1 00:00:00,401 --> 00:00:04,481 Too often artificial intelligence in marketing is about volume, pushing out 2 00:00:04,481 --> 00:00:06,971 more content, more ads, more noise. 3 00:00:07,451 --> 00:00:11,231 But today we'll hear a different story, one where AI is used like 4 00:00:11,231 --> 00:00:15,821 a scientist measuring, testing and refining to ensure every marketing 5 00:00:15,821 --> 00:00:17,681 dollar drives real results. 6 00:00:18,041 --> 00:00:21,401 It's a smarter, more precise way to grow without the waste. 7 00:00:21,851 --> 00:00:24,851 My name is Alexandre Nevski, and this is Innovation Tales. 8 00:00:26,300 --> 00:00:28,880 Navigating Change one story at a time. 9 00:00:28,910 --> 00:00:33,110 We share insights from leaders tackling the challenges of today's digital world. 10 00:00:33,380 --> 00:00:36,890 Welcome to Innovation Tales, the podcast exploring the human 11 00:00:36,890 --> 00:00:38,750 side of digital transformation. 12 00:00:39,246 --> 00:00:42,966 In this episode, we're tackling a big challenge in marketing attribution. 13 00:00:43,536 --> 00:00:47,076 How do you know which campaigns actually drive sales and which 14 00:00:47,076 --> 00:00:48,786 ones are wasting your budget? 15 00:00:48,816 --> 00:00:53,076 We will break down why ad platforms often give misleading data. 16 00:00:53,556 --> 00:00:58,296 How AI can reveal what really works and why testing, not tracking people 17 00:00:58,296 --> 00:00:59,856 is the key to better decisions. 18 00:01:00,576 --> 00:01:05,076 Our guest today is Zeke Camusio, founder of Data Speaks, an AI 19 00:01:05,076 --> 00:01:09,156 powered analytics platform that helps businesses spend smarter. 20 00:01:09,936 --> 00:01:13,956 With a background in economics and data science, he spent 20 years 21 00:01:14,106 --> 00:01:18,156 helping companies move beyond guesswork and make data-driven decisions. 22 00:01:19,041 --> 00:01:21,891 If you've ever wondered whether your marketing dollars are going 23 00:01:21,891 --> 00:01:25,221 to the right places, this episode will give you the answers. 24 00:01:25,581 --> 00:01:28,961 Without further ado, here's my conversation with Zeke Camusio. 25 00:01:29,675 --> 00:01:30,870 Zeke, welcome to the show. 26 00:01:31,092 --> 00:01:31,722 Thank you, Alex. 27 00:01:31,722 --> 00:01:32,652 It's good to be here. 28 00:01:32,862 --> 00:01:35,082 Well, I'm so glad you found the time for this conversation. 29 00:01:35,682 --> 00:01:41,142 I have been looking for an angle to cover AI impact on marketing, but I 30 00:01:41,142 --> 00:01:44,682 needed an expert like you, someone with a background in data science. 31 00:01:45,047 --> 00:01:48,347 Let's make sure we bring the audience along however, and set the context 32 00:01:48,347 --> 00:01:50,407 while providing a few definitions. 33 00:01:51,517 --> 00:01:54,277 We agreed to discuss marketing attribution. 34 00:01:54,817 --> 00:01:58,417 So how did you come to focus on this capability and how do you 35 00:01:58,417 --> 00:02:01,642 explain the importance to people outside of the marketing team? 36 00:02:01,900 --> 00:02:02,455 Of course. 37 00:02:02,545 --> 00:02:08,785 So in any business, regardless of what kind of business it is, usually 38 00:02:08,785 --> 00:02:14,065 what happens is that you have a set of activities you do that you hope 39 00:02:14,395 --> 00:02:17,305 to drive a certain outcome, and then you have the outcome itself. 40 00:02:17,635 --> 00:02:22,875 So an example of that is if it's B2B, maybe you're trying 41 00:02:22,875 --> 00:02:24,435 to get leads from your website. 42 00:02:24,465 --> 00:02:28,065 If you have an e-commerce store, maybe you're trying to get sales. 43 00:02:28,635 --> 00:02:31,305 Regardless of the outcome, there are certain things that you do. 44 00:02:31,305 --> 00:02:34,455 You advertise on LinkedIn, on Google, on Facebook. 45 00:02:34,455 --> 00:02:36,225 You go to trade shows. 46 00:02:36,225 --> 00:02:36,765 You. 47 00:02:37,105 --> 00:02:43,125 send emails, do podcasts, and none of these, activities, even though 48 00:02:43,125 --> 00:02:47,895 they have data that they provide in dashboards, none of them can give you 49 00:02:47,895 --> 00:02:52,454 a real understanding of the impact they're having on your outcome. 50 00:02:52,454 --> 00:02:56,134 So if you got a million dollars in revenue at the end of the month, how do 51 00:02:56,134 --> 00:03:00,694 you know how much of that was driven by Google, Facebook, Instagram, and so on. 52 00:03:01,024 --> 00:03:05,414 And if you don't know that, how can you possibly decide what's the best 53 00:03:05,414 --> 00:03:07,934 way to invest your marketing dollars? 54 00:03:08,154 --> 00:03:09,624 So that's the problem we solve. 55 00:03:09,624 --> 00:03:14,004 We give marketers the ability to track what drives their conversions 56 00:03:14,004 --> 00:03:17,364 and revenue so they can invest in the right channels and campaigns. 57 00:03:18,114 --> 00:03:21,804 And were you already in marketing when you started your latest venture? 