:06:06,240 --> 00:06:08,760 more robust segmentation solution? 114 00:06:09,840 --> 00:06:11,760 Yeah, yeah, absolutely. And this is, this is such a 115 00:06:11,760 --> 00:06:14,640 great topic. Actually there are many such 116 00:06:14,640 --> 00:06:17,360 examples of it. And as we speak, like the one 117 00:06:17,360 --> 00:06:22,720 that comes to my mind is a customer where like they have 118 00:06:22,720 --> 00:06:24,200 like a traditional segmentation model. 119 00:06:25,240 --> 00:06:28,760 The customer had implemented a customer segmentation model for 120 00:06:28,760 --> 00:06:31,800 their own customers across 2 dimensions, like how much their 121 00:06:31,800 --> 00:06:35,360 customers spent money over a year and what is the price they 122 00:06:35,360 --> 00:06:38,680 were like paying, right? Like a price index, which 123 00:06:38,680 --> 00:06:41,480 effectively kind of measures like, hey, this person paid 124 00:06:41,680 --> 00:06:45,080 below the average price paid by all customers or above, right. 125 00:06:46,480 --> 00:06:49,240 And if you think about these two attributes, think of it as Skype 126 00:06:49,320 --> 00:06:52,120 plot and these two dimensions will create like 4 quadrants, 127 00:06:52,120 --> 00:06:55,880 right? You break the customer base by 128 00:06:55,880 --> 00:06:58,680 spent and then you further break it by the price index, right? 129 00:06:58,680 --> 00:07:03,160 So you have 4 bad clusters or or or segments, if you will. 130 00:07:04,000 --> 00:07:06,920 Now the salesperson used to come to them and say, like, hey, I 131 00:07:06,920 --> 00:07:10,360 don't really like this pricing because I feel like these two 132 00:07:10,360 --> 00:07:12,280 customers are selling situations are similar. 133 00:07:12,520 --> 00:07:15,920 I feel the price guidance for one is very low and we are 134 00:07:15,920 --> 00:07:18,720 leaving money on the table, whereas others is very high. 135 00:07:19,280 --> 00:07:22,720 And we are kind of like at the risk of not winning the 136 00:07:22,720 --> 00:07:25,280 business, right. And this was a problem that like 137 00:07:25,280 --> 00:07:28,800 kept on repeating and and that sort of also erodes the trust 138 00:07:29,400 --> 00:07:31,560 from the end user, the person who is using the pricing 139 00:07:31,560 --> 00:07:32,920 solution. So the customer like the 140 00:07:32,920 --> 00:07:36,000 business sponsor came to us and asked like, hey, like this is 141 00:07:36,000 --> 00:07:37,720 what we are hearing. And I kind of agree with what 142 00:07:37,720 --> 00:07:40,320 they're saying is in their argument like, can you like make 143 00:07:40,320 --> 00:07:46,040 sense of why it is happening. So again, like looking at the 144 00:07:46,040 --> 00:07:48,440 transaction data, it's very difficult to kind of like have a 145 00:07:48,440 --> 00:07:50,680 scarab water and eyeball and say like, does it make sense? 146 00:07:51,840 --> 00:07:54,040 So that was like one of the first use case of data science, 147 00:07:54,040 --> 00:07:57,760 like years ago where I just like ran a unsupervised crossing 148 00:07:57,760 --> 00:08:00,560 algorithm. So basically instead of telling 149 00:08:00,560 --> 00:08:04,400 the system like how many clusters of segments of data you 150 00:08:04,400 --> 00:08:07,000 need to create, go figure how many should be there. 151 00:08:08,000 --> 00:08:10,880 And the machine came back and said like, OK, there are optimal 152 00:08:11,040 --> 00:08:13,840 5 clusters instead of the four that the traditional 153 00:08:13,840 --> 00:08:15,320 segmentation tree was creating, right? 154 00:08:16,160 --> 00:08:19,000 And then we did some further analysis and we found that there 155 00:08:19,000 --> 00:08:22,680 was a cluster of customers or transactions that were not the 156 00:08:22,680 --> 00:08:24,720 top spenders or not the bottom spenders. 157 00:08:24,720 --> 00:08:27,280 They were right in the middle. And then there was another set 158 00:08:27,280 --> 00:08:29,560 of customers that were like not paying the top dollar but also 159 00:08:29,560 --> 00:08:31,720 not paying the rock bottom. So they were like right in the 160 00:08:31,720 --> 00:08:34,000 middle. And you partition that middle 161 00:08:34,000 --> 00:08:37,200 cluster into four segments, the way the traditional segmentation 162 00:08:37,200 --> 00:08:40,520 was done, 2 customers with a very similar selling situation, 163 00:08:40,520 --> 00:08:43,039 one goes into a high price cluster, other goes in a low 164 00:08:43,039 --> 00:08:45,080 price cluster, right. And that was the source of the 165 00:08:45,080 --> 00:08:48,440 problem. So we did some further analysis 166 00:08:48,440 --> 00:08:53,400 and work with the customer to kind of give them like hey, 167 00:08:53,440 --> 00:08:56,760 instead of like doing this customer segmentation across 168 00:08:56,760 --> 00:08:59,600 these two attributes like in the segmentation tree maybe we have 169 00:08:59,600 --> 00:09:02,040 one attribute that like clusters customers and you have those 170 00:09:02,040 --> 00:09:04,320 five clusters more a data science driven cluster 171 00:09:04,320 --> 00:09:07,760 segmentation and that like low behold the problem was kind of 172 00:09:07,760 --> 00:09:10,720 solved, right. So that was one good example. 