In this episode, we take a look at how AI turns your sharing into business growth.

Grant
Okay, welcome, everybody to another episode of ClickAI radio. So I'm very excited to have here with me today ShareThis business development leader. I think it got that right. Michael Gorman, business development leader. But before I go any further, Michael, would you introduce yourself?

Michael
You bet. Grant. Yeah, it's great to be here. Like you said, I oversee business development, but also product and marketing at ShareThis. I've been I've been with ShareThis for a couple of years. In that role. I have a background in data, really, data and analytics has been my passion. Also media and marketing sort of themes. I've worked for big data companies like Axiom, I've worked for an email marketing leader, digital impact, they got bought by Axiom. That's how I got there. And I've also worked for big consulting firms. And for ESPN back in the earlier days of my career.

Grant
Oh, wow. Could you maybe give us a play by play? I bet you could write ESPN. Interesting. Wow.

Michaels
It was a fun period. I was like years eight through 11 of the history of of ESPN, which, so is a fun time to be there.

Grant
How fun. All right. And he did some some consulting roles as well. So data and analytics, huh? Yeah. Right. All throughout all throughout the career. So what led you into this work was ShareThis what was it was the journey there?

Michael
Well, one thing is that, that I've worked with our CO CEO on the past, at axiom, so we knew each other, but ShareThis is a really, really special data asset. In a lot of ways, and within the world of the of the advertising that I've worked in for quite a few years. It's it was well known. So when I had an opportunity to do a little consulting for them, I jumped into it. And that led to the to the role. It's a Yeah, sure this is, you know, it's Well, shall I tell you a bit about the company? Or is that?

Grant
Yeah, yeah. I mean, yeah, tell me a little bit about how it got started. And its purpose and sort of the vision of it.

Michael
You know, well, like a lot of companies, it started with one purpose and, and things evolved a little bit over time, it, it started off in the early days of social networks, when Facebook was still a new idea and mind MySpace was, was beginning to slow down, it was with the idea of making it easy for any website to make to make it easy for their users to share content to all the social networks that they might have an interest in. And so a developer with a simple, taking, you know, taking a piece of code and pasting it on their website that they could then have sharing. We and so it was one of two or three tools that really started in those early days and became a leader in the space. We actually have a how to still maintain a trademark on that little little V on the site there. Yeah, I mean, that's what you're known for. Yeah. So it's a sign if that's there, it's a sign that sharing is you know, sharing tools are present. It's essentially the balance value for the for the publisher for the owner of the site who doesn't have to does no work to have sharing available will get some analytics as a result, sharing is valuable because it makes it attracts more people to the site new users more more content. And, and so it's it's grown up naturally. And we're, you know, so really well established. But a number of business models were tried over the years, but but about five years ago, we started focusing, moving towards being 100% about our data, is that really as a special asset, we have around 3 million publishers using us sort of our live arm 3 million now, that's been pretty stable, you know, half to three quarters and in the rest of the world, a quarter in the United States, a little biased towards English language, but we have every language in the world represented among the users on the sites. And, and so that data and and we'll talk more about this when we get into things like, you know, the the technology in the AI. Yeah, but we're really just, you know, it's like a window into what, what people are what's on people's minds? What are they looking for? What are they searching about online, and we can, you know, discern trends and also, you know, make sure that advertising is more relevant for for users.

Grant
So I have a question for you on that. So you've, of course, are familiar with the terminology of neuromarketing, right. And, you know, as a way of sort of tracking, how are people interacting with a site, right, and where do they go? And where do they point and click and, you know, there's organizations that look at, you know, extracting what the user is doing on the site, this feels like this starts to come into that world right that day. I mean, I don't know that it's tracking every single movement, but it's tracking, obviously, the event of I want to share something. Any thoughts on that?