58 00:03:22,614 --> 00:03:22,944 Yeah. 59 00:03:22,944 --> 00:03:23,094 Yeah. 60 00:03:23,094 --> 00:03:28,924 I, this is all I've done since I, graduated high school, about 25 years ago. 61 00:03:29,194 --> 00:03:34,324 Had my several businesses on my own and had a marketing agency for a few 62 00:03:34,324 --> 00:03:37,684 years, and when it was acquired in 2015. 63 00:03:38,524 --> 00:03:43,784 I was working with a few, clients, doing consulting and this one client 64 00:03:43,784 --> 00:03:51,284 that saw a metric for Facebook ads known as ROAS or return on ad spent, 65 00:03:51,734 --> 00:03:54,464 and the ROAS said it was five. 66 00:03:54,564 --> 00:03:58,484 What that should mean is that you could invest if, $10,000 67 00:03:58,574 --> 00:04:00,554 and get $50,000 in revenue. 68 00:04:01,274 --> 00:04:02,114 So that happened. 69 00:04:02,114 --> 00:04:08,354 He invested $10,000 on that channel, and he got $20,000 in revenue back, and he 70 00:04:08,354 --> 00:04:13,084 was, I. and we know we didn't manage the campaign or anything, but he was really 71 00:04:13,084 --> 00:04:18,484 confused by the fact that he got a 2X return on investment instead of 5X as the 72 00:04:18,484 --> 00:04:21,994 platform reported, and he didn't know why. 73 00:04:21,994 --> 00:04:24,484 I had been working in digital marketing for a long time. 74 00:04:24,484 --> 00:04:28,184 By then, I knew that the customer journey is very, complex. 75 00:04:28,184 --> 00:04:34,334 People interact with you across multiple channels, multiple devices, and no one 76 00:04:34,334 --> 00:04:39,884 platform can give you a true understanding of the influence on, revenue in this 77 00:04:39,884 --> 00:04:43,379 case, because none of these platforms has access to the other platforms. 78 00:04:43,794 --> 00:04:48,664 for example, if somebody clicks on a, an on an Instagram ad and Google 79 00:04:48,664 --> 00:04:52,624 ad, those two platforms will take credit for the same sale because 80 00:04:52,624 --> 00:04:53,854 they don't talk to each other. 81 00:04:54,034 --> 00:04:57,854 Neither one of them has access to your source of truth, which is your web sales. 82 00:04:58,394 --> 00:05:04,319 And, as a result, I noticed that it wasn't only this client, but a lot of our clients 83 00:05:04,319 --> 00:05:09,239 were having an issue with not being able to know what kind of return investment 84 00:05:09,269 --> 00:05:12,689 they were actually getting for the dollars they were putting into marketing. 85 00:05:12,959 --> 00:05:14,369 So that gave me the idea. 86 00:05:14,419 --> 00:05:19,219 it became really evident at that point that, it was a, really big problem 87 00:05:19,219 --> 00:05:21,349 that marketers all over the world had. 88 00:05:21,679 --> 00:05:23,669 And they had just accepted that. 89 00:05:23,974 --> 00:05:26,554 there was no reliable way to figure that out. 90 00:05:26,954 --> 00:05:30,344 So that, that's what inspired me to start Data Speaks. 91 00:05:30,979 --> 00:05:38,149 And I guess, traditional marketing before digital was very much magic anyways 92 00:05:38,149 --> 00:05:43,519 in, in many respects because yeah, you don't necessarily have the data, you 93 00:05:43,519 --> 00:05:48,074 have to extrapolate, you have to, to sample I guess With digital technology 94 00:05:48,074 --> 00:05:53,708 now you have a lot more opportunities to leverage, the signals you're getting back. 95 00:05:53,798 --> 00:06:00,628 But tell us more about like how easy or difficult it was to find 96 00:06:00,628 --> 00:06:04,258 a, a product or platform that, that can, provide this capability. 97 00:06:05,293 --> 00:06:05,893 Yeah, of course. 98 00:06:05,903 --> 00:06:08,753 It's very relevant to what you said before. 99 00:06:08,803 --> 00:06:12,443 Before the internet, we didn't have pixels. 100 00:06:12,473 --> 00:06:16,943 For example, Google Analytics has a pixel that helps you track website usage. 101 00:06:17,753 --> 00:06:20,903 Facebook ads, Google ads, every platform gives you a pixel. 102 00:06:21,723 --> 00:06:24,063 And prior to the internet we didn't have that, right? 103 00:06:24,063 --> 00:06:29,373 So we had to essentially rely on modeling, media mix modeling, and 104 00:06:29,763 --> 00:06:33,753 it worked for media that where you couldn't track individual users 105 00:06:34,083 --> 00:06:35,703 or individual customers, right? 106 00:06:35,713 --> 00:06:37,926 it was perfect for radio, tv, print. 107 00:06:38,843 --> 00:06:43,318 And the internet then gave us the illusion that, oh, now we can track 108 00:06:43,318 --> 00:06:45,568 things on a much deeper level. 109 00:06:46,258 --> 00:06:51,478 And although there's some truth to that, I think it also gave us the illusion 110 00:06:51,478 --> 00:06:55,298 that we could get data that is way more accurate than it actually is. 111 00:06:55,298 --> 00:07:01,418 So when I started looking into this, I essentially kinda 112 00:07:01,418 --> 00:07:02,498 saw the writing on the wall. 113 00:07:02,498 --> 00:07:06,558 I saw that a lot of the startups that were trying to solve this problem 114 00:07:06,618 --> 00:07:09,078 was were taking the wrong approach. 