173 00:09:12,800 --> 00:09:15,680 Another good example actually goes back to customer specific 174 00:09:15,680 --> 00:09:19,640 price, B to B, like as you know, it's very negotiated bilateral 175 00:09:19,640 --> 00:09:24,880 conversations, right. So segmentation will give you a 176 00:09:24,880 --> 00:09:27,560 segment price, right? But again, there are situations 177 00:09:27,560 --> 00:09:29,840 within the segment that there are customers who pay the rock 178 00:09:29,840 --> 00:09:32,480 bottom or below the floor price. Like in typical price guidance, 179 00:09:32,480 --> 00:09:34,600 you have a stock price which is like the best price that you 180 00:09:34,600 --> 00:09:37,040 expect to get. There's a floor price before 181 00:09:37,160 --> 00:09:39,600 below which you don't want to go and then there's a target price 182 00:09:39,600 --> 00:09:42,880 right in the middle, right. And there will always be 183 00:09:42,880 --> 00:09:44,240 transactions that are below floor. 184 00:09:44,240 --> 00:09:45,800 There will always be transactions that are above 185 00:09:45,800 --> 00:09:48,440 stock, right. And we were dealing with a 186 00:09:48,440 --> 00:09:54,240 customer who was like typically into selling raw materials and 187 00:09:54,240 --> 00:09:59,280 stabilizing products for food and beverage industry and they 188 00:09:59,280 --> 00:10:02,520 were like, hey, I want to build customer specific pricing for my 189 00:10:02,520 --> 00:10:05,320 customers because I feel like hey, we are losing business or 190 00:10:05,320 --> 00:10:07,160 we are not increasing prices fast enough. 191 00:10:07,960 --> 00:10:11,320 And a typical implementation of customer specific prices is like 192 00:10:11,320 --> 00:10:13,760 people look at like, OK, what is this customer within this 193 00:10:13,760 --> 00:10:16,920 segment has paid and where are they related to the segment 194 00:10:16,920 --> 00:10:18,240 prices? They are like way below the 195 00:10:18,320 --> 00:10:20,440 floor price, maybe slowly increase them to floor if 196 00:10:20,440 --> 00:10:22,360 they're between floor and target, increase them close to 197 00:10:22,360 --> 00:10:24,400 target, right. So it's again a rules based 198 00:10:24,400 --> 00:10:26,960 pricing, not a super scientific way of doing it, right. 199 00:10:27,880 --> 00:10:30,360 And the customer was like hey, can you come up with a more 200 00:10:30,360 --> 00:10:33,080 scientific way. So like what we implemented was 201 00:10:33,080 --> 00:10:36,480 like a very hyper focused segmentation based on not the 202 00:10:36,480 --> 00:10:39,240 segmentation attributes that we use to segment the business, but 203 00:10:39,360 --> 00:10:42,880 instead of like within the segment itself, what is excuse 204 00:10:42,880 --> 00:10:44,840 me a hyper segmentation that we can implement. 205 00:10:46,320 --> 00:10:49,000 And I used like machine learning methods for that hyper 206 00:10:49,000 --> 00:10:50,360 segmentation that I'm talking about. 207 00:10:51,200 --> 00:10:54,280 And we were able to leverage like their purchase behaviour, 208 00:10:54,280 --> 00:10:57,520 who are their end customer to assess like how much is their 209 00:10:57,520 --> 00:11:00,760 willingness to pay. And in this example we learned 210 00:11:00,760 --> 00:11:05,560 that all things equal you have 3 customer, 1 customers. 211 00:11:06,080 --> 00:11:09,080 End user is like adult food industry like you and I, right? 212 00:11:09,880 --> 00:11:13,720 Eating like packaged food and then similar customer is making 213 00:11:13,720 --> 00:11:16,000 baby food and similar customer is making pet food. 214 00:11:16,640 --> 00:11:19,880 The willingness to pay for baby food manufacturer and the pet 215 00:11:19,880 --> 00:11:22,400 food manufacturer was significantly higher compared to 216 00:11:22,440 --> 00:11:25,720 adult food manufacturer. Oh wow, that's interesting. 217 00:11:25,720 --> 00:11:28,160 And and in that case what we were able to do is like OK, 218 00:11:28,200 --> 00:11:30,960 instead of going by like where they are in the pricing segment, 219 00:11:30,960 --> 00:11:32,720 let's go by their willingness to pay. 220 00:11:33,080 --> 00:11:36,680 So we were able to give higher price increases to the baby and 221 00:11:36,680 --> 00:11:40,440 pet food manufacturers compared to like adult food 222 00:11:40,440 --> 00:11:42,480 manufacturers. So like these are like hidden 223 00:11:42,480 --> 00:11:45,960 patterns that are like not visible or even like the 224 00:11:45,960 --> 00:11:49,760 traditional segmentation methods will kind of like fail at 225 00:11:49,760 --> 00:11:52,400 identifying these trends and patterns in the data which is 226 00:11:52,400 --> 00:11:53,840 like machines are very good as this. 227 00:11:53,960 --> 00:11:56,920 Sure, that's really cool. That's a good point. 228 00:11:56,920 --> 00:12:01,640 You make leaning on machine learning for these algorithms 229 00:12:01,640 --> 00:12:06,120 and for these hyper focus segmentation components to 230 00:12:06,120 --> 00:12:10,200 further study the willingness of customers and how their their 231 00:12:10,200 --> 00:12:11,960 willingness to I guess purchase certain products. 232 00:12:12,560 --> 00:12:15,480 You know there's there's a lot that goes into segmentation. 