Michael
Yeah, that's really interesting. I mean, there's a lot of different ways to make inferences about about people, we tend to focus a bit more on the on the broad, the broader picture, that the thing that's that, I mean, there's, like you say, so many choices. But the thing about online content is, it's very rich. So when a person visits a site, there's a lot of things there, there's a lot of things on the page they're looking at. And so what we've really focused on is using the page as a source of clues about what a person is interested in, we also might look at the link in and out of the page, and get a clue from, say, a search term as well, that's a that's useful, and clearly when someone shares, you know, content that's that that sort of zoned in on exactly what they care about on the page. But we've opted more for the broad picture of focusing, you know, taking all that richness and attributing some probability of interest that for you, for user to the things that are on the page. And that way we can we have just such a broad, you know, broad palette to work with. And I think also from the point of view of, of, you know, user consent and user experience, it means that what we're actually collecting is is relatively light, it's just that this user was on this page at this time. And any inference we make is not based on what he or she did, or how are their eye movement, there's no no personal collection, we just have the that event, and we get all the all the power.

Grant
So it's when they were there. Is it anything about how they got there? Or where are they left?

Michael
Yeah, exactly. We do. We do use the inbound links and outbound links when we can get them. And that sometimes, as I said, yields a search term, those can that was sort of part of the of the link the part of the information that what came with the user, you know, the referring search term or so that so there's some some useful data there as well.

Grant
Yeah. So so when you collect this, and then that's got to be a massive repository, I think I saw somewhere else and I'm looking at, was it three terabytes of raw data and 100 million keywords in 200 languages a day? Is that right?

Michael
It sounds roughly right. I haven't counted it lately. But, yeah, you're right. But But yeah, we we see about half a billion, you know, unique, what we call events, something, you know, something happened at a point in time, visits a share per day.

Grant
This is a grounds for, you know, a playing field for AI, right, just you have so much data. So tell me what it is you learn from it with the AI, right? What kinds of problems are you looking to solve? As you and I know, when we pursue AI, we, it would tend to be better served if we're going after a particular question or thought in mind. Now, obviously, we get surprised with AHA insights from Ai. But going intentionally after something makes a lot of sense. Can you give a scenario the kinds of things that you're looking for?

Michael
Well, the I would say that the theme that has worked for us so far, is to try to do is to focus on being the able to represent and reflect human interest, what are people interested in? And yeah, and so. So we, we use, and I guess where the AI comes in is that we use the latest techniques of language analysis and language modeling. So we capture all of the linguistic content on the page and then we represent it in a number of ways. What are all the prominent keywords? What are the what are the entities that are you know more that are Unusual, you know, a brand name, a celebrity name, a business name? What are the what is this page about the concept? Or what are? What are some of the concepts that accurately describe what this page is about. And then we have some standard categorization techniques are basically a taxonomy of topic interest topics that we we screen for, you know, and and it's not, it's not a yes, one of the nice things about this is it's not a, a, it's not a, it, we don't have to decide one thing, you know, we were able to say, all of the prominent keywords, and all of the interesting entities and several concepts and all the categories that this page is about. So it could be a page, it's about, you know, mountain climbing and and what shall we say? And Utah, and the, or the American West and, and road vehicles? And, you know, and beverages, you know, skiing or whatever? Right, right. Exactly.

Grant
Yeah, so some form of an ontology there, right, that allows you to sort of connect these together?

Michael
Yeah, we used a number of techniques that you said, One is, we built a custom ontology, using relative and you know, we're, we're not a huge company. So we, we try to wherever we can do something open source or free as the entry point we do that. And so we, we use some Wikipedia, it's slash DBPedia is a source for us. And, as is some Google free offerings that help us sort of the provide the raw material for building our customer ontology. We've also take great advantage of some of the latest open source language modeling tools. One is when it goes by the name of the Google released one, I forget what the what the acronym stands for, but one that's called Bert, and then more recently, one that's called Muse. Yeah, we use muse. Okay, that, that allows us to represent anything, either a word or a sentence, or the whole page as a as a set as a vector of 500 numbers. And if two pages have the same values for those 500 vectors, then they are about the same thing. Yeah, you got you have some affinity there right now, even though in practice, they might be in different languages use totally different, you know, different sets of words, but they're still about the same thing. That's, that's, that's really, for us that technology has been a real breakthrough. Because it's we've been sometimes keywords and can be very, you know, they can be false positives or No, yeah, negative.