115 00:07:09,108 --> 00:07:12,198 They were trying to track individual users across every 116 00:07:12,198 --> 00:07:15,408 device, every channel with pixels. 117 00:07:15,408 --> 00:07:18,798 But you know, these users were demanding more privacy. 118 00:07:18,798 --> 00:07:24,228 So there were more users on using ad blockers, enabling privacy settings. 119 00:07:24,288 --> 00:07:29,488 And essentially if you wanna, track people that don't want to be tracked, 120 00:07:29,488 --> 00:07:30,958 you're gonna have a really tough time. 121 00:07:31,748 --> 00:07:36,858 I started looking at, how this was done in many other scientific communities 122 00:07:36,858 --> 00:07:41,258 and how different areas of science were looking for these correlations 123 00:07:41,258 --> 00:07:43,148 and causal, causal relationships. 124 00:07:43,208 --> 00:07:48,838 And, it became evident to me that really what you wanted to know as a marketer 125 00:07:49,198 --> 00:07:56,128 is what happens to my revenue for every dollar that, that I invest in X? 126 00:07:56,128 --> 00:07:58,168 It could be Google ads, Facebook, whatever. 127 00:07:58,918 --> 00:08:04,374 And we formulated a hypothesis, which is, tracking the increases or 128 00:08:04,374 --> 00:08:08,504 decreases in, in budget, would be a much better way to help you understand 129 00:08:08,504 --> 00:08:10,929 the impact of that on your revenue. 130 00:08:11,449 --> 00:08:16,534 So to give you an example, if you have a shop where you have a hundred customers 131 00:08:16,559 --> 00:08:22,234 a day and you start running radio ads, and now you get 120 customers a day, 132 00:08:22,684 --> 00:08:25,654 then you know that's helping you, right? 133 00:08:26,274 --> 00:08:30,674 if you start ru running radio ads and you still have a hundred customers a day, 134 00:08:30,694 --> 00:08:33,059 then the amount of impact is very low. 135 00:08:33,509 --> 00:08:35,039 That's a very simple example. 136 00:08:35,039 --> 00:08:39,554 And what makes this so complex is that there's a lot of different channels. 137 00:08:39,554 --> 00:08:43,044 We had to analyze the individual impact of each. 138 00:08:43,624 --> 00:08:49,144 What we do is we essentially look at each state every single day. 139 00:08:49,144 --> 00:08:54,044 So for example, Colorado, what happens to your, if you 140 00:08:54,044 --> 00:08:56,054 decrease your Google Ads spend. 141 00:08:56,564 --> 00:08:57,764 What happened to your conversions? 142 00:08:57,764 --> 00:08:59,864 Did it go down or up or stay the same? 143 00:09:00,164 --> 00:09:01,604 If you went down by how much? 144 00:09:01,844 --> 00:09:06,904 If you doubled, Facebook ads for Maine, what happens to your sales in that state? 145 00:09:07,314 --> 00:09:12,394 Now if you look at one state one day, you know there's no, no way you can 146 00:09:12,394 --> 00:09:14,494 make any inferences based on that. 147 00:09:15,074 --> 00:09:20,174 but with machine learning, if you have 50 daily observations on a daily basis, 148 00:09:20,554 --> 00:09:22,714 over time you see very clear patterns. 149 00:09:22,904 --> 00:09:28,548 It's literally taking the scientific process of experimentation and 150 00:09:28,548 --> 00:09:35,898 applying it to an area where, yeah, the alternative approach runs into, 151 00:09:36,108 --> 00:09:37,968 like you said, a lot of privacy issues. 152 00:09:38,013 --> 00:09:39,693 How do you control variables though? 153 00:09:39,693 --> 00:09:42,183 Because it's not the scientific experiment done in the lab. 154 00:09:42,183 --> 00:09:43,743 It's out there in the real world. 155 00:09:44,103 --> 00:09:48,978 I guess that must be like a methodological constraint that, you 156 00:09:48,978 --> 00:09:51,798 must work with your clients to respect. 157 00:09:52,888 --> 00:09:53,158 Yeah. 158 00:09:53,218 --> 00:09:56,488 So there are two ways that, we are, our model works. 159 00:09:56,538 --> 00:10:00,018 you can do design experiments or natural experiments. 160 00:10:00,198 --> 00:10:03,408 So with the design experiment, essentially you say, I wanna 161 00:10:03,408 --> 00:10:06,058 test, say Google ads, for example. 162 00:10:06,058 --> 00:10:12,728 I. So what the algorithm does is it makes a suggestion for what 163 00:10:12,728 --> 00:10:15,518 markets you should test and for how long you should run a test. 164 00:10:16,128 --> 00:10:21,488 Maybe it says, in Colorado and Maine, for 22 days, you can 165 00:10:22,088 --> 00:10:23,258 do one of these two things. 166 00:10:23,258 --> 00:10:27,128 so you can do a holdout where you say, I'm just gonna shut it off to 167 00:10:27,128 --> 00:10:28,988 see how much my conversions drop. 168 00:10:29,558 --> 00:10:33,868 Or you could say, I'm gonna scale it up by, 25-50% and see 169 00:10:33,868 --> 00:10:35,308 how much my sales increase. 170 00:10:35,488 --> 00:10:40,288 The algorithm does a really good job helping you find the markets that have the 171 00:10:40,288 --> 00:10:45,188 highest statistical significance, but the lowest, negative impact on the business. 172 00:10:45,188 --> 00:10:48,938 So if you have a lot of customers in California and New York, it's 173 00:10:48,938 --> 00:10:50,138 not gonna mess with those markets. 