1169: Why Some AI Initiatives Succeed (& Others Fail) w/ Ankur Goyal

WEBVTT 1 00:00:00.040 --> 00:00:04.799 Wouldn't it be nice to have several fault leaders in your industry know and Love 2 00:00:05.000 --> 00:00:10.230 Your brand? Start a podcast, invite your industries thought leaders to be guests 3 00:00:10.349 --> 00:00:15.429 on your show and start reaping the benefits of having a network full of industry 4 00:00:15.429 --> 00:00:25.699 influencers. Learn more at sweet phish MEDIACOM. You're listening to be tob growth, 5 00:00:26.100 --> 00:00:30.500 a daily podcast for B TOB leaders. We've interviewed names you've probably heard 6 00:00:30.539 --> 00:00:34.420 before, like Gary Vander truck and Simon Senek, but you've probably never heard 7 00:00:34.500 --> 00:00:38.729 from the majority of our guests. That's because the bulk of our interviews aren't 8 00:00:38.770 --> 00:00:43.289 with professional speakers and authors. Most of our guests are in the trenches leading 9 00:00:43.329 --> 00:00:48.170 sales and marketing teams. They're implementing strategy, they're experimenting with tactics, they're 10 00:00:48.250 --> 00:00:52.759 building the fastest growing BTB companies in the world. My name is James Carberry. 11 00:00:52.799 --> 00:00:56.320 I'm the founder of sweet fish media, a podcast agency for BB brands, 12 00:00:56.439 --> 00:00:59.640 and I'm also one of the cohosts of this show. When we're not 13 00:00:59.719 --> 00:01:03.600 interviewing sales and marketing leaders, you'll hear stories from behind the scenes of our 14 00:01:03.640 --> 00:01:07.510 own business. Will share the ups and downs of our journey as we attend 15 00:01:07.590 --> 00:01:11.590 to take over the world. Just getting well? Maybe let's get into the 16 00:01:11.670 --> 00:01:21.700 show. Welcome back to BTB growth. I am your host for today's episode. 17 00:01:21.819 --> 00:01:26.060 Nikki Ivy was sweet fish media guys I've got with me today on core, 18 00:01:26.180 --> 00:01:30.099 Goo Yall, who is the founder and CEO of empyra. On core. 19 00:01:30.180 --> 00:01:32.780 How you do when today? I'm doing really well. Thanks so much 20 00:01:32.780 --> 00:01:34.489 for having me. Thank you for coming, guys. He really is doing 21 00:01:34.609 --> 00:01:38.890 really well. Like I'm looking at his face and he's smiling really big and 22 00:01:38.969 --> 00:01:45.290 there's this positive energy that cannot be contained by my computer screen. So strap 23 00:01:45.329 --> 00:01:51.239 yourselves in. So we're going to be talking today about why some AI initiatives 24 00:01:51.319 --> 00:01:55.799 succeed and others fail, and and uncle is going to give us his his 25 00:01:55.959 --> 00:01:57.640 insights there, and I can't wait to dig into that. But before we 26 00:01:57.879 --> 00:02:01.069 do on core, I would love it if you would just give us all 27 00:02:01.109 --> 00:02:05.629 a little bit of background on yourself for what you and the folks that impyre 28 00:02:05.670 --> 00:02:09.189 are up to these days. Absolutely so. I'm the founder and CEO of 29 00:02:09.349 --> 00:02:15.979 IMPYRA and at Impyra we're building software that uses AI to provide business intelligence for 30 00:02:16.139 --> 00:02:21.500 visual information like photos, documents, and more. And you know, in 31 00:02:21.580 --> 00:02:28.259 a nutshell, as our world becomes increasingly digital, organizations are recognizing the importance 32 00:02:28.340 --> 00:02:34.210 of visual information, like you know, images, videos, documents, sound, 33 00:02:34.650 --> 00:02:38.169 etc. But prior to Impyra, actually managing the stuff was a very 34 00:02:38.289 --> 00:02:45.159 burninsome task, and our product uses ai to not only help organizations manage visual 35 00:02:45.240 --> 00:02:49.680 data but also extract intelligence out of it and impact the top line for their 36 00:02:49.719 --> 00:02:53.800 business. Before Impyra, I was an early engineer and VP of engineering at 37 00:02:54.280 --> 00:03:00.030 men sequel, and there we actually develop of technology to help customers do similar 38 00:03:00.150 --> 00:03:04.349 things, but with structure data, and structure data is data that you know, 39 00:03:04.430 --> 00:03:07.830 fits into a spreadsheet or a traditional database. And as you and you 40 00:03:07.909 --> 00:03:13.259 know the listeners, no, getting photos and documents into a spreadsheet just doesn't 41 00:03:13.259 --> 00:03:16.939 work. And that's why I started in here. I I love it. 42 00:03:17.020 --> 00:03:21.379 So you guys are you're doing the work that sort of keeps our our daytoday 43 00:03:21.419 --> 00:03:24.620 going right, as some of the some of the behind the scenes necessary stuff 44 00:03:24.659 --> 00:03:31.330 that folks who are not good at any of that stuff neat and benefit from. 45 00:03:31.610 --> 00:03:36.889 So thanks for God's work out here. Before we really get into it, 46 00:03:37.009 --> 00:03:38.650 I kind of just want to get at a high level, talk about 47 00:03:39.129 --> 00:03:45.240 you know, given the height around Ai these days, what are you, 48 00:03:45.360 --> 00:03:47.719 know, some of your high level recommendations for people who are trying to make 49 00:03:49.240 --> 00:03:51.719 sense of it all? And how do we how we know, like, 50 00:03:51.840 --> 00:03:54.990 what's real versus was high? Absolutely I think part of the reason for the 51 00:03:55.069 --> 00:04:00.789 hype is that a I actually has a massive impact in pretty much everyone's daytoday 52 00:04:00.870 --> 00:04:04.909 life. I mean we all use software like Google. Maybe a little bit 53 00:04:04.949 --> 00:04:13.340 less facebook nowadays, but instagram, interest Google maps, even apples software nowadays, 54 00:04:13.780 --> 00:04:17.180 and you know, our experiences with the software are heavily powered by AI. 55 00:04:17.420 --> 00:04:20.699 Every time you click on a search result in Google, it learns a 56 00:04:20.740 --> 00:04:25.490 little bit about your preferences and it learns a little bit about itself. But 57 00:04:25.649 --> 00:04:29.930 the reality is this software just hasn't made its way into our work life. 58 00:04:30.370 --> 00:04:33.730 One of the most common things that we hear from our customers is that it's 59 00:04:33.769 --> 00:04:38.839 really easy to figure out who is in a photo on facebook, but if 60 00:04:38.879 --> 00:04:42.480 they take a bunch of photos at an event, for example, it's impossible 61 00:04:42.560 --> 00:04:46.199 to actually figure out, you know, all the pictures of a particular person 62 00:04:46.519 --> 00:04:49.480 or even label you know who's in what picture, and so I think that's 63 00:04:49.959 --> 00:04:54.430 where the the issue is with hype. In reality, in the consumer world 64 00:04:54.470 --> 00:04:58.509 it's present everywhere and it's been really successful, but in the enterprise world it 65 00:04:58.629 --> 00:05:01.189 just it hasn't made its where. It's way there yet, and what we 66 00:05:01.310 --> 00:05:06.620 see is that this is primarily due to the way that the software is delivered, 67 00:05:08.100 --> 00:05:11.660 and I'm happy to talk about that and a little bit of detail. 68 00:05:12.060 --> 00:05:15.100 If you think about Google, for example, they have lots and lots of 69 00:05:15.139 --> 00:05:19.060 data about the entire Internet on every single time you search for something, they 70 00:05:19.100 --> 00:05:23.529 learn a little bit about you as well, and that's really great, because 71 00:05:23.569 --> 00:05:28.569 the user interface is able to capture your energy and your attention and learn from 72 00:05:28.610 --> 00:05:30.610 you as a result. You know. On the other hand, the software 73 00:05:30.689 --> 00:05:35.920 that enterprises tend to consume is usually driven by some kind of API, AP 74 00:05:36.160 --> 00:05:42.160 onions, application programming interface, and when code is talking to you know, 75 00:05:42.240 --> 00:05:46.160 another company's Code, not only do you lose out on the ability to learn 76 00:05:46.639 --> 00:05:50.870 from all of the data that exists in an enterprise, but you actually lose 77 00:05:51.550 --> 00:05:56.350 on the API side how that data is used and so nick be as an 78 00:05:56.430 --> 00:06:00.149 example, if you are an API and I send you one picture from a 79 00:06:00.269 --> 00:06:02.709 party, it's going to be really hard for you to