233 00:12:17,360 --> 00:12:19,160 But can you highlight a few other instances where you 234 00:12:19,160 --> 00:12:23,000 effectively utilize, you know, AI or data science to tackle 235 00:12:23,000 --> 00:12:25,840 complex pricing challenges? Because it kind of would be 236 00:12:25,840 --> 00:12:29,080 great to hear about the diversity of problems you you 237 00:12:29,080 --> 00:12:31,240 can address in this particular space. 238 00:12:31,920 --> 00:12:35,360 Yeah, absolutely. So I mean one thing I alluded to 239 00:12:35,360 --> 00:12:40,040 like customers, end users willingness to lean on data. 240 00:12:40,560 --> 00:12:43,640 So people are harvesting a lot of data and like previously like 241 00:12:43,640 --> 00:12:46,720 I said, you would have some transaction data, you would have 242 00:12:46,720 --> 00:12:48,640 some customer data and then things like that. 243 00:12:48,640 --> 00:12:53,240 And then you build a model Right now with more and more data 244 00:12:53,240 --> 00:12:55,560 coming in and more and more products being available to the 245 00:12:55,560 --> 00:12:59,400 customers, the scope of data has been really exploded, right. 246 00:13:00,160 --> 00:13:04,480 And if you continue using the traditional methods, then you do 247 00:13:04,480 --> 00:13:07,800 take a significantly longer time to find pressing solution or 248 00:13:07,800 --> 00:13:10,640 recommendations. Like there are two challenges 249 00:13:10,640 --> 00:13:14,160 with the advent or or the adoption of the data culture if 250 00:13:14,160 --> 00:13:16,240 you will, right. One is like the scale of the 251 00:13:16,240 --> 00:13:18,760 problem has become much, much, much bigger and the scope has 252 00:13:18,760 --> 00:13:22,400 become much, much, much bigger. So like there was a time when 253 00:13:22,400 --> 00:13:24,760 like I had a customer who had like millions of segments 254 00:13:24,760 --> 00:13:27,080 because they were talking like millions of skews, right. 255 00:13:27,840 --> 00:13:31,600 And with customers like those, what you have is typically you 256 00:13:31,600 --> 00:13:35,920 have 8020 rule pretty much everyone actually where 80% of 257 00:13:35,920 --> 00:13:38,080 your revenue comes from 20% of your products, right. 258 00:13:38,960 --> 00:13:42,000 And if you were leaning on traditional segment, sorry, 259 00:13:42,080 --> 00:13:44,960 elasticity methods, right, let's just use elasticity as an 260 00:13:44,960 --> 00:13:46,120 example. We don't want to talk 261 00:13:46,120 --> 00:13:50,720 segmentation. And if you're trying to compute 262 00:13:50,760 --> 00:13:53,120 elasticity for those millions of segments, you're talking about 263 00:13:53,120 --> 00:13:54,720 running million regressions, right? 264 00:13:55,320 --> 00:13:57,640 And those are really time consuming. 265 00:13:57,680 --> 00:14:01,320 And again, like I talked about, they're very prone to outliers, 266 00:14:01,520 --> 00:14:04,280 80% of your business, like I said, also 20% of products. 267 00:14:04,360 --> 00:14:08,240 Still, that means that almost a large chunk of your business 268 00:14:08,240 --> 00:14:11,480 doesn't have enough transactions to accurately estimate 269 00:14:11,480 --> 00:14:14,560 elasticity. In that case, what happens is 270 00:14:14,560 --> 00:14:17,920 like if you're running the regressions, first of all for a 271 00:14:17,920 --> 00:14:22,920 thin or sparse segment where you do not have enough data, your 272 00:14:23,000 --> 00:14:27,320 results can swing from one way to other with like injection of 273 00:14:27,320 --> 00:14:30,760 newer data, right? Like we can estimate like, hey, 274 00:14:30,920 --> 00:14:34,320 for this particular SKU, oh it's not elastic and then you have 275 00:14:34,320 --> 00:14:36,640 like few months of transaction data coming and all of a sudden 276 00:14:36,640 --> 00:14:39,760 there is few outlier and it goes from inelastic to super elastic. 277 00:14:40,280 --> 00:14:42,320 That is not good. You don't want any of those. 278 00:14:44,080 --> 00:14:45,680 So. And the second thing is like if 279 00:14:45,680 --> 00:14:48,960 you are having like too much products, compute elasticities 280 00:14:48,960 --> 00:14:51,200 like millions of regressions. It's time consuming right? 281 00:14:51,240 --> 00:14:56,720 So to address these challenges we brought in like 2 methods, 282 00:14:56,720 --> 00:14:59,400 right? One method was to be able to 283 00:14:59,400 --> 00:15:02,120 work with less amount of data. So most of the data science 284 00:15:02,120 --> 00:15:04,640 algorithms are data hungry. They need a lot of data to give 285 00:15:04,640 --> 00:15:07,200 good protections. But there are certain algorithms 286 00:15:07,200 --> 00:15:10,040 that like deal with like less amount of data. 287 00:15:10,040 --> 00:15:12,400 So for example a segment doesn't have enough data. 288 00:15:12,720 --> 00:15:16,280 Instead of like collecting data or climbing above the product 289 00:15:16,280 --> 00:15:19,360 hierarchy and computer regressions or elasticity at a 290 00:15:19,400 --> 00:15:22,880 product, some are higher in the product hierarchy, right? 291 00:15:22,960 --> 00:15:27,240 You could like look at the closest queues that are like 292 00:15:27,760 --> 00:15:30,080 similar to this queue we are dealing with which doesn't have 293 00:15:30,080 --> 00:15:31,720 enough data. So there are scientific 294 00:15:31,720 --> 00:15:35,840 hierarchical methods which will go up and up until it reaches a 295 00:15:35,840 --> 00:15:41,160 certain data density and is able to compute LSCS accurately. 