174 00:10:50,138 --> 00:10:54,438 It's gonna find markets that are smaller markets, but still, tell you a lot 175 00:10:54,468 --> 00:10:56,758 about the overall, country in this case. 176 00:10:57,508 --> 00:11:02,368 So that essentially helps you measure, the incrementality if you scale it up, 177 00:11:02,398 --> 00:11:04,888 or how much you lose if you, turn it off. 178 00:11:05,698 --> 00:11:09,058 Sometimes it's not possible to do an experiment and there's 179 00:11:09,058 --> 00:11:10,048 a lot of reasons for that. 180 00:11:10,048 --> 00:11:12,005 Sometimes, our clients don't really wanna. 181 00:11:12,085 --> 00:11:16,505 mess even with the small markets or, it's just not practical for whatever reason. 182 00:11:17,205 --> 00:11:22,580 So our model still needs to learn from, what happens with natural experiments. 183 00:11:22,640 --> 00:11:23,630 So it's very normal. 184 00:11:23,660 --> 00:11:26,750 even like in terms of demand, you are not gonna spend this 185 00:11:26,780 --> 00:11:28,370 the exact same amount every day. 186 00:11:28,420 --> 00:11:30,790 So even if you take one state, let's take Colorado. 187 00:11:31,650 --> 00:11:32,940 You're not gonna spend the same every day. 188 00:11:32,940 --> 00:11:35,040 So some days you're gonna spend more because people 189 00:11:35,040 --> 00:11:36,240 are searching more for stuff. 190 00:11:36,240 --> 00:11:39,550 Some, sometimes, people are not scrolling social media, so 191 00:11:39,550 --> 00:11:40,990 there's gonna be less demand. 192 00:11:41,090 --> 00:11:43,910 , And then over time you also increase or decrease budgets, 193 00:11:43,910 --> 00:11:47,670 sometimes pretty substantially, so that's what it, it analyzes. 194 00:11:47,670 --> 00:11:51,955 And again, it's, if it looks at one day, one state, could just be a coincidence. 195 00:11:51,955 --> 00:11:56,615 But if there's, say, if consistently every time you increase a dollar 196 00:11:56,615 --> 00:11:59,520 here you see $5 more revenue. 197 00:11:59,880 --> 00:12:02,270 And that happens over time. 198 00:12:02,300 --> 00:12:07,400 And based on that, the system makes predictions that turn to be very accurate. 199 00:12:07,760 --> 00:12:10,010 Then, that's a really good indication for the system. 200 00:12:10,400 --> 00:12:16,220 So our system allows for what in, in science is called randomized 201 00:12:16,220 --> 00:12:19,280 control trials where you have a control or a treatment. 202 00:12:19,890 --> 00:12:23,160 but even if you don't have the ability to do that, it still 203 00:12:23,160 --> 00:12:25,320 learns from natural experiments. 204 00:12:26,380 --> 00:12:30,880 I'm so glad that, I got to talk to you about marketing because I think, 205 00:12:31,010 --> 00:12:36,250 what I see, I. These days where, generative ai, for example, is being 206 00:12:36,250 --> 00:12:39,520 applied to marketing, it's in order to just push a lot of things, it's just 207 00:12:39,570 --> 00:12:41,535 to throw things and see what sticks. 208 00:12:41,535 --> 00:12:46,445 I think this is a very, clean, elegant approach. 209 00:12:46,895 --> 00:12:50,925 And before we move on and talk about the features specifically, but I'm 210 00:12:50,925 --> 00:12:58,050 really curious, how long did it take for you to get to a point where your 211 00:12:58,170 --> 00:13:00,570 clients can rely on the platform. 212 00:13:00,570 --> 00:13:05,550 And did you have to go for some time with more of a, like a hybrid 213 00:13:05,550 --> 00:13:09,090 manual consultancy plus the platform? 214 00:13:09,390 --> 00:13:14,910 Yeah, so we engaged about 20 different brands and agencies, in a pilot, and 215 00:13:14,910 --> 00:13:19,440 it took us about three years to build a solution to a point where we could predict 216 00:13:19,440 --> 00:13:25,080 with at least 95% accuracy what's gonna happen if you invest a dollar in whatever. 217 00:13:25,700 --> 00:13:28,620 and, yeah, they were, really gracious with us. 218 00:13:28,620 --> 00:13:32,990 They understood that we were building something that was never, didn't exist 219 00:13:33,140 --> 00:13:37,790 to the, this degree and didn't allow for this real time ingestion of data. 220 00:13:38,080 --> 00:13:42,530 That was converted into insights that helps you optimize right away, so they 221 00:13:42,530 --> 00:13:47,820 knew that we were going after a, a big challenge that all marketers have wasn't 222 00:13:48,000 --> 00:13:51,060 gonna be easy to solve it, but they were very good at providing feedback. 223 00:13:51,460 --> 00:13:56,500 And not only feedback on the user interface that allow us to, may make 224 00:13:56,500 --> 00:13:58,270 it more intuitive, but also they. 225 00:13:58,670 --> 00:14:01,190 made re implemented the recommendations. 226 00:14:01,190 --> 00:14:04,390 Our platform, was provided and saw the results. 227 00:14:04,390 --> 00:14:08,410 And that, that's what helped us calibrate the models to a point where 228 00:14:08,600 --> 00:14:12,610 we said, okay, we're ready to open the doors and, take in more clients and 229 00:14:12,940 --> 00:14:14,380 yeah, it's been fantastic since then. 230 00:14:15,270 --> 00:14:16,200 Super exciting. 