Grant
I mean, there, yeah, there's nothing that governs some, you know, webpage designer to, you know, say, hey, are they using the actual right keywords? Right?

Michael
Yes, or even? Or even? How do you a lot of words have multiple meanings? How do you disambiguate to get the right one? Yeah. So this this, embedding technology, this Muse model helps us do that. And then Facebook is given we use a tool, they think it's called Facebook. Ai similarity search. Yeah. And both of these are open source tools, y'all you have to put in the effort and have the knowledgeable people to master their use. And that allows us because great, it's great that you've now got all these numbers you can compare, but that's a lot of numbers. That's you half a billion a day, you know, and we have we see 600 million unique pages every month. So so how do I great, I want to rank the 600 million pages to see which ones are most about skiing in Utah. Yeah, that's, you know, how do I do that quickly, and then and affordably? So fate, the Facebook tool helps us a lot with that.

Grant
So let me ask you a question that So so far, you've been talking about leveraging AI technologies to help you get your arms around that sheer volume of data on a daily basis and to try to extract some meaning and semantics and understanding from it. That's a good point that's on the side of ShareThis and the benefits to ShareThis. What about it from pivoted to the other side? What does it mean to it is, you know, I talk a lot with small medium organizations, how does that benefit them? What takeaways or values come over to help them through something like that?

Michael
Well, what the I mean, the industry that we started with, is was is advertising and programmatic online advertising as a place where we make our solution available. And so we were at this point, probably the leading source of the ability to target ads based on interest. So if if A small business were doing online display advertising and they went to Google's, if they use Google's platform or trade desk, or any of the major platforms, and they searched on, I want to find people interested in skiing in Utah, our data would be one of their choices to find that. And so it's designed to provide a broad set of individuals who in the last 30 days have shown some interest in that topic. And it could be, you know, it might be at the level of skiing, and they might, then they might, but but the nice thing about it is that we we've, I mean, it's hard, this is harder for the stats, that's what's available for the smaller business. That's, it's, it's right off the shelf, you can, you can use $1 worth or $10 worth or $100 worth if it works for you. But then on the big company side, we use some of those tools I talked about for is, well, what if, what if we don't actually have ski in Utah, we just have skiing, right? Well, we well, for an advertiser can can say, well, I need to skiing in Utah. In fact, I need to, you know, skiing in snow. But what is the alter? You know, we can create a segment using keywords and, and topics that is just about that is exactly what they need.

Grant
So if I were to look at maybe an advertising opportunity, leveraging, you know, this great insight that you have, does it allow me to target specific demographics, specific locations or locales? So like, you know, you're able to?

Michael
Absolutely, it's pretty much anything you could, I mean, because every kind of website needs sharing, we have our, our customer base, our base of publishers use our tool is pretty representative of the internet as a whole. And so if your interest is travel, we've got sites that are about, you know, traveling Las Vegas, traveling to Europe traveling to do outdoor activities, if you're interested in financial products, we can we can find things, you know, content that relates to whatever be at a mortgage or or FinTech to know. And we we represent those in about 1500 standard audiences that we distribute every day. And every day, the nice thing about our data, compared to a lot of datasets is we refresh it every day. Yeah,

Michael
I mean, it's every second, right? I mean, yeah, it could be, you know, people talk about real time, and we were always looking for people who've got a real time use case. But yeah, at this point, the the most frequently we refresh for a client, the customer is up by a by his hourly.

Grant
Oh, it's hourly, okay, that's, that's still really up to date. Yeah. I mean, if you had hourly insights on what the what's in the mind of people are the consumers that's really fresh data?

Michael
Yeah, yes. Yeah. Yeah, one of the areas that we that we are moving towards is trying to go beyond advertising and inform other activities like demand forecasting, you know, how much should we order for a store in a given location? Well, our data about how much interest is being shown on the products of that store, and in that store, in that area, we can sort that way, and provide that as an input.