296 00:15:41,600 --> 00:15:43,920 The good thing about these hierarchical regressions which 297 00:15:43,920 --> 00:15:48,640 are like Bayesian in nature, you can have robust results which 298 00:15:48,640 --> 00:15:51,320 doesn't change with like advent of a few data points, sorry, 299 00:15:51,360 --> 00:15:53,160 injection of few data points in the model. 300 00:15:54,120 --> 00:15:56,400 So stable robust results. So that was one set of 301 00:15:56,400 --> 00:15:59,640 algorithms that we implemented like hey algorithms that are 302 00:15:59,640 --> 00:16:02,680 like less data hungry and still have robust results. 303 00:16:03,240 --> 00:16:06,680 The other question or problem was about scale, right. 304 00:16:07,120 --> 00:16:10,720 So previously what you you would do in traditional methods is 305 00:16:10,720 --> 00:16:15,680 just rely on accurately estimating elasticity of top or 306 00:16:15,680 --> 00:16:19,120 faster in skews and just like give some elasticity number at a 307 00:16:19,120 --> 00:16:22,120 higher level for the remaining skews, right. 308 00:16:22,360 --> 00:16:24,960 So for you to be able to run that regression, you need a 309 00:16:24,960 --> 00:16:27,880 method. So like traditional older 310 00:16:28,160 --> 00:16:32,440 methods that and this elasticity is relied on algorithms that 311 00:16:32,520 --> 00:16:36,280 used to take a long amount of time. 312 00:16:36,280 --> 00:16:39,480 Let's just put like in some cases 50 hours for a million 313 00:16:39,480 --> 00:16:41,840 regressions. Yeah yeah. 314 00:16:41,960 --> 00:16:44,800 And we implemented newer algorithms sampling techniques 315 00:16:44,800 --> 00:16:47,360 like. I mean, the new robot sampling 316 00:16:47,360 --> 00:16:51,720 technique is called like a new no U-turn sampling NUTS, NUTS is 317 00:16:51,720 --> 00:16:55,600 able to converge that same 50 hours of work in less than five 318 00:16:55,600 --> 00:16:57,080 hours. So it's almost like you'll see 319 00:16:57,080 --> 00:17:00,640 90% reductions in runtime. And it's basically data science 320 00:17:00,680 --> 00:17:04,359 and machine learning algorithms that learn things faster and 321 00:17:04,359 --> 00:17:07,240 converge to a solution faster. So you can solve the problem of 322 00:17:07,240 --> 00:17:08,720 scale. You can solve the problem of 323 00:17:08,800 --> 00:17:11,839 accuracy being affected by few data points using these 324 00:17:11,839 --> 00:17:15,839 techniques. I can imagine that comes in. 325 00:17:15,920 --> 00:17:19,640 That's very convenient in today's today's world as far as 326 00:17:19,640 --> 00:17:22,720 the amount of data we have in certain industries. 327 00:17:23,839 --> 00:17:25,880 Now you've been instrumental in assisting various companies 328 00:17:26,200 --> 00:17:29,360 including Fortune 500 companies and building and refining their 329 00:17:29,360 --> 00:17:33,200 pricing programs. How do you even measure the 330 00:17:33,200 --> 00:17:36,720 tangible value of your solution? Like how do you measure what 331 00:17:36,720 --> 00:17:39,040 that, what that brings to the table? 332 00:17:39,040 --> 00:17:42,400 Are there any specific metrics or analytical frameworks you can 333 00:17:42,400 --> 00:17:44,720 share with us? Yeah, absolutely. 334 00:17:44,720 --> 00:17:48,000 I mean and honestly like in my experience, not much attention 335 00:17:48,000 --> 00:17:50,920 has been paid to this topic in the bTB space. 336 00:17:51,840 --> 00:17:54,760 Typically companies implement pricing strategies and don't 337 00:17:54,760 --> 00:17:57,760 really have a scientific method to quantify what is the benefit 338 00:17:57,760 --> 00:17:59,480 I'm getting out of this pricing program, right. 339 00:18:00,320 --> 00:18:03,760 So, and if you don't measure what what you're, what you're 340 00:18:03,760 --> 00:18:04,960 working with, what is the benefit? 341 00:18:04,960 --> 00:18:09,120 There is no way, no good way to improve the that process, right, 342 00:18:09,120 --> 00:18:12,200 The pricing strategy and in essence of which you are just 343 00:18:12,200 --> 00:18:14,840 like you cannot continuously improve, right. 344 00:18:14,840 --> 00:18:18,880 The continuous improvement relies on effectively measuring 345 00:18:18,880 --> 00:18:22,720 it. And this is a problem I had to 346 00:18:22,720 --> 00:18:25,600 deal with multiple times. So I leaned on my operations 347 00:18:25,600 --> 00:18:29,520 research background to and to this background to solve for 348 00:18:29,520 --> 00:18:35,240 this problem, right. So I leaned on traditional 349 00:18:36,160 --> 00:18:38,440 statistical techniques to develop store tests. 350 00:18:38,440 --> 00:18:41,800 So basically what you could do is, the crux of this method is, 351 00:18:41,800 --> 00:18:45,120 is that you can select a relatively small amount of 352 00:18:45,120 --> 00:18:48,840 products and a relatively small amount of stores to be 353 00:18:48,840 --> 00:18:51,800 statistically representative of the whole fleet or the whole 354 00:18:51,800 --> 00:18:53,520 business that we're talking about, right? 