231 00:14:16,740 --> 00:14:21,970 All right, let's talk about what actually is required for such a platform to work? 232 00:14:22,280 --> 00:14:26,390 I think you've already mentioned a few things, but let's just summarize them 233 00:14:26,390 --> 00:14:30,530 in, in, in one place to facilitate, the listening experience for our audience. 234 00:14:30,870 --> 00:14:34,470 What are the core features that you would say are necessary for this platform? 235 00:14:35,325 --> 00:14:37,825 Yeah, so we need access to all your data. 236 00:14:37,825 --> 00:14:41,125 So every platform these days has an API. 237 00:14:41,245 --> 00:14:46,415 So we will connect to your Google analytics, Facebook ads, email, so on, we 238 00:14:46,415 --> 00:14:50,735 have about 250 connectors at the moment, so we connect to pretty much anything. 239 00:14:51,045 --> 00:14:55,275 We create automated data pipelines that ingest all the data, put 240 00:14:55,275 --> 00:14:56,595 that into our data warehouse. 241 00:14:57,045 --> 00:15:02,495 We have a ton of different processes in place, clean the data, look for outliers 242 00:15:02,495 --> 00:15:11,045 issues and, the way we approach it is, we actually build one, attribution model 243 00:15:11,045 --> 00:15:12,935 for every client that we work with. 244 00:15:13,295 --> 00:15:16,715 So if you are selling furniture, you're not gonna get the same attribution as 245 00:15:16,715 --> 00:15:18,905 somebody selling accounting services. 246 00:15:18,905 --> 00:15:21,670 What we do is the first, we connect your data. 247 00:15:21,670 --> 00:15:23,470 We start building the data pipelines. 248 00:15:23,470 --> 00:15:28,770 But in parallel to that, we have three meetings, one hour each with our 249 00:15:28,770 --> 00:15:35,610 clients, and we focus on your customer segments, product mix and media mix. 250 00:15:35,730 --> 00:15:40,290 So we need to understand what are all the different personas that buy your product? 251 00:15:40,650 --> 00:15:42,150 What products do you actually sell? 252 00:15:43,060 --> 00:15:48,329 And what are all the different channels that you use online, offline channels? 253 00:15:48,710 --> 00:15:49,910 Where do you sell your products? 254 00:15:49,910 --> 00:15:50,960 Only on your website. 255 00:15:50,960 --> 00:15:54,120 Amazon, Walmart, Target, other retail stores. 256 00:15:54,900 --> 00:15:58,710 Are you doing, non trackable marketing activities? 257 00:15:58,710 --> 00:15:59,850 Are you going to trade shows? 258 00:15:59,850 --> 00:16:01,680 Are you appearing on podcasts? 259 00:16:01,760 --> 00:16:05,020 So we have to learn everything about your business. 260 00:16:05,020 --> 00:16:10,870 And then we, based on that, we create custom model that explains your business 261 00:16:10,900 --> 00:16:14,210 in particular and with all these elements that we already talked about. 262 00:16:14,580 --> 00:16:20,790 By how much does the lack of a pre-trained model delay the, how long does it take 263 00:16:20,790 --> 00:16:25,450 for a client who expresses, commits to this approach and then when they 264 00:16:25,450 --> 00:16:28,830 start, when they can see results, isn't the lack of a pre-trained model 265 00:16:28,890 --> 00:16:30,600 like lengthening this quite a bit? 266 00:16:31,348 --> 00:16:31,828 Yes. 267 00:16:31,888 --> 00:16:35,608 if you use a, an, outside the box solution, you could probably be up 268 00:16:35,608 --> 00:16:37,698 and running in four or five days. 269 00:16:38,098 --> 00:16:43,918 With us it takes six weeks and we believe that it's much more important to have 270 00:16:43,918 --> 00:16:46,618 accurate data than to have data quickly. 271 00:16:46,718 --> 00:16:51,203 Most of our clients manage hundreds of thousands of dollars, or millions 272 00:16:51,203 --> 00:16:55,793 of dollars and they make decisions of moving maybe 2 million from here to here. 273 00:16:56,313 --> 00:17:02,193 So they can wait a couple of weeks knowing that the insights are gonna be worth it. 274 00:17:02,193 --> 00:17:04,803 They wanna make sure that they make the right decisions. 275 00:17:05,303 --> 00:17:10,688 It doesn't take that long if, for example, if you think about the media mix modeling 276 00:17:10,758 --> 00:17:12,858 pre-internet, it used to take six months. 277 00:17:13,308 --> 00:17:18,348 And the problem is that not only are you looking at data from six months 278 00:17:18,348 --> 00:17:22,908 ago, but also like you don't get to make decisions for six months. 279 00:17:22,908 --> 00:17:28,148 So in this case, once it's built, it will start like ingesting data automatically. 280 00:17:28,148 --> 00:17:31,928 So yeah, it takes six weeks to, for the setup, but after that, 281 00:17:31,988 --> 00:17:33,683 it's gonna be up to date a hundred 282 00:17:33,983 --> 00:17:35,263 It's not, it's not a snapshot. 283 00:17:35,263 --> 00:17:36,613 it's learning as it goes. 284 00:17:36,663 --> 00:17:37,093 Absolutely. 285 00:17:37,263 --> 00:17:38,073 Every single day. 286 00:17:38,133 --> 00:17:41,428 So every day is learning and adjusting and getting better and better. 