Grant
That makes that makes a lot of sense. You know, there's, there's some retail organizations I've worked with with AI. And obviously, it always comes back to or not always, but most of it comes back to the supply chain, right, getting further and further left in terms of their their demand forecasting. And if they were able to understand you know, where that interest lies, it does almost gets to, oh, I know, this is a stretch in terms of language, but it's kind of a sentiment analysis, a play on that. Right. It's the ability Yeah, the ability to say I understand what the sentiment is in terms of where their interests are. And if I understood what that was, in terms of particular set of products or other things I'm offering, and I could get that further into my, into my supply chain, that would be really valuable to Yeah,

Michael
I mean, it's nice that you mentioned that we do we do actually score the sentiment of the content on the page. So we're sentiment is useful, either to only talk to the people who are in favor or opposed or the middle, we can we can build an audience that or provide that as a data element as well.

Grant
Yes. See, that's that's powerful to understand the the sentiment of the page itself, even how people are talking about it, or what they're doing, have you ever ran into the ability to use it in terms of IP tracking, right. So in other words, if there is an organization that had a certain set of IP and, and and really, yeah, they felt like oh, my IP, I've lost control my intellectual property, it's showing up in other places.

Michael
Oh, that's interesting. You know, I was thinking of I was thinking of the I the the IP address the Internet Protocol address. Yeah. Should have been more clear. Yeah, I'd love to answer that question. But that wasn't what you were asking.

Well, yeah, answer. Oh, we'll start with intellectual property. Yeah. One sec. Regarding intellectual property? You know, we have it. Let me think about that. Let me give you the scenario. I had, one of the things I've thought about that we haven't taken on it, you know, is that is, is using using intellectual property as a data set? Yeah. If if we were to, to read to do the same kind of analysis I talked about earlier on trademarks. Yeah, it could mean be the means for discovering which, what sites were about branded products by seeing the correspondence between the trademark and the, because that's always you run into difficult How do you tell something's a brand? When is Jaguar a brand? You know? Exactly.

Grant
Yeah. Yeah, it's a fascinating problem. I had a company reach out to me and say, Hey, can you develop something in this area, and we did some work on that. I called it smart catch, but they were looking to protect their IP, their intellectual property, which was, we've got this corpus of information. And, and we've got others that are, you know, getting access to it and are promoting it, you know, elsewhere out into the, you know, online universe there, or metaverse. And, and I want to be able to discover, you know, when it's opportunistic, and you can use, you know, SERP and other technologies to try to find some of that stuff and do lots of scraping. But that's got its own challenges in terms of a solution. And where you've got this opportunity to listen. Right, right, to observe what people are sharing and to the to compare that against a corpus of protected material, right?

Michael
Kind of an intro you're giving, you're giving me a product idea. Seriously, one of the things that we've done this year, is to create what we what we call, you know, similarity scoring. So similarity, and that's gonna cause Yeah, you can literally give someone who was curious about the dispersing dispersion of intellectual property, give us a domain. Yep. And, or a, you know, the piece of content that describe their, their stuff, and we would rank our sites for which ones had it most. Right. And, you know, whatever the top 100, you know, and you know.

Grant
What I found interesting on that, when I built the initial piece on that was that I found that, in some of the discovery, in some cases, what I found was a foe. And in other cases, it was a friend. Exactly right. That, you know, okay, just because I found it doesn't mean it's an enemy. But, but it might be, and so you want to then notify them? Is this? Is this someone that's an ally or not? Anyway, interesting thought?

Michael
Because I think I think that sometimes there is a, you know, I don't know, there's a presumption that fraud detection or a bad actor detection is, is, you know, worth more, etc. But I do find that in a lot of cases, the pro cases are actually, you know, sometimes you just by suppressing something, you do more yourself more harm than good. Yeah. Yeah. Right. Right. That's another I wanted to touch on the other meaning of it. Yeah.