guess who's in that 80 00:06:02.829 --> 00:06:08.060 picture without looking at, you know, maybe all the pictures at the Party 81 00:06:08.180 --> 00:06:12.620 and what the guest list was and maybe some prior information about people. And 82 00:06:12.939 --> 00:06:15.259 even if you guess and you tell me the answer and I maybe show it 83 00:06:15.420 --> 00:06:18.610 to someone inside of my company, if you're an API, I'm not going 84 00:06:18.649 --> 00:06:20.730 to tell you whether or not you got it right or wrong, and so 85 00:06:20.850 --> 00:06:25.129 you really don't have an opportunity to learn. And that's kind of where we 86 00:06:25.250 --> 00:06:30.370 see the big difference between enterprise and consumer right, which ends up being the 87 00:06:30.490 --> 00:06:34.639 difference between the real and the high right exactly. The high team tends to 88 00:06:34.680 --> 00:06:40.079 be what we engage with as consumers and the real tends to be what we 89 00:06:40.279 --> 00:06:44.839 stand to be able to benefit from if we apply it as enterprises. I 90 00:06:44.920 --> 00:06:47.629 like the way that you that you laid that out. So once we get 91 00:06:47.670 --> 00:06:51.550 to this, what we understand the difference between, we're able to discern the 92 00:06:51.750 --> 00:06:56.509 real from from the hipe. Can you give us some examples of what as 93 00:06:56.589 --> 00:07:02.620 successful AI initiative looks like and how you know that? Absolutely so. I 94 00:07:02.740 --> 00:07:08.819 think the number one most important thing is to have a feedback loop, and 95 00:07:08.939 --> 00:07:14.379 a feedback loop is one where the software is able to learn every single time 96 00:07:14.540 --> 00:07:18.689 someone engages with it about whether or not it made a valid or invalid guess, 97 00:07:18.810 --> 00:07:23.290 and I think there's some actually really good examples of this that are starting 98 00:07:23.290 --> 00:07:29.329 to pop up in the enterprise. One company that has been really successful in 99 00:07:29.410 --> 00:07:33.120 a totally different space as a company called Demisto, which is a security product 100 00:07:33.519 --> 00:07:41.839 that surfaces potential security threats to an internal security team and actually from how the 101 00:07:41.959 --> 00:07:48.350 security team learns and interacts with those anomalies, Demisto software is able to learn 102 00:07:48.670 --> 00:07:53.269 from those as well and therefore the next time it's able to provide a better 103 00:07:53.310 --> 00:07:56.910 recommendation. It's kind of like, you know, Pandora provides a better music 104 00:07:56.990 --> 00:08:01.220 recommendation every time you give a thumbs up or thumbs down. Yeah, yeah, 105 00:08:01.459 --> 00:08:05.779 and I won't get into why I'm still soil the Pandora that kids call 106 00:08:05.860 --> 00:08:09.819 me old because they're like when Uspni Fu and like whatever, just let me 107 00:08:09.899 --> 00:08:13.529 do my thing. They give me the same hard time about facebook. Actually, 108 00:08:13.529 --> 00:08:16.329 they're like only old people on facebook and I'm like whatever anyway. So 109 00:08:18.050 --> 00:08:22.689 so, yes, no, you're absolutely right that those are the sort of 110 00:08:22.730 --> 00:08:28.199 framework implications of what we what we still stand to see as far as the 111 00:08:28.319 --> 00:08:31.759 future of what enterprises are going to be doing with Ai, and it's so 112 00:08:31.879 --> 00:08:35.559 that it's no wonder you're so excited about it. There's it's it's the frontier 113 00:08:35.639 --> 00:08:39.440 out here. There's so much yet to happen. It is an exciting time, 114 00:08:39.519 --> 00:08:43.710 as time to be alive right. And so once we've gotten to this 115 00:08:43.789 --> 00:08:50.590 point where we're separating real from hype, we can recognize what a successful initiative 116 00:08:50.710 --> 00:08:52.950 is. What are what are some of your thoughts on some of the reasons 117 00:08:54.429 --> 00:09:01.620 why some initiatives succeed and and and some fail. And you know, are 118 00:09:01.740 --> 00:09:05.980 those sort of the things that are true for for enterprise and direct consumer companies? 