355 00:18:54,880 --> 00:18:58,960 And and then implement a pricing program or a pricing strategy in 356 00:18:58,960 --> 00:19:01,360 that narrow set of products and store combination. 357 00:19:01,760 --> 00:19:05,200 Look at the benefit against another similar set of stores 358 00:19:05,200 --> 00:19:07,800 and products to see how much of A lift that you are getting out 359 00:19:07,800 --> 00:19:11,280 of it. And what happens from this is 360 00:19:11,280 --> 00:19:14,720 the key objective of this design is to achieving a kind of 361 00:19:14,720 --> 00:19:19,280 balance where the selected products and the selected stores 362 00:19:19,280 --> 00:19:23,040 are representative like I said and small enough so that if 363 00:19:23,040 --> 00:19:26,000 something goes S, they're not affecting the whole business, 364 00:19:26,000 --> 00:19:28,560 right. And how you do it is you want to 365 00:19:28,560 --> 00:19:33,120 select products that are and needed to ensure equivalence 366 00:19:33,120 --> 00:19:36,080 across like multiple dimensions, like business divisions. 367 00:19:36,080 --> 00:19:38,200 Like if you have like multiple business divisions, you want to 368 00:19:38,200 --> 00:19:39,720 have equivalence. You don't want to select 369 00:19:39,720 --> 00:19:43,680 products that are top revenue across and a business division 370 00:19:43,680 --> 00:19:45,880 is not represented. Similarly, you want to select 371 00:19:45,880 --> 00:19:48,360 products across product categories, different product 372 00:19:48,360 --> 00:19:51,880 velocities, product profitabilities and metrics like 373 00:19:51,880 --> 00:19:55,000 margin markup, even seasonality. So you can like slice the 374 00:19:55,000 --> 00:19:56,440 products based on those dimensions. 375 00:19:56,960 --> 00:19:59,720 Similarly stores you can slice the stores and select stores 376 00:19:59,720 --> 00:20:03,480 from based on region, traffic, profitability and and 377 00:20:03,480 --> 00:20:06,760 performance metrics like store labor, size of the store 378 00:20:07,240 --> 00:20:11,560 location, volume that that that store deals with. 379 00:20:12,120 --> 00:20:14,360 And then you can kind of form with those pieces of 380 00:20:14,360 --> 00:20:16,920 information, you can formulate the optimization problem like, 381 00:20:16,920 --> 00:20:24,280 hey, with these dimensions of products and excuse me stores. 382 00:20:25,520 --> 00:20:29,560 Find me a combination that is representative. 383 00:20:29,560 --> 00:20:34,520 Why fleet in terms of volume, in terms of revenue, in terms of 384 00:20:34,520 --> 00:20:37,240 representativeness across the dimensions that I talked about 385 00:20:37,920 --> 00:20:41,440 and and and and it will not be a perfect match. 386 00:20:41,440 --> 00:20:45,200 Like for example, let's just say your PROC Division One is 50% of 387 00:20:45,200 --> 00:20:47,280 your business. These methods can get you to 388 00:20:47,280 --> 00:20:51,240 like 48 percent 49% representativeness right and and 389 00:20:51,240 --> 00:20:54,840 and once you have these you're able to get like a near 390 00:20:54,840 --> 00:20:56,720 representativeness. And typically with these kind of 391 00:20:56,720 --> 00:20:59,440 tests you can simultaneously test like multiple pricing 392 00:20:59,440 --> 00:21:03,160 strategies and like in statistical speak they're like 393 00:21:03,240 --> 00:21:05,400 AB test. But the design of it is like a 394 00:21:05,400 --> 00:21:08,160 little more nuanced to achieve equivalence. 395 00:21:08,920 --> 00:21:11,320 And yeah. And once you have these design, 396 00:21:11,320 --> 00:21:15,040 these tests, you can like come up with multiple pricing 397 00:21:15,040 --> 00:21:18,960 strategies, implement those at a smaller scale, understand the 398 00:21:18,960 --> 00:21:22,080 benefit out of it and then like depending on which ones are 399 00:21:22,080 --> 00:21:26,080 working versus not, you scale up right and quantify the benefit 400 00:21:26,080 --> 00:21:27,400 of it. And if it's not working, you can 401 00:21:27,400 --> 00:21:30,480 continuously assess like, OK, does my pricing program kind of 402 00:21:30,480 --> 00:21:32,760 needs a refinement because the benefit that I was seeing over a 403 00:21:32,760 --> 00:21:35,760 period of time has kind of stagnated or like going down. 404 00:21:35,800 --> 00:21:38,040 So it just gives you a continuous read into how your 405 00:21:38,040 --> 00:21:43,120 strategy is doing and and then and you can continuously keep on 406 00:21:43,120 --> 00:21:45,040 refining it. OK. 407 00:21:45,080 --> 00:21:48,320 Now let me ask you this as well in correlation to what you just 408 00:21:48,320 --> 00:21:52,640 said, the selecting products that don't aren't too heavily 409 00:21:52,640 --> 00:21:55,440 involved with the revenue. You know if it doesn't sell too 410 00:21:55,440 --> 00:21:58,120 much, you know it's not going to make make the company go belly 411 00:21:58,120 --> 00:21:59,920 up. How often do you run these type 412 00:21:59,920 --> 00:22:04,760 of tests? Actually you can run it once in 413 00:22:04,760 --> 00:22:07,360 a while like for example if you have a new strategy you can run 414 00:22:07,360 --> 00:22:11,880 this test like as a one go see like hey one time six weeks, 12 415 00:22:11,880 --> 00:22:15,320 weeks test and you need to have like the product store and the 416 00:22:15,320 --> 00:22:18,600 