NetBase | Interview with its Chief Innovation Officer & Co-Founder – Michael Osofsky
In Mountain View (CA), we meet Chief Innovation Officer and Co-Founder of NetBase, Michael Osofsky. Michael talks about his story how he came up with the idea and founded NetBase, how the current business model works, as well as he provides some advice for young entrepreneurs.
Martin: Hi. Today we are in Mountain View at the NetBase office. Hi, Michael. Who are you and what do you do?
Michael: Nice to meet you, Martin. So I’m Michael Osofsky and I’m the Chief Innovation Officer and Co-Founder here at NetBase. I’ve been here since the beginning and I have taken on this role of innovation really so that we could add executives to the team who had 20 and 30 years of experience, and I’ve focused on bridging the customer and the technology, inventing new products and that kind of thing.
Martin: What did you do before NetBase? And what made you come up with this business idea?
Michael: So I was MBA student at MIT Sloan School of Management. I chose that because I really wanted to learn how to do innovation the right way, and there are a lot of, believe it or not, social scientists who have researched innovation and come up with some best practices for that. So while I was there, that’s really when the idea for NetBase came together. It originally started off as this idea I was going to create this enormous database of the world’s unmet needs and the world’s technologies and we were going to randomly pair them up and produce innovations.
So it turned out that another company was already doing that, yet2.com, and since then the founder, Ben duPont, has become a friend. And so we actually had coffee the other day and talked about how NetBase had to pivot, and we decided that instead of filling a database of needs and technologies and trying to produce innovation that way, we recognized that the database already existed. It was the web and social media. The word didn’t even exist at the time, but what we figured out we needed to do is go out to the web and harvest the information that we need for marketers and R&D professionals, extract it using a technology called natural language processing, and put it into a searchable database to help people do marketing better, innovation better, R&D better.
Martin: Cool. Michael, how long did it take for you to do the pivot from the first idea and then realizing, “Actually another company is doing something similar. Maybe I want to pivot.”? How many months did it take?
Michael: So I’d say before we officially founded the company, I’d call our founding moment really when we got our first customer and tested the idea out. So several months before that, maybe six months before that, my co-founder and I were really incubating it as a project in our respective MBA programs, and so we’d take different classes and we’d build out different aspects. So there was one class in particular taught by Eric von Hippel at MIT, and he had some research he was sharing around, user innovation and how users oftentimes are the ones who come up with innovations but along the way they leave this digital trail mark leading to their pot of gold, what they have discovered. And that digital trail mark entails writing posts on forums and maybe writing articles describing their unmet needs, describing the inventions that they’ve made along the way. So what I realized in a moment of glory really was this wonderful moment just realizing this in one of his classes. It was like, “Wow. Okay, that’s what we’re going to do. The database I needed was already there.” It was all at the Internet and there’s this digital path that innovators have left. We could just snatch all that up.
We had to have some kind of technology then to extract it. And that ended up being a huge project that we had to raise funding to get. But something I’m very proud of about this company is that when we… At some point, you have to set up your bank account, right? So the first deposit that went in wasn’t venture capital chasing after the latest fad, and it wasn’t even government grants, because a lot of our competitors started that way for some kind of pie in the sky research. There was this middle road where we had found a customer who got value from our idea. We prototyped it for them, we tried it out, we delivered our results to them, and they paid us voluntarily actually. We didn’t even ask them. So that first check that went into our bank deposit was for cold hard cash that we earned from our customer, and that set the basis of one of our core values here at NetBase is that we deliver value for customers and for our shareholders and our partners. Everybody’s got to win in order for you to have a healthy business.
Martin: Cool. Michael, walk me through the prototyping that you did for this one client. So first, how did you find this customer? And second, this prototype, what type of data sources did you use and what type of value did you create for this customer?
Michael: So it was a customer that…they were an IP consulting firm and their job, they’d be hired by a large chemical company that might have invented something. In this case, it was these uniform microspheres, these tiny little spheres that you could encapsulate drugs or other chemicals. What was key about them is that when you ingest them, you could time release whatever was inside. And so this invention had been invented for one purpose, but the market for that invention was too small really to recoup. So a common problem is called intellectual capital management. You have to find alternative uses. And so that goes back to our database of market needs. Here was a technology and we had to find other market needs. Other times we’d have the opposite. We might have a specific market need that we’d need to find technologies to solve that problem. The nexus of technologies and market needs, that’s innovation. And so the innovation problem for this customer was that direction – have a technology and find the market need.
So typically this would have been done with lots of manual research, lots of just poring over documents, maybe go do some focus groups. These projects can cost maybe $30,000 for an IP consulting firm to do this kind of work. So our approach was to use publicly available data on the Internet to suck it all in and pull it down. So we didn’t have the resources to pull it down, so what I had to do instead was this really crazy hacking of Google, where we would just blast Google with all of these queries. It would be like we were trying to find unmet needs, so we’d figured out these certain phrases like “it’s difficult to,” “I have a problem x,” all of these different phrasings for words that we’d find in a problem statement. So we blast Google, blast Google, blast Google, pull all that data down, and that’s all the data that was market needs, and then we’d scrub it, looking for needs that could be met by the benefit for the properties of this uniform microsphere.
And so we found opportunities for different foods. I can’t talk about it in too much detail, but suffice it to say what happened is we delivered the results and we had done this as a free project because we had been introduced by the MIT venture mentoring service to this company and they ended up paying for it because inevitably what probably happened was they went into the meeting with the client and facilitated a brainstorm but whenever they got stuck, they probably had a crib sheet under the table with all of our ideas, all of our unmet need markets that we had found for them. So they paid us a whopping $2000 and we thought, “Wow, this is exciting.” So that $2000 then became the basis of our pricing for our next customer. And so with our next customer, we got some introductions from the authors of some great books and intellectual capital management, “Rembrandts in the Attic” and books like that.