119 00:09:07.820 --> 00:09:11.169 Absolutely, I think you know, broadly speaking, in enterprise software they're 120 00:09:11.690 --> 00:09:16.970 two kinds of initiatives that can succeed. One type of initiative is where you 121 00:09:16.049 --> 00:09:20.049 cut the bottom line and another type of initiative is where you increase the top 122 00:09:20.210 --> 00:09:26.679 line, and in both cases I think having an Roi understanding with your customer 123 00:09:26.799 --> 00:09:31.200 from the very beginning and working together actually on measuring things is the key to 124 00:09:31.320 --> 00:09:35.000 figuring out what's going to be successful and what's not. So I'll give you 125 00:09:35.039 --> 00:09:39.549 an example of each one as it relates to the bottom line, one of 126 00:09:39.590 --> 00:09:43.950 the most common things that software like hours is able to help companies with is 127 00:09:43.149 --> 00:09:50.269 data entry. So there are a lot of tasks that extremely creative and intelligent 128 00:09:50.389 --> 00:09:54.220 people inside of companies are kind of bogs down with, which are, you 129 00:09:54.299 --> 00:09:58.139 know, looking at images or pdf and literally retyping what they see into a 130 00:09:58.220 --> 00:10:03.220 screen. And actually most companies that we work with have a pretty good sense 131 00:10:03.259 --> 00:10:07.809 of how much time gets spent engaging in these tasks and what we're able to 132 00:10:07.889 --> 00:10:13.049 do is work with those companies and work with the teams inside of these companies 133 00:10:13.330 --> 00:10:18.330 to understand one you know, is it even feasible for artificial intelligence to be 134 00:10:18.370 --> 00:10:22.279 a good solution to this problem? The best question to ask when someone you 135 00:10:22.399 --> 00:10:26.000 know wonders can ai help, is how would a human do this? And 136 00:10:26.120 --> 00:10:30.759 if you can't piece together a good story for how a human would do it, 137 00:10:30.840 --> 00:10:33.720 it's actually going to be really hard for an algorithm too as well. 138 00:10:33.080 --> 00:10:37.870 If the answer is a spreadsheet, then maybe ai is the answer, uh, 139 00:10:39.309 --> 00:10:41.429 because it be hard for a human to do it. No, no, 140 00:10:41.750 --> 00:10:43.350 that's a great point. I mean one of my favorite examples of something 141 00:10:43.389 --> 00:10:48.309 that's hard for an algorithm to do is to look at a picture and identify 142 00:10:48.470 --> 00:10:52.539 what clothes or what skew numbers someone is wearing. So that's something that we 143 00:10:52.700 --> 00:10:54.940 hear a lot about that. Actually it's something we help our customers with quite 144 00:10:54.940 --> 00:11:00.100 a bit, especially customers, and you know, retail and consumer package goods. 145 00:11:00.539 --> 00:11:03.220 And you know, if you took a picture of me right now wearing 146 00:11:03.299 --> 00:11:07.809 this Patagonia and you asked me to look at it and say what Patagonia sweatshirt 147 00:11:07.850 --> 00:11:11.049 is this, you know it would be really challenging for me to just look 148 00:11:11.090 --> 00:11:15.409 at the picture and figure that out and similarly it's really, really challenging to 149 00:11:15.450 --> 00:11:20.080 train and Algorithm to actually solve that problem. On the other hand, if 150 00:11:20.240 --> 00:11:24.159 you took a picture of me and I had access to all of the photos 151 00:11:24.159 --> 00:11:28.320 you were shooting that day and what skews you were shooting that day and a 152 00:11:28.480 --> 00:11:33.110 bunch of the surrounding maybe the the label which has some information about the product, 153 00:11:33.149 --> 00:11:37.710 all of these things are lying around a studio when you're shooting pictures of 154 00:11:39.429 --> 00:11:43.870 Patagonia sweatshirts, and actually all of these clues can be fed into an artificial 155 00:11:43.909 --> 00:11:48.100 intelligence algorithm to figure out things, you know, much, much, much 156 00:11:48.100 --> 00:11:52.460 more accurately, and so that's kind of the process we engage in, especially 157 00:11:52.500 --> 00:11:56.539 related to data, and try to figure out, you know, whether this 158 00:11:56.700 --> 00:12:00.460 is a feasible problem to solve in the first place and be if we partner 159 00:12:00.659 --> 00:12:03.009 with our customer and solve that, how much time it's actually going to save 160 00:12:03.450 --> 00:12:09.090 their you know, highly creative teams to work on much more meaningful things. 