287 00:17:41,668 --> 00:17:41,908 Yeah. 288 00:17:42,383 --> 00:17:42,683 Amazing. 289 00:17:42,953 --> 00:17:46,923 And so what are the key challenges to in, in training models, 290 00:17:46,943 --> 00:17:48,353 models for each client like this? 291 00:17:48,968 --> 00:17:49,388 Yeah. 292 00:17:49,388 --> 00:17:51,639 I mean there's, there are a lot of challenges. 293 00:17:51,689 --> 00:17:56,644 We have a process, that it's, we, it's a, the model of validation process. 294 00:17:56,644 --> 00:18:01,584 So we build the model with all the assumptions that we had plus, 295 00:18:01,584 --> 00:18:03,684 conversations we have with our clients. 296 00:18:04,174 --> 00:18:07,264 We input all the data, run a couple of diagnostics. 297 00:18:07,924 --> 00:18:09,514 I would say about half of the time. 298 00:18:09,719 --> 00:18:13,169 It passes everything with great grades from the get go. 299 00:18:13,169 --> 00:18:17,389 But most of the time there's something there that we're not quite happy with. 300 00:18:17,389 --> 00:18:21,259 and usually what it is we forgot to include some factor. 301 00:18:21,339 --> 00:18:23,561 A very common one is promotion. 302 00:18:23,689 --> 00:18:25,639 So you did a black Friday promotion. 303 00:18:25,639 --> 00:18:27,889 We forgot to model that. 304 00:18:27,889 --> 00:18:31,574 So we see that something that was predicting to be here, it actually, 305 00:18:31,604 --> 00:18:33,094 it, ends up being much higher. 306 00:18:33,314 --> 00:18:33,614 Okay. 307 00:18:33,614 --> 00:18:36,644 That's, one promotion that we forgot to build into something. 308 00:18:37,034 --> 00:18:41,008 Or maybe, what is this bump around September last year? 309 00:18:41,034 --> 00:18:41,364 Oh yeah. 310 00:18:41,364 --> 00:18:43,224 We got, featured in this magazine. 311 00:18:43,224 --> 00:18:48,034 So it's getting a initial set of insights, talking to our clients about things that 312 00:18:48,034 --> 00:18:53,404 our model wasn't able to predict well, and make sure that we are accounting 313 00:18:53,404 --> 00:18:54,664 for those things moving forward. 314 00:18:55,554 --> 00:18:59,994 How does the artificial intelligence, the machine learning that, you've 315 00:18:59,994 --> 00:19:05,899 built into the platform now change the role of, the marketer, the strategist? 316 00:19:06,229 --> 00:19:09,689 Both of course on your side and your customers, in their organization. 317 00:19:10,039 --> 00:19:11,629 There are two key aspects of it. 318 00:19:11,679 --> 00:19:14,559 One is accuracy and the other one is access. 319 00:19:14,739 --> 00:19:18,819 You can have the best AI model in the world, like very pleasant to chat 320 00:19:18,879 --> 00:19:22,099 with, but if it has the wrong data, you're gonna make the wrong decisions. 321 00:19:22,099 --> 00:19:26,259 So first and foremost, we wanna make sure that our model is very good at 322 00:19:26,259 --> 00:19:30,189 capturing that underlying reality and the causal relationships in the data. 323 00:19:30,639 --> 00:19:34,276 So the key characteristic of a great model is that it should be able to 324 00:19:34,276 --> 00:19:38,276 predict a ton of different scenarios, even scenarios that he hasn't seen before. 325 00:19:38,776 --> 00:19:43,851 So maybe you spend a thousand dollars a day on, on Google, or $1,500 a day, 326 00:19:43,851 --> 00:19:45,621 but you've always been in that range. 327 00:19:46,101 --> 00:19:48,291 What would happen if you spend $5,000 a day? 328 00:19:48,771 --> 00:19:53,096 Like the model has never seen that, but it should still be able to predict 329 00:19:53,096 --> 00:19:56,126 with a high degree of accuracy, things that it has never seen before. 330 00:19:56,126 --> 00:19:59,286 And that's how we, he knows that it's, understanding the 331 00:19:59,286 --> 00:20:00,696 true impact of that channel. 332 00:20:01,206 --> 00:20:02,466 So accuracy is key. 333 00:20:02,536 --> 00:20:07,251 The quality of our decisions, the quality of our outcomes are directly 334 00:20:07,251 --> 00:20:10,651 related to the quality of the data that, aids those decisions. 335 00:20:11,341 --> 00:20:13,081 The second one is access. 336 00:20:13,081 --> 00:20:17,641 And what I mean by access is the average brand in the United States has 12 337 00:20:17,641 --> 00:20:23,701 different platforms, and that means you have to log to 12 different places and 338 00:20:23,701 --> 00:20:25,921 look at each one and start analyzing. 339 00:20:27,226 --> 00:20:30,376 And yeah, sure if you have infinite time, you could do it. 340 00:20:30,406 --> 00:20:33,316 It's really not a matter of, you're not smart enough, so 341 00:20:33,316 --> 00:20:34,696 you need AI to do it for you. 342 00:20:34,696 --> 00:20:39,386 But it is more of a matter of, you have usually somewhere between one 343 00:20:39,596 --> 00:20:41,216 and 10 million data points a day. 344 00:20:41,216 --> 00:20:46,296 There's no way that anybody can go through all that and look for patterns. 345 00:20:46,296 --> 00:20:52,741 So AI is very good at spotting problems, opportunities and helping you prioritize. 