duration and it's a statistical formula that feeds that and 417 00:22:18,600 --> 00:22:21,000 you'll tell, it will tell you like hey if you run these tests 418 00:22:21,000 --> 00:22:23,520 for these long number of days with these number of products 419 00:22:23,520 --> 00:22:27,960 and stores and you expect to observe X percentage of lift. 420 00:22:28,880 --> 00:22:30,600 Here is what how long it's going to take. 421 00:22:30,640 --> 00:22:35,280 You can run these tests as a one off or you can have a continuous 422 00:22:35,280 --> 00:22:38,680 like a as they call it like a control study where you always 423 00:22:38,680 --> 00:22:42,040 keep a small set where you're not putting any of your current 424 00:22:42,040 --> 00:22:44,960 pricing strategy and then compare it and continuously keep 425 00:22:44,960 --> 00:22:46,160 on checking. That is what I was talking 426 00:22:46,160 --> 00:22:48,520 about. Like you can have these tests 427 00:22:48,520 --> 00:22:51,440 that can be one off to test the strategy or these tests. 428 00:22:51,520 --> 00:22:54,240 This test could be like continuously running over a time 429 00:22:54,240 --> 00:22:56,800 horizon and you can continuously refine it. 430 00:22:56,800 --> 00:23:00,560 If there is like a product that is not included or is a new in 431 00:23:00,560 --> 00:23:03,200 your assortment, you can refine the assortment mix and then keep 432 00:23:03,200 --> 00:23:04,360 going. OK, OK. 433 00:23:04,360 --> 00:23:06,720 And that's more of like a controlled testing that makes 434 00:23:06,720 --> 00:23:10,280 sense now as it pertains to adoption, you know, driving 435 00:23:10,280 --> 00:23:14,120 adoption especially within sales teams, it's often a challenge 436 00:23:14,120 --> 00:23:17,320 when introducing a new pricing strategy as many of us may may 437 00:23:17,320 --> 00:23:22,040 already know are aware of what are some strategies or 438 00:23:22,040 --> 00:23:25,040 experiences you have successfully navigated this 439 00:23:25,040 --> 00:23:29,280 hurdle and ensured seamless integration within teams. 440 00:23:30,280 --> 00:23:32,920 Yeah, this is just a good example actually. 441 00:23:33,240 --> 00:23:37,560 Good question actually and a few that come to my mind and some of 442 00:23:37,560 --> 00:23:42,280 my customers have extensively spoken about it on public 443 00:23:42,280 --> 00:23:46,360 forums. So basically if you don't ask 444 00:23:46,480 --> 00:23:54,240 anyone, I mean there is always a balance between trying to get 445 00:23:54,240 --> 00:23:58,800 the highest price or not losing the business, right. 446 00:23:59,640 --> 00:24:04,560 So you have to incentivize those instead of incentivizing volume. 447 00:24:04,560 --> 00:24:08,520 For example, if your incentives or adopting the price is like, 448 00:24:08,520 --> 00:24:10,760 hey, how many transactions you can close, it's a race to 449 00:24:10,760 --> 00:24:12,400 bottom. Everyone will go to the lowest 450 00:24:12,400 --> 00:24:15,240 price that you're recommending and try to close as many deals 451 00:24:15,240 --> 00:24:17,840 as possible. However, if you're able to 452 00:24:17,840 --> 00:24:22,680 incentivize like, hey, if you are like selling at a stock 453 00:24:22,680 --> 00:24:25,240 price, your incentive is going to be widely different than if 454 00:24:25,240 --> 00:24:27,160 you're selling at a floor price, right? 455 00:24:28,280 --> 00:24:32,880 So we had some customers who were trying to optimize gross 456 00:24:32,880 --> 00:24:35,720 margin dollars while incentivizing number of deals 457 00:24:35,800 --> 00:24:39,120 done and that was like counterintuitive defining your 458 00:24:39,120 --> 00:24:42,320 sales compensation program is a good way to kind of Dr. 459 00:24:42,320 --> 00:24:45,760 adoption. The other is simply asking 460 00:24:45,760 --> 00:24:48,800 people for like why do they need a price exception. 461 00:24:49,160 --> 00:24:51,280 So for example, we had a customer where like they were 462 00:24:51,280 --> 00:24:55,720 seeing a lot of price override. And what they did was they added 463 00:24:55,760 --> 00:24:59,480 a simple widget where you have to select from like 5 or 6 464 00:24:59,880 --> 00:25:02,600 regions. Why you think the price is not 465 00:25:02,600 --> 00:25:05,440 right and why you need the exception And you cannot leave 466 00:25:05,560 --> 00:25:06,840 others that you cannot leave blank. 467 00:25:06,840 --> 00:25:09,520 You have to give a reason. And those are concrete reasons. 468 00:25:09,920 --> 00:25:12,880 And all of a sudden the adoption increased because if you people 469 00:25:12,880 --> 00:25:16,400 can select others or not give a reason, they will always like oh 470 00:25:16,400 --> 00:25:21,240 right, right. And yeah, And I had one more 471 00:25:21,240 --> 00:25:24,280 customer actually. They were like pretty good in 472 00:25:24,280 --> 00:25:27,400 the sense like what they did was basically when you have a price 473 00:25:27,400 --> 00:25:30,280 that is displayed for the salesperson and if you are not 474 00:25:30,280 --> 00:25:33,560 feeling confident about, there is a widget to explain, you hit 475 00:25:33,560 --> 00:25:35,320 that explain button. It's just like hey, your 476 00:25:35,320 --> 00:25:38,400 transaction is similar to these transactions that have happened 477 00:25:38,400 --> 00:25:41,840 in the last six months or a year and this is the price they have 478 00:25:41,840 --> 00:25:44,760 realized. So it and that kind of gave 479 00:25:44,760 --> 00:25:46,880 confidence to the salesperson, like hey, the price that I'm 480 00:25:46,880 --> 00:25:50,280 going to offer is not going to end up in the lost business. 