Yeah. Now IP address. Yeah, yeah. So So an IP address is one of the four or five things that we capture for each case. And there's a lot that you can tell from an IP address, like, it can be translated into a location of origin, we approximate we resolve that to within half a mile. So that it's still relatively privacy compliant, and you know, not too revealing, but it certainly helps understand, you organize the data by where it's coming from example. And so the one use that is, has been an important one for us is business to business. So we, we have a number of the major companies that are in the business to business world license our data as one source where they're able to see people from a from an intellect Internet Protocol address that is owned by or been associated with a particular company. Oh, and then see what sites that that IP address is showing interest in? Oh, it just can be. Yeah, so it can be a signal that oh, it seems like you know, Chevron is interested in a new CRM system because they're you know, there's there's a big spike in that kind of traffic Awesome. Yeah, that's awesome. Yeah. Talk about so almost like a lead management. Yeah, solution for sure. That's, that's powerful. Yeah, to do that. that. Oh, there. Yeah. And that's yeah. And IP in general, I think the location implications are a really well, it's how I can, how we can do that demand forecasting I mentioned earlier, it's about looking at the origin of the data.

Grant
So some of the AI solutions that I've built take into consider location. So So in other words, okay, but in what I've been doing is more around, oh, some transaction occurred? Where was that transaction initiated? From? Oh, this, you know, here's the IP address. Okay, I know that where they are on the planet. Now, tell me what the context of what's taking place in you know, at that location? What is what's the weather like, right, what are other events that are taking place in that location? And then then use an AI to help draw inferences on, you know, to what degree are those factors affecting it? It sounds like you might be doing some similar things with that

Michael
I well, I think we could be a great contributor to any solution that was along those lines. I was adding that dimension of what are people looking at? What are people interacting? What topics? Are people in this location more engaged by then people in general, fascinating those comparisons?

Grant
Yeah, it's fascinating is okay. Very good. All right. So let me ask you on. Okay, so we've gone from the the big corpus of what you're collecting on a daily basis, or hourly, actually, hour by hour. And then we talked about the impact to, you know, maybe businesses organizations, when when is there a particular case or outcome that you feel like you could talk about some specific example where some organization used the advertising from that? What you did, and it had this sort of impact or effect on them? Do you have any sort of case study like that? Well, it's,

Michael
I guess that some of the ones that are coming to mind, I think, I mean, there's some of it's very straightforward. Yeah. An advertiser, like Western Union, is looking for people who want to make payments, you know, at a distance, I mean, wire wire transfers and payments, and we offer people showing interest in wire transfer, so that the simple act of being able to get your message in front of people who have recently shown interest in it is the is the, you know, it just doesn't need no explanation. We've taken that though, one of the things we did this year that I'm proud of is we were inspired by some of the events of last summer, to get more try to take a more active role and figure out what our data was good for. Beyond commercially, and, and we ended up creating a data for good part new part of our taxonomy, we call data for good. And so people interested in social justice loving people entered interested in veterans issues people wanted in. And so and those those segments, you know, have gotten are getting a growing amount of usage by advertisers who either, you know, wanting to demonstrate their commitment to the court to a cause, like, or to find or teachers or to, you know, communicate, right people who have concerns of that kind. So that's been one. Yeah. Another kind of it's, it's not in the mainstream of what we do. But we've, I think this data could be really great as a as a resource for educational institutions. So we've actually a major business school has has is testing I've taken a take taken a subsidiary six months of our data, and they're looking at using it in a project that they have to investigate unemployment. So fascinating. How could you How could you see earlier unemployment trends in a in a location or region that could help the for the process of forecasting the unemployment rate, and it sort of feed into it? Because I've, what I've, I think that lots of people govern organizations included, are somewhat frustrated by the fact that, you know, traditional means of forecasting that were invented before there were personal computers or barely work computers. Take a long time, you get to find out that 40 days after the month, what happened in the month, I love both data can be used to generate that much more quickly.

Grant
Yeah, Michael, that's I love how you're bringing that up. It seems like it has both the opportunities for not only the capitalistic aspects, but the altruistic aspects of this, the values and benefits that can help society and be pulled out of that. I think that's awesome. So all right. I've thrown a lot of questions at you. So let me ask you this, if you will. To direct direct my listeners to where to go to learn more, where would you send them?