346 00:20:52,741 --> 00:20:56,096 In our case we have a checklist of 150 different things that it 347 00:20:56,096 --> 00:20:59,666 checks for, and there's all this conditional logic built into it. 348 00:20:59,666 --> 00:21:03,596 So it checks for this, and if this is a yes, then it goes, checks for this. 349 00:21:03,596 --> 00:21:05,456 But if it, this is a no checks for this. 350 00:21:05,456 --> 00:21:07,281 So it has all that built into it. 351 00:21:07,771 --> 00:21:12,631 Plus whatever you as a user add to the system, because you want things 352 00:21:12,691 --> 00:21:14,791 different than the guy next to you, right? 353 00:21:15,301 --> 00:21:21,451 So the value in that is that, not only does it analyze all that and 354 00:21:21,451 --> 00:21:25,321 filters out the noise and leaves you with a signal, is even if you had 355 00:21:25,321 --> 00:21:28,761 150, insights, that's not actionable. 356 00:21:28,761 --> 00:21:32,071 It is way too much for anybody to really, act on. 357 00:21:32,071 --> 00:21:38,121 So it has to help you understand what's not relevant and what are the three to 358 00:21:38,121 --> 00:21:42,611 five things that you need to be doing, paying attention to right now, that 359 00:21:42,611 --> 00:21:44,201 are gonna have the greatest impact. 360 00:21:44,721 --> 00:21:50,306 So if you go and ask a question, you don't have to learn how to use a user interface. 361 00:21:50,306 --> 00:21:53,696 Like you had to click here and then here, and where was that? 362 00:21:53,696 --> 00:21:56,756 Next time you forget, with AI you just ask a question and 363 00:21:56,756 --> 00:21:57,866 it just gets you the answer. 364 00:21:57,866 --> 00:22:01,976 So that's being able to access these 12 different, platforms and 365 00:22:01,976 --> 00:22:03,416 ask questions in plain English. 366 00:22:03,416 --> 00:22:06,036 Maybe you don't know what a certain metric is called. 367 00:22:06,036 --> 00:22:08,796 You call it revenue and the platform calls it sales. 368 00:22:09,126 --> 00:22:11,556 The AI understands that revenue is sales. 369 00:22:11,573 --> 00:22:13,523 So that context is critical. 370 00:22:13,681 --> 00:22:14,916 That's a big part of it. 371 00:22:14,916 --> 00:22:18,636 But yeah, the other part is even if you're not asking a question, having 372 00:22:18,636 --> 00:22:25,111 a system that can process data million of times faster than any human. 373 00:22:25,351 --> 00:22:31,271 And can then give you that information so you as a human who knows the 374 00:22:31,271 --> 00:22:35,731 business, has the context, knows about the strategy, can then, with that 375 00:22:35,731 --> 00:22:38,041 context, know what decision to make. 376 00:22:38,351 --> 00:22:41,781 I think that's, that's something that is, is probably the, one of 377 00:22:41,781 --> 00:22:45,051 the highest competitive advantages that you can have as a business. 378 00:22:45,511 --> 00:22:51,281 The core of the message there is that to compete, to be, successful, 379 00:22:51,581 --> 00:22:56,771 businesses must learn to buddy up, right? 380 00:22:56,861 --> 00:23:02,451 Their employees, their experts, I think initially you were talking about 381 00:23:02,451 --> 00:23:08,511 how human oversight of the models is critical because our ability to 382 00:23:08,511 --> 00:23:10,191 take things in context of course. 383 00:23:10,671 --> 00:23:16,641 Then you've also mentioned how there is both from an automation perspective when 384 00:23:16,641 --> 00:23:21,691 there's way too many platforms, but also way too many data just coming at you. 385 00:23:22,021 --> 00:23:26,331 And so it's important to learn to use these tools to augment your, 386 00:23:26,581 --> 00:23:28,621 analytical, essentially capability. 387 00:23:28,621 --> 00:23:29,941 You've mentioned that as well. 388 00:23:30,604 --> 00:23:35,244 And what's interesting about this is that, most business leaders or marketers are 389 00:23:35,244 --> 00:23:40,254 gonna agree with the fact that accurate data helps you make good decisions. 390 00:23:40,654 --> 00:23:43,454 And I would say accurate data that is readily available. 391 00:23:43,814 --> 00:23:47,104 But the problem is that it's not always easy to have 392 00:23:47,764 --> 00:23:50,044 accurate data or available data. 393 00:23:50,044 --> 00:23:53,734 But if we are thinking not even five years into the future, two 394 00:23:53,734 --> 00:23:56,774 years into the future, every competitor is gonna have that. 395 00:23:57,234 --> 00:24:01,144 And I think that's because we already believe that if I have the right 396 00:24:01,144 --> 00:24:04,114 information in front of me, I'll be able to make better decisions. 397 00:24:04,604 --> 00:24:08,814 Making sure that every person in your organization that needs to 398 00:24:08,814 --> 00:24:10,524 make decisions, has access to that. 399 00:24:10,524 --> 00:24:12,579 It's probably one of the best investments that you can make. 400 00:24:13,041 --> 00:24:15,036 Excellent point on which to finish. 401 00:24:15,336 --> 00:24:18,486 But before we do, I actually have a couple of questions I always ask. 402 00:24:18,486 --> 00:24:23,916 So the first one is, what is a book, a tool or a habit that has made 403 00:24:23,916 --> 00:24:27,276 a particularly significant impact on yourself in the last 12 months. 