481 00:25:50,480 --> 00:25:52,640 So it was more like building confidence. 482 00:25:52,640 --> 00:25:55,640 So the multiple things people can do, but what I've seen is 483 00:25:57,080 --> 00:25:59,920 rewarding the right things and not rewarding the wrong things 484 00:25:59,920 --> 00:26:01,800 like other things that are not aligned with your business or 485 00:26:01,800 --> 00:26:03,880 pricing strategy is the most effective way. 486 00:26:05,080 --> 00:26:06,760 That's interesting, that's good to know. 487 00:26:08,480 --> 00:26:11,760 Now I want to move briefly aside from adoption and just looking 488 00:26:11,760 --> 00:26:14,800 at beyond you know things as aspects beyond pricing. 489 00:26:15,640 --> 00:26:19,080 Are there other areas within the realm of B to B commerce where 490 00:26:19,080 --> 00:26:22,000 you foresee the integration of data science and AI playing a 491 00:26:22,000 --> 00:26:24,680 pivotal role? Any emerging trends or 492 00:26:24,680 --> 00:26:27,080 opportunities that may be catching your eye? 493 00:26:28,120 --> 00:26:30,040 Yeah, Yeah. I mean, certainly and then like 494 00:26:30,160 --> 00:26:35,640 I have seen that happen with some of my customers and like a 495 00:26:35,640 --> 00:26:38,160 good example is just staying with the segmentation team, like 496 00:26:38,840 --> 00:26:42,960 you could use machine learning to cluster your customers into 497 00:26:42,960 --> 00:26:45,360 different profiles, right. And I'll give you a more 498 00:26:45,360 --> 00:26:48,120 relatable example. So think about it like your 499 00:26:48,120 --> 00:26:51,760 customer has like builders, manufacturers kind of thing and 500 00:26:51,760 --> 00:26:54,240 then one of your customer base, you're able to identify like, 501 00:26:54,240 --> 00:26:58,880 hey, these guys seem to be doing transactions that make it look 502 00:26:58,880 --> 00:27:00,920 like they are floating contractors, right. 503 00:27:01,480 --> 00:27:04,400 So you're able to create personas of different customers 504 00:27:04,400 --> 00:27:06,760 that you have and machine learning algorithms. 505 00:27:06,760 --> 00:27:10,640 There are a bunch of those that can accurately classify your 506 00:27:10,640 --> 00:27:13,400 cluster, those set of customers. Now think about just one 507 00:27:13,400 --> 00:27:15,000 customer, like floating contractors. 508 00:27:15,360 --> 00:27:18,200 Sure. Now in the floating contractor, 509 00:27:18,760 --> 00:27:20,200 what would a floating contractor buy? 510 00:27:20,240 --> 00:27:23,640 Will buy tile, will buy you reducers, transition pieces, 511 00:27:23,640 --> 00:27:25,240 stair noses, all the good things, right? 512 00:27:26,240 --> 00:27:29,320 So you have a persona that hey, of these customers within my 513 00:27:29,320 --> 00:27:33,600 flooring contractor, a customer base, pretty much everyone buys 514 00:27:33,600 --> 00:27:39,680 40 to 50% of their sales is, is on flooring type, 20% is 515 00:27:39,960 --> 00:27:42,320 reducers, stair noses, accessories, whatever you will. 516 00:27:43,040 --> 00:27:46,000 And then once you have that you can use again use machine 517 00:27:46,000 --> 00:27:48,720 learning algorithms, sort of even simple analytics really to 518 00:27:48,800 --> 00:27:52,080 say like, OK, are there any anomalies in these clusters? 519 00:27:52,280 --> 00:27:55,920 Like is there a customer who is buying pretty much the same 520 00:27:55,920 --> 00:27:58,400 number of floor tiles or same percentage of floor tiles that 521 00:27:58,480 --> 00:28:01,000 other contractors are buying, but is not buying reduced 522 00:28:01,000 --> 00:28:02,720 transition pieces, accessories, right. 523 00:28:03,320 --> 00:28:05,760 And you can identify areas of growth there. 524 00:28:07,440 --> 00:28:10,040 So that's that's like a very good use case of like clustering 525 00:28:10,040 --> 00:28:12,680 similar customers together and then within that cluster 526 00:28:12,680 --> 00:28:16,080 identifying anomaly to find areas where you can grow your 527 00:28:16,080 --> 00:28:18,680 business that's. Cool. 528 00:28:19,320 --> 00:28:20,920 Yeah, yeah. And then there's another 529 00:28:20,920 --> 00:28:22,480 example. Actually, I worked on this one 530 00:28:22,480 --> 00:28:25,160 with another customer. So I had a client who stocked 531 00:28:25,160 --> 00:28:28,400 like standard parts and also built custom parts like 532 00:28:28,400 --> 00:28:30,840 electrical panels and and things like that, right. 533 00:28:31,400 --> 00:28:34,240 And many times like customers would come to them and say like, 534 00:28:34,240 --> 00:28:36,880 hey, I'm looking for this part XYZ part right. 535 00:28:37,720 --> 00:28:41,920 And that would often be the case that they didn't have the exact 536 00:28:41,920 --> 00:28:45,320 same part but they had like a perfectly good substitute of it 537 00:28:45,320 --> 00:28:49,320 in the stock and it was very hard for them, manually 538 00:28:49,320 --> 00:28:52,440 cumbersome and error prone to have that mapping. 