161 00:12:11.570 --> 00:12:16.600 Imagine it, a spreadsheet filled with rows and rows of your sales enablement assets. 162 00:12:16.879 --> 00:12:20.320 You've devoted two years organizing this masterpiece, only for it to stop making 163 00:12:20.360 --> 00:12:26.039 sense. This was Chad tribuccos reality. As the head of sales enablement at 164 00:12:26.120 --> 00:12:30.789 glint, a linkedin company, he's responsible for instilling confidence in his sales reps 165 00:12:30.830 --> 00:12:33.669 and arming them with the information they need to do their jobs. However, 166 00:12:33.870 --> 00:12:39.549 when his glorious spreadsheet became too complex, he realized he needed a new system. 167 00:12:39.149 --> 00:12:43.509 That's when Chad turned to guru. With Guru, the knowledge you need 168 00:12:43.549 --> 00:12:48.340 to do your job finds you. Between Guru's Web interface, slack integration, 169 00:12:48.740 --> 00:12:54.419 mobile APP and browser extension. Teams can easily search for verified knowledge without leaving 170 00:12:54.460 --> 00:12:58.330 their workflow. No more siload or staled information. Guru acts as your single 171 00:12:58.450 --> 00:13:03.370 source of truth. For Chad, this meant glent sales reps were left feeling 172 00:13:03.529 --> 00:13:07.610 more confident doing their jobs. See why leading companies like glint, shopify, 173 00:13:09.009 --> 00:13:13.639 spotify, slack and more are using guru for their knowledge management needs. Visit 174 00:13:13.840 --> 00:13:20.320 BB growth dot get gurucom to start your thirty day free trial and discover how 175 00:13:20.480 --> 00:13:28.549 knowledge management can empower your revenue teams. Give us your your top line tips 176 00:13:28.669 --> 00:13:33.870 for successfully implementing Ai, just for the folks who are maybe just starting out 177 00:13:33.870 --> 00:13:35.950 in this and want to know how to how to start off on the right 178 00:13:35.990 --> 00:13:41.269 foot. Absolutely, I think from the perspective of the vendor, it's very, 179 00:13:41.350 --> 00:13:45.299 very important that you have at least one person on your team who understands 180 00:13:45.620 --> 00:13:48.860 the technology, and when I say that I mean someone with a background in 181 00:13:48.059 --> 00:13:52.820 statistics or machine learning who can help you figure out whether this problem that you're 182 00:13:52.860 --> 00:13:56.210 going to help a customer with is something that actually could be solved with artificial 183 00:13:56.250 --> 00:14:01.490 intelligence in the first place. The second thing, from the customers perspective, 184 00:14:01.929 --> 00:14:05.889 is to make sure that you have a really good understanding of the process that 185 00:14:05.049 --> 00:14:09.320 goes into accomplishing a certain task, and that means what people actually have to 186 00:14:09.360 --> 00:14:13.559 do and how much time they spend on each task to actually get something from 187 00:14:13.600 --> 00:14:18.039 start to finish, and I think if you build out that map of the 188 00:14:18.120 --> 00:14:22.720 process and you partner with a company that has, you know, people that 189 00:14:22.919 --> 00:14:30.389 understand statistics, then it's a really straightforward process to understand how artificial intelligence technology 190 00:14:30.429 --> 00:14:33.990 can go in and solve your problem. And I think that not only is 191 00:14:35.070 --> 00:14:39.379 there a lot of off the shelf artificial intelligence software that can that can solve 192 00:14:39.460 --> 00:14:43.019 this problem, there's a very vibrant ecosystem of startups, including but not limited 193 00:14:43.059 --> 00:14:48.340 to Impura, that have different strengths and can help you solve different kinds of 194 00:14:48.379 --> 00:14:50.899 problems. I love it absolutely. Thank you for laying that out. I 195 00:14:52.019 --> 