404 00:24:28,036 --> 00:24:32,008 I start each day, visualizing what I want the day to, to look 405 00:24:32,008 --> 00:24:33,358 like, and then planning it. 406 00:24:33,488 --> 00:24:38,128 So every call I, I have, I wanna make sure that I know what I 407 00:24:38,128 --> 00:24:39,448 want to get out of the call. 408 00:24:39,508 --> 00:24:43,768 That I make sure that the other person want gets what they want out of the call. 409 00:24:44,338 --> 00:24:48,008 Every minute that I spend working, I wanna make sure that I'm working on the 410 00:24:48,008 --> 00:24:50,318 most impactful thing at any given time. 411 00:24:50,798 --> 00:24:56,119 And before doing that, I was just coming to the office and things come 412 00:24:56,119 --> 00:24:57,589 at you that you're responding to. 413 00:24:57,619 --> 00:25:01,549 So therefore you're not never really taking the time to, to say, this is the 414 00:25:01,549 --> 00:25:03,379 most important thing I need to do today. 415 00:25:03,379 --> 00:25:08,229 So now I still leave about two hours of time open time for everybody 416 00:25:08,229 --> 00:25:09,699 else that needs something from me. 417 00:25:09,699 --> 00:25:13,629 But I make sure that I start the day by doing the thing that is gonna 418 00:25:13,629 --> 00:25:15,579 move the business forward most. 419 00:25:16,479 --> 00:25:19,794 That's a very good one, very deliberate and mindful of you. 420 00:25:20,304 --> 00:25:25,934 And the last question is, is there maybe one thing that you would expect 421 00:25:25,934 --> 00:25:29,054 to remain constant among all the other things that are changing in the 422 00:25:29,054 --> 00:25:31,094 moment, let's say 10 years from now? 423 00:25:31,394 --> 00:25:37,316 The, the human quality, and the importance it has in, in, in business. 424 00:25:37,316 --> 00:25:42,927 Sure, we are gonna have AI systems that have access to all the data that 425 00:25:43,076 --> 00:25:46,376 that we need access to all our tools that we'll be able to send you a 426 00:25:46,376 --> 00:25:51,006 notification, slack or email or, give you recommendations for, hey, like this 427 00:25:51,006 --> 00:25:52,536 happened, what should I do about it? 428 00:25:52,536 --> 00:25:53,046 And so on. 429 00:25:53,436 --> 00:25:57,821 But I think that, that brings a lot of concerns to people and I think that 430 00:25:58,211 --> 00:26:01,721 when we think about what machines are good at, the machines are not good at, 431 00:26:01,771 --> 00:26:06,111 there's a certain aspect of connecting with somebody else and being real and 432 00:26:06,111 --> 00:26:11,526 being, open about the challenges you have and what, the goals that you have 433 00:26:11,526 --> 00:26:14,166 and why, certain things are, complex. 434 00:26:14,166 --> 00:26:16,451 And I think that's never gonna go away. 435 00:26:16,541 --> 00:26:20,201 I, I hope that it never goes away, but, it's really at the core of who we are. 436 00:26:20,221 --> 00:26:25,771 So I think that's that's something that has always served me, to put relationships 437 00:26:25,771 --> 00:26:32,071 first, to think long term, to invest in people and that's not gonna go anywhere. 438 00:26:32,101 --> 00:26:35,001 We are human beings having a human experience. 439 00:26:35,001 --> 00:26:36,561 I don't expect that to ever change. 440 00:26:37,104 --> 00:26:38,064 Such a great answer. 441 00:26:38,364 --> 00:26:39,259 Thank you very much, Zeke. 442 00:26:40,464 --> 00:26:40,969 Thank you, Alex. 443 00:26:41,828 --> 00:26:44,828 Zeke shared two big ways AI is changing marketing. 444 00:26:45,218 --> 00:26:48,458 First by making data more accurate so companies can see 445 00:26:48,548 --> 00:26:49,988 what actually drives sales. 446 00:26:50,378 --> 00:26:55,748 Second, by improving access, turning complex data into simple, useful insights. 447 00:26:56,138 --> 00:26:59,768 But as Zeke pointed out, AI isn't here to replace marketeers. 448 00:26:59,858 --> 00:27:02,458 It's here to help them make better decisions. 449 00:27:03,088 --> 00:27:04,318 If you want to dive deeper. 450 00:27:04,408 --> 00:27:08,278 Zeke's team has a free resource, the Data Speaks Playbook, which 451 00:27:08,278 --> 00:27:12,808 lays out 20 years of best practices for using data to drive growth. 452 00:27:12,928 --> 00:27:15,688 You can download it at dataspeaks.ai. 453 00:27:16,288 --> 00:27:20,638 As always, we'll have more exciting topics and guest appearances lined up, so 454 00:27:20,638 --> 00:27:25,378 stay tuned for more tales of innovation that inspire, challenge and transform. 455 00:27:25,978 --> 00:27:27,723 Until next time, peace. 456 00:27:28,934 --> 00:27:31,274 Thanks for tuning in to Innovation Tales. 457 00:27:36,014 --> 00:27:39,704 Get inspired, connect with other practitioners and approach the 458 00:27:39,704 --> 00:27:41,744 digital revolution with confidence. 459 00:27:42,674 --> 00:27:46,064 Visit innovation-tales.com for more episodes. 460 00:27:47,404 --> 00:27:48,424 See you next time.