Michael
Well, I would, I would love them to visit our site, because and in particular to, you know, to ShareThis.com, look, look at our news and our, our blogs, we we basically we publish both as you know, as a demonstration of our the value of our data. And and it's just a general service, we publish a lot of educational and informative information about trends in the economy, and, and public interest generally about how to do marketing well about trends in data. So so we we, we try to be a resource for people and I love I'd love people to visit that content, sometimes. Some of the best stuff is is not on on the nightly news. It's like putting some of it out. I could also you know, I can give you some examples. It would be fun. I go right ahead. Knowing that knowing this audience I we are getting a sense of who maybe was listening is interested in the show, I asked our team to identify some current trends. Yeah, I guess as we come to the end of 2021. Yeah. And so so we put these together. So what one is that, that, that while the world isn't, we're seeing the trend of the gradual resumption of events in person events, even though COVID continues to cycle up and down against the backdrop of COVID. So as of August, for example, 77% of advertised events were in person events, there was a period where, you know, year and a half ago, there was there, they basically no almost having anything, it was just shut down. It was virtual or nothing. That's interesting. So as we adapt, we are adapting. And so as you as you think about should I make plans for a virtual vet, should I invest in advertise? Should I invest in participating in virtual event? Yep, don't count them out. Even if you're nervous, you know, they, they're coming back steadily. Another thing, pattern we observed in finance, that again, you know, COVID is inevitably one of the backdrops to what any of us are thinking about, but people are continuing to be engaged with saving money. So, it so as you think about what, oh, you know, what is what's going on in the in the economy? As the, as virus uptake increases, as one of the things to extract is, is increased saving? And so if that's a, again, depending on your business, how that factors in if savings is your business? Yeah. When your could be good, good to you. If if, and then let's see, what's another one? Let's see. You know, we've heard a lot about supply chain issues. And you know, what, but what, if your retailer what a consumers most worried about? When and so the top concern is shortages and out of stock, and 51% a second costs, inflation and rising prices at 28%. And then staffing issues like worker shortages and strikes, 14, and last last of all shipping delays. So it's thinking about communication strategies, what's on people's minds that might make them not come to the store? That sort of thing? So I'm not surprised. Yeah, yeah. So and we're, we're putting out new new stuff of this kind every, every month in the blog. And and I firstly, look, I think we did we have Superbowl trends out, as of yesterday, I think.

Grant
So it's already started to build right. That's right. That's, that's amazing. So So you gather it on an hourly basis, and then you do the AI on it

Michael
Truthfully, truthfully, Grant, it's being gathered continuously. Okay, that's, that's what I thought, yeah, I thought we built we build it as it happens, okay. We literally, you know, record a record for each thing. That's, that's, that's filled out all the way with all the data that will that will need eventually. And then once an hour, we some or as we frequently as our we'll sum it up into a distribution and push it to someone but the most people get their get their data delivered overnight. Amazing. It's picking it up on their AWS bucket. Like Well, this is

Grant
Fascinating. Any final comments as we wrap up here?

Michael
Well, you know, I guess that I hope I've given you a sense of the I mean, AI is critical to our business. We are you know, we When we started on this track, we were about a 50 person company, we're approaching 100 person company. And so you don't have to be, you know, IBM to use AI AI to build a great business. So it's a combination of finding the right tools and a core of of talent, the right kind of talented people, and you can and and then, frankly, sustained effort over a period of years and you can build a business that is really hard to replicate, without without it, so very hard. Right. That's, that's my thought. That's, that's

Grant
Wonderful. Well, Michael, thank you so much for taking your time today. I appreciate you sharing your insights and guidance with us today, everyone. Thanks for joining another episode of ClickAI Radio and until next time, go get some ShareThis.com.

Thank you for joining Grant on ClickAI Radio. Don't forget to subscribe and leave feedback. And remember to download your free ebook, visit ClickAIRadio.com now.