AR Data Sciences
This Month’s Q & A: April 2021
Chairman and CEO, Inuvo
WE CONSUMERS ARE strange birds who sometimes don’t even know our own minds. That used to make it hard for marketers to pin us down, but now, with the help of its revolutionary IntentKey technology, Little Rock (and San Jose)-based advertising and Artificial Intelligence company Inuvo can read our ever-changing motives like a book. We asked Inuvo’s chairman and CEO Rich Howe to tell us how he knows exactly what we’re thinking.
Before we talk about Inuvo and AI, tell me about your career before this.
Prior to jumping in and starting to build Inuvo into what it is today, I worked at Acxiom, hired by Charles Morgan. But there’s an interesting story that actually predates my time at Acxiom, and maybe it’s important because it’s germane to this discussion.
Before Acxiom, I was a senior executive at a company called Fair Isaac Corporation (FICO), based in San Jose. We invented the credit score. I was in a strategic role in addition to an operating role, and one of the things I was advocating for was a merger between FICO and Acxiom. In my mind, FICO had probably the largest collection of analytically minded employees of any company on the planet. Think Ph.D. mathematicians. And Acxiom had the largest repository of consumer information.
This is like peanut butter and jam, I thought, so I put the two CEOs together, and that’s how I met Charles. The merger never happened, by the way, and there’s a whole host of interesting stories associated with that. But I did get to build a relationship with Charles while I was at FICO. Later we connected again, and Charles said, “Look, I’ve got a lot of things going on here at Acxiom that are related to trying to maneuver the business to be more aligned with the future than the past.”
At that time, Acxiom was mostly offline in terms of the marketing and the data use for the company, the services, so he was thinking about buying some digital assets and trying to innovate products in the digital area. “Why don’t you come over here and help me make that happen?” he said. Then he empowered me by making me the Chief Marketing Officer, even though my job was really more strategic and operational as a result of being assigned P/L responsibilities for the businesses where innovation was most required—one of them being the data business.
What’s your educational background—marketing?
No. I have two degrees in engineering. My undergraduate degree is in structural engineering and my master’s is actually more in computer science. But even as an undergraduate, I was involved in writing software and using computers to solve how to build complex structures.
In my career before FICO, I started a couple of software businesses and sold one of them. I’d also been an executive in a pioneering AI firm—a company called HNC Software that was started out of San Diego, California. HNC invented a technology that today is really ubiquitous. If you get a call from your bank asking if your last number of credit card transactions were actually you, that’s actually a neural net we invented at HNC that looks at credit card transactions and has this uncanny ability to predict whether or not a transaction is fraud or not.
We invented that technology and commercialized it, and it was hugely successful. We’re talking about the mid-1990s, so it was one of the very early commercial applications of Artificial Intelligence technology. Maybe this background explains why Inuvo’s business is headed in an AI direction.
Your website says, “Inuvo AI solutions align brand messaging with appropriate content and consumer intent.” How is that different from the old ways of marketing and advertising?
Before, the method that was used—and still is used by many—was to take in lots of different files that have consumer information in them. It could be name and address, occupation—there’s a whole collection of things.
At Acxiom, for example, we created the paradigm of going out and purchasing data from hundreds, if not thousands, of different sources. Some of them were public record sources—driver’s license records, real estate records—and others were not. You could buy a list of consumer names and information from various companies, and the technology that was developed by Charles and others was the way to bring all that information together in one place—to have a common way to identify individuals across all those files. Because it was necessary to say, “Hey, I just found a record in file number 600. How do I match that to what I have already for this person?” And that was a product called AbiliTec.
It was just unbelievably leading-edge technology at the time. But looking back now, you could say that it was limited in the sense that just the processes required to go get all that data from all kinds of different sources was cumbersome. Then bringing it together was equally cumbersome. It created delays in the—let’s call it “freshness” of the data, because not all the sources of information have the same timelines on them.
The best you could do was to structure your data into a set of categories for consumer interests, and then try to fit everything you know about that consumer—age, gender, zip code, the brand of car they drive—into those buckets. Then once you have that data file, which you’re always rebuilding, you try to make it available so people can buy it, because you’re in the business of selling that data.
At the time, it was fabulous, let me tell you. We killed it, because Acxiom became a roughly $1.2 billion annual revenue business. Now we look back and go, “Wait a second, there’s a couple of problems with this.” One problem is you’re not really getting to the true intent of the consumer, when you’re limited to hundreds of high-level pieces of demographic data, like, “These people are interested in technology,” or, “These people are interested in fashion.” What specific part of technology are they interested in? What part of fashion? Wouldn’t it be better if the data was more “real-time” instead of structured and more buyer-signal based?