539 00:28:52,440 --> 00:28:55,360 Like hey, I have this product, here are the 15 other products 540 00:28:55,360 --> 00:28:57,680 that are a perfectly good substitute and can be used in 541 00:28:57,680 --> 00:29:00,760 its place. But for them to manually 542 00:29:00,760 --> 00:29:04,480 maintain that mapping across multiple products and as their 543 00:29:04,480 --> 00:29:07,360 assortment grew or the product offerings increased, it was 544 00:29:07,360 --> 00:29:09,800 becoming harder and hard for them to keep that mapping or 545 00:29:09,800 --> 00:29:11,840 like maintain the list of substitute products that could 546 00:29:11,840 --> 00:29:16,000 go right. So like I worked on like 547 00:29:16,000 --> 00:29:18,800 building a machine learning solution for them that will 548 00:29:18,800 --> 00:29:21,880 create this mapping by kind of identifying product attributes. 549 00:29:21,880 --> 00:29:24,680 Like you have a free form text description like hey, this is X 550 00:29:24,880 --> 00:29:30,280 world, switchboard, whatever. And then you have product 551 00:29:30,280 --> 00:29:33,680 attributes, descriptions like 3 phone texts like I talked about. 552 00:29:34,200 --> 00:29:38,000 And it would create a similarity score based on like, OK, these 553 00:29:38,000 --> 00:29:40,240 products are similar to this product and this is how, like 554 00:29:40,240 --> 00:29:42,960 you'd create manually create a mapping but give that rule to 555 00:29:42,960 --> 00:29:45,960 machine to be able to faster and accurately figure it out. 556 00:29:46,680 --> 00:29:49,880 And that would create like a good list of substitute products 557 00:29:49,920 --> 00:29:51,520 and in this case, what they have to do. 558 00:29:51,520 --> 00:29:54,720 I mean of course you cannot take human in the loop, like human 559 00:29:54,720 --> 00:29:56,040 out of this loop. Sure. 560 00:29:56,040 --> 00:29:59,520 But the task reduced from like I define products to like looking 561 00:29:59,520 --> 00:30:01,240 at, hey, here is my machine's recommendation. 562 00:30:01,240 --> 00:30:03,960 Yeah, 90% of these are fine. These percent are not right. 563 00:30:04,400 --> 00:30:06,920 They introduced their workload and it was like a good use case 564 00:30:07,000 --> 00:30:10,480 of machine learning and making your product offerings more 565 00:30:10,480 --> 00:30:12,840 robust, right. OK. 566 00:30:13,520 --> 00:30:14,080 Yeah. That's good. 567 00:30:14,760 --> 00:30:20,000 Now Mr. Vivek Anand with Gap, I want to, I want to close out and 568 00:30:20,000 --> 00:30:21,800 thank you so much for your time today. 569 00:30:22,520 --> 00:30:25,520 Before I let you go, you you did provide a lot of key insights, a 570 00:30:25,520 --> 00:30:30,520 lot of knowledgeable wisdom if you will in the AI and pricing 571 00:30:30,920 --> 00:30:35,720 and data science realm as regards to bTB pricing lease for 572 00:30:35,720 --> 00:30:38,520 those who are listening and are are one are interested in 573 00:30:38,520 --> 00:30:42,280 learning more about you and you know your resources, what you 574 00:30:42,280 --> 00:30:45,400 stand for maybe where can they go to learn more about you. 575 00:30:47,200 --> 00:30:49,240 That's a good question. I mean, like, I'm available on 576 00:30:49,240 --> 00:30:51,560 LinkedIn, so like a good place is to just connect with me on 577 00:30:51,560 --> 00:30:54,680 LinkedIn and like research to me with your questions that you 578 00:30:54,680 --> 00:30:56,240 have. And I'm more than happy to 579 00:30:56,240 --> 00:30:59,360 engage. But yeah, I mean I also am 580 00:30:59,360 --> 00:31:02,000 looking to like create a forum for myself. 581 00:31:02,000 --> 00:31:04,800 Like I like I mentioned previously, I not only deal with 582 00:31:04,800 --> 00:31:07,600 pricing, I also deal with inventory like buying and like 583 00:31:07,600 --> 00:31:11,200 Internet issues which are again like pretty prevalent in the B 584 00:31:11,200 --> 00:31:14,200 to B space. And I'm going to like I'm 585 00:31:14,200 --> 00:31:16,720 planning to actually. So I don't have a name yet. 586 00:31:16,720 --> 00:31:20,160 I have a a domain of myself where I will be like sharing 587 00:31:20,160 --> 00:31:23,640 these ideas, but LinkedIn is probably the best way to reach 588 00:31:23,640 --> 00:31:24,800 out to me or communicate with me. 589 00:31:25,720 --> 00:31:28,520 LinkedIn and and kind of keep up with you on the latest updates 590 00:31:28,520 --> 00:31:30,840 about this project you're you're working on personally that that 591 00:31:30,840 --> 00:31:34,000 might be really cool. So staying, staying to know, I'm 592 00:31:34,000 --> 00:31:35,800 sorry, go ahead. Yeah, I think that, yeah, 593 00:31:35,840 --> 00:31:37,760 absolutely I can. LinkedIn will be the first place 594 00:31:37,760 --> 00:31:40,480 where I'll like talk about this project where I have like these 595 00:31:40,480 --> 00:31:43,520 things like codes and and use cases and things like that. 596 00:31:43,680 --> 00:31:46,040 OK, OK, cool. Well, for those who are 597 00:31:46,040 --> 00:31:49,440 listening, stay in tune because he has great things in store for 598 00:31:49,440 --> 00:31:52,400 you all in the times to come. Until next time. 599 00:31:52,520 --> 00:31:53,760 We'll see you all later. Bye, Bye.