00:14:54.769 think it's going to be super valuable for those listeners of ours who are wanting 196 00:14:54.809 --> 00:14:58.929 to dip their toe in the water the right way when it comes to an 197 00:14:58.970 --> 00:15:03.970 AI initiative. But now that I've successfully picked your brain and seeing what I 198 00:15:03.009 --> 00:15:05.769 can get out of it, is time for you to tell us about what 199 00:15:05.809 --> 00:15:09.559 you're putting in it. So on, coore, tell us about a learning 200 00:15:09.639 --> 00:15:13.919 resource that you have been engaging with that is informing your approach. This just 201 00:15:13.960 --> 00:15:18.919 got you excited these days? Absolutely. I think the biggest thing I've learned 202 00:15:18.440 --> 00:15:24.710 since starting the company is the importance of design in enterprise software nowadays. To 203 00:15:24.789 --> 00:15:28.470 be able to get good user feedback and make your model work really well, 204 00:15:28.870 --> 00:15:33.309 you have to build a user experience that really, really engages users. And 205 00:15:33.389 --> 00:15:37.340 actually, if you ask our customers why they bought our software, they don't 206 00:15:37.379 --> 00:15:41.419 say ai or database, you know this for that. They say user experience. 207 00:15:41.740 --> 00:15:45.740 So for me I'm spending a lot of time learning as much as I 208 00:15:45.779 --> 00:15:48.500 can about design the user experience. We have a great team here at Impia, 209 00:15:50.009 --> 00:15:52.889 but it's something that's just really important and there's a really great book called 210 00:15:52.929 --> 00:15:58.009 Sprint, how to solve big problems and test new ideas in just five days. 211 00:15:58.210 --> 00:16:02.970 It was written by a few folks at Google ventures and it was recommended 212 00:16:03.009 --> 00:16:07.879 to me by mercy, who was the director of Product at slack and ran 213 00:16:07.960 --> 00:16:11.120 their growth team, you know, while ago, and actually use that framework 214 00:16:11.360 --> 00:16:14.480 to help slack in the early days. I love it. I love when 215 00:16:14.519 --> 00:16:18.159 we get the book recommendation that I haven't heard of yet that I can swa 216 00:16:18.320 --> 00:16:22.470 out and add to my audible library. It's amazing. I can't wait. 217 00:16:22.549 --> 00:16:26.629 I know that that just like me. Ever right listening has become fast fans 218 00:16:26.669 --> 00:16:29.070 of yours and they're going to want to know how to keep up with you. 219 00:16:29.230 --> 00:16:32.830 So tell us how people can connect with you. Definitely. So we're 220 00:16:33.259 --> 00:16:37.940 on twitter. You can reach out to us at IMPYRA HQ, and you 221 00:16:37.980 --> 00:16:41.019 can also reach out to us over email. Feel free to reach out to 222 00:16:41.220 --> 00:16:45.740 info at Impyracom, or you can reach out to me directly at my first 223 00:16:45.779 --> 00:16:49.129 name, a N K. You are at Impyracom. I'm also on, 224 00:16:49.330 --> 00:16:53.529 you know, Linkedin and twitter at a N K R Gil. So happy 225 00:16:53.570 --> 00:16:57.090 to connect with what anyone who's interested. Good stuff. Thank you so much 226 00:16:57.690 --> 00:17:03.799 this enthusiasm that you brought to this conversation and the expertise. I'm really likewise 227 00:17:03.920 --> 00:17:07.559 for and we'll have to have you on again because I got a lot more 228 00:17:07.559 --> 00:17:10.759 questions. This is something I know nothing about and you, like I said, 229 00:17:10.799 --> 00:17:14.960 you made it nice and understandable from me. So until then, thank 230 00:17:15.000 --> 00:17:17.910 you so much. Have a good one. I for thank you. It's 231 00:17:17.910 --> 00:17:22.869 been a lot of fun. We totally get it. We published a ton 232 00:17:22.950 --> 00:17:26.230 of content on this podcast and it can be a lot to keep up with. 233 00:17:26.750 --> 00:17:30.859 That's where we've started. The BB growth big three a no fluff email 234 00:17:30.940 --> 00:17:34.660 that boils down our three biggest takeaways from an entire week of episodes. Sign 235 00:17:34.740 --> 00:17:41.859 up today at Sweet Fish Mediacom big three. That sweet PHISH MEDIACOM Big Three