It used to sound like hundreds of bits of data on somebody was a lot, didn’t it?
You’re absolutely right, and that’s the irony of it. I can even remember us high-fiving and saying, “Wow, look how much we know! Think how much better the advertising experience could be with this!” But the reality is, that demographic and categorical data doesn’t let marketers hone in on the smaller sets of people who are hyper interested in something way narrower than, say, the general category of “fashion.” Optimization is all about getting to the smaller, more intentional data sets. There are reasons we buy what we buy; the challenge is to find those reasons.
Inuvo was formed with the goal of totally reinventing the process, to make it 100 percent a function of AI-generated, real-time data and modeling. And at the same time, because the world’s changed since we were at Acxiom back in the day, we needed to be more sensitive to consumer privacy issues. So we asked ourselves, is there a way to do this without having to take in all this third-party data and, at the same time, create a different, better solution that acts like a compass by pointing toward the real intent of the consumer?
How do you go about such a reinvention?
When you’re thinking up these kinds of projects, the classic scenario is either build it or buy it. But sometimes you can’t buy it because it doesn’t exist.
As it turned out, I had a relationship with an individual by the name of Terry Opdendyk, one of the early venture capitalists in the Bay Area. Terry’s firm is called ONSET Ventures, which invests in startups—one of which involved an interesting technology created in a machine-learning lab at UCLA. The people who did that, including the original professor, started a company to try to commercialize that technology, and they were capitalized by ONSET.
But at a certain point in 2017, the ONSET fund that had made the investment in that business was going to close, which meant that all of the assets in that fund typically get liquidated. While Terry had a particular interest in this business, the decision had been made not to put any more money into it. He and I cut a deal by which Inuvo acquired all of the technology, the patents, and the rights to this thing.
So tell me—what is “this thing”?
The core of the UCLA research was about creating a new source of data and understanding using the Internet as the teacher. That’s the high construct for what was going through these very, very smart people’s minds. Unfortunately, with the money that ONSET had given them, they had also branched off and started building other software technology that was related to that core piece of tech, but had a different focus, you might say. As a result, they were never as successful as they could have been—because they hadn’t focused on the one differentiating piece, the AI piece that we saw as the secret sauce.
That piece is very much akin to the way our brains work, at least as best anyone has figured out. The analogy is that our brains are these gigantic libraries, and all these neurons within our brains are like the books in that library. Our brains know how to access and link all of these books, and that’s how we have this ability to know things, and to figure out things.
Once we got hold of that technology, we spent a couple of years working toward a commercialization of the technology that effectively mimicked that model of the brain for what we call “concepts.” What is it that humans understand? Well, we understand a language—language is the basis of everything we learn and know. We set out to map all of the concepts in the English lexicon. A concept could be a person, a place, a thing, an emotion, a sentiment. For example, “I love you” is a concept. Our brains know what that means when somebody says that to another person.
The original UCLA science analyzed the English language and said, Hey, there are 24 million really important concepts in the English language. The question was, could the machine be taught to figure out the probability between all of these concept connections—much like our brains do when we hear, see, or read something and can immediately associate that with all the other information we’ve learned?
We mapped out the interrelationships between every one of these important concepts. We did it through analytics and crawling the Internet. We’re talking about billions of pages of content and trillions of relationships between words. That’s an important concept in this thing. The Internet was the teacher for how our AI brain would work, because it effectively canvases all of humanity’s knowledge. And what we’ve created from that may be the single largest collection of concept interrelationships about everything that ever existed.
Wow, hard to get my brain around that concept. How does it work in marketing?
There are lots of examples, and they bring into play why intent is so interesting when you can predict it. In fact, we launched our patented product called the IntentKey in 2019. The name implies what we do, which is determine the key to the intent of the consumer. The product a consumer purchases is the what; this is more about the why.
Think about a concept like sleep, and just pause for a second; think, sleep, sleep, sleep. Already, there are a dozen things going through your head. Frankly, there are thousands of things that are associated with sleep—and that’s how these connections work. Are you thinking about sleeping pills? Sleep quality? Is it health? Are you thinking about sleep because you’re worried about your health, and you know if you sleep more, you’re going to be healthier?
You can take those threads and keep going, and every single one of those threads actually has another thread, like another book. For example, if I’m thinking about sleep, it just so happens that it’s because I’m going to be travelling soon, and the reason I’m interested in looking at sleep has less to do with sleep and more to do with travel. Hey, I better get one of those neck pillows because I’m going on a long plane flight to Europe, and I want to sleep.
So the context changes almost immediately, which is exactly how our brains work. There are lots of pathways, and when you know the pathway or the context, you can actually get to what the real consumer intent was.
I’ll give you another example of this using the same sleep analogy—and a real customer example for us, by the way. I’m not going to identify the client, but let’s just say they’re in the headphone business, and they had a product that fits in your ear at night and makes soothing sounds to help you sleep. It’s called a sleep bud. So how do they find the perfect customer for this product?
Like a lot of conventional marketers who are burdened by the paradigm that we created at Acxiom, their thought process was limited by what they know. OK, well, I guess we’ll go buy a list of people “types” who like headphones. If they like headphones, maybe they’ll be interested in buying this.
The fact that people who buy this sleep bud might have some association with people who like to buy headphones isn’t all that interesting, and they probably don’t need Inuvo to help them figure that out. There are a whole bunch of companies that do that already.
What’s different about us is, we start at the end of the sales process. We look at who’s actually buying the client’s product, and then our AI figures out why. It’s about the intentions of the various audience groups, not groups of consumer types.
For our sleep bud client, our technology was able to figure out very quickly that one of the reasons people were buying those sleep buds was because they had noisy pets who were keeping them up at night. And not just any noisy pets, but a couple of specific kinds of noisy pets—short-nosed dogs like pugs and bulldogs. If you’ve ever had a pug or a bulldog, you know those dogs have breathing problems, so they’re noisy and they snore. It’s something related to their palates.
Our IntentKey technology was able to find that whole thread and connect it to sleep quality. Then our tech says, “Okay, let’s find everybody who has a pug or a bulldog, because that’s who’s buying this thing.” And it automatically finds the client a group of people highly focused in that area and puts the ad in front of that audience—instead of putting it in front of people who like headphones, but don’t have sleep problems, so they could care less.
That’s pretty amazing.
It is. Especially because most marketing today is very trial-and-error oriented. As a marketer, you’ve got these demographic categories to pick from, so you pick one and start marketing to that category online. But you know going in that the category is too broad, so you’re going to have to keep adjusting. You run it, you see whether or not you’re getting enough people buying your thing, and if they’re not, you tweak it. And you keep going through that cycle to try to improve your performance.
Our tech is more of what I call an adaptive technology. You don’t need as many humans in this model. And frankly, humans can’t act fast enough or have the related context to see the trend changes, because buyers are always changing. Today it’s pugs and bulldogs for why people are buying those sleep buds, but then, suddenly, you find that it’s an audience living way north in Alaska. And you’re like, What the hell? What does that have to do with my sleep buds? And the answer is, the sun stays up something like 20 hours a day in Alaska, and that’s probably why a lot of Alaskans are buying the product. Think about seasonality alone: Many businesses have products or services whose buyers vary throughout the year, and our technology adapts automatically to these situations.
Do you have clients saying, “Alaska? That’s crazy!” I mean, do you have to do a lot of client convincing?
It’s true, they always have to be convinced in our early sales cycles because our technology is a paradigm shift, and most clients think they’ve heard everything about how to do marketing. But when they get the information from us, the insights of what it is that’s actually driving the sales, they’re usually pleasantly surprised by it.
You talk about how the AI identifies what’s going on right now. Is there a timeframe for it to successfully identify who your client should be marketing to?
You should be marketing to that consumer when they’re in the market to buy, and that’s one of the advantages we have as a result of the way our data is generated from this AI. Again, going back to how we did it in the past, it was cumbersome, it was late, and there was nothing really intentional that we knew about the buyers. Ours is just the opposite. We can show our clients that they’ve got people right now, for whatever reason, wanting to buy this or that product. So just put it in front of them when they’ve got their hand up saying, “Right now!”
In a nutshell, that's why we’re killing it. Because we’ve reinvented this industry through Artificial Intelligence, and the results have been off the charts. In 2020, we beat our clients’ goals by 46 percent. That’s a big number in marketing.
But while beating our clients’ goals by 46 percent is fantastic, I'll tell you what makes it even more fantastic to me. When we call on a potential client, you can bet they’ve already got somebody doing this. I mean, marketing isn’t new, right? When we beat our clients’ goals by 46 percent, we’re probably also beating the best competitor we've got, because that's whose performance they’ve given us as a benchmark or goal to displace.
I’ve been around technology long enough to know that incremental advantages don't get you where you need to go. When I see numbers like 46 percent across the board for a client base, that tells me we’re beating the market by 46 percent. And that’s the kind of performance you can build another billion-dollar business on, like we built at Acxiom.