And so in our second pricing negotiation, my co-founder did the business aspect of this, Jonathan Spier. He thought he’d be brave and he thought he’d ask for double that. And so our prospect heard the price was $4000 and sat back. I think my co-founder was pretty nervous at this point, but it turns out the guy was so surprised because it was way too low. He said, “Well, I would have paid $20,000 for this. Let’s settle on $12,000.” So Jonathan brought home our first 12K deal. Every deal after that was 15K. So that became our standard price and that was in our forecasting models for the longest time.
The other thing that I was going to mention is how the prototyping worked. So I talked about how we harvested the data. We developed this kind of Google blast approach, then we’d do a lot of manual scrubbing, but then the output would be we needed it to look like software because we wanted it to be a software company. So what we’d do is we would format a result set screen that looked like you’ve just pressed the search button but it was all hard coded results that we had manually filtered data and carved out our own little snippets from the tags, but to the end user it looked like it was a result set. So this is why we got a check for $2000 because we weren’t trying to build the perfect system right off the bat and we weren’t just trying to deliver a PowerPoint. We went middle road. We manually got the data, we massaged it, we made it look like software.
And why this was so key is because we had seen this passes the wallet test. Will the customer open their wallet and take out money and give it to you? We knew that if we could give this result set and the way it looked to an R&D team, our own software developers, they’d have something to aim for, something specific and concrete. So that was our first specification. Go make this more robust. Take out the manual stuff. And we even came up with a mantra for the team. We said, “Well, it took us 40 hours to develop this result set,” and we said, “Okay, guys, this is pain time, and all of you here, you’re here as our Tylenol, our painkiller. Go reduce that time. Make it smaller and smaller.” And through the course of many, many iterations of strong development, we got that down to push button.
BUSINESS MODEL OF NETBASE
Martin: Michael, if you look at NetBase today, how does it work? And so what type of data sources are you using, like Twitter, Facebook, Google, whatsoever? And how are you then analyzing the data? Is it only that you’re using NLP or also graph analyzers or some other kind of analytics in order to take insights for matching this kind of technology with their needs?
Michael: The data is, like you said, Twitter, Facebook, Instagram. Tumblr is becoming a popular channel. Public areas of these sites, we don’t get any private data. So we also get the comments off of ecommerce websites and that kind of thing. So that’s all of the data. And from there, yes, we apply natural language processing. It’s very deep. So for example, this can of Coke that you have, what do you like about Coke?
Martin: Actually, a good question. Maybe just the taste.
Michael: You like the taste. Martin, believe it or not, some people like the fact that Coca-Cola is harsh.
Michael: Yes. It burns their throat, and they love it. So if somebody says that, we’re able to figure out that they’ve said something positive despite the fact that it has a bunch of negative words on it. So we don’t treat that tweet like a bag of words. We look at the combination of the words, we parse it, and we figure out, “Okay, this is a positive.” And what’s positive is that it’s harsh or that it burns their throat, and that’s considered something positive. The competitors that we have in the market, they’ll most likely label that as a negative just because of its prevalence of negative words. But natural language processing gets at the core. So we go very, very deep, but also what’s special about our technology is that it’s very fast. Within 11 seconds of somebody tweeting how they love how Coke burns their throat, we’ve not only obtained it from Twitter but we’ve pushed it through our natural language processing and we’ve surfaced it on the user interface. Somebody might have a live dashboard, a live pulse running about Coke and it’ll show up, labeled appropriately, putting the right analysis and all that stuff.
So mostly it’s natural language processing. We’re also beginning to do some image analysis. We do some statistical analysis, too, words and phrases that are important. But where we’re a little different from our competitors is that if somebody’s talking about Taco Bell, you’re not going to see Taco up here and Bell down here in the word cloud or something. It’s Taco Bell. We know that that’s a phrase. So we’re pretty good at that sort of thing. So those are some of the analysis.
Martin: Michael, what type of products or use cases are you delivering to what type of customers?
Michael: A lot of our customers are consumer insight groups. They traditionally go to agencies for focus groups and surveys and that kind of thing, and we’re faster, better and a lot less expensive to do that because a focus group can only look at a handful of people, so very small sample size. Plus you have groupthink and the different kinds of biases that come into the research. Now, every technique of understanding consumers has its own biases. Some people put these biometric centers on your head and they make you shop around the store. Well, you can get a positive or negative read on that maybe with some degree of accuracy, but you don’t maybe really know what’s going on for them. What are they thinking? What are their thoughts? We get at that.
And anyway, if I can’t convince our customer that we’re better because of these benefits, we always have the fallback of this is a heck of a lot faster. To set up the focus group or survey, it’s going to take you months and you can only afford to do it once or a couple of times. Within 11 seconds of you tweeting about that Coke, we’ve got that insight produced, so it’s very fast and very, very inexpensive. And inexpensive means that you can get insights about a lot of other things. So if you’re Coca-Cola, you don’t have to just look at what people think about your brand. You could also look at your competition. How many competitors do you have? There are so many soft drinks out there. And then you can think about things like the ingredients. What do people think of high fructose corn syrup? Or what do they think of sodas in general? Or what do they think about the Olympics? That gets into another use case, helping you understand consumers so that you can generate better creative and better targeting.
Another use case is audience marketing where maybe I’m Walmart and I’ve got millions of people talking about me. Well, maybe they’ve mentioned Walmart once in a year, but what else are they talking about? Where else are they shopping? What are they talking about? Eating and drinking? What do they drive? What do they watch? If we know that, that’s going to tell us where should we be advertising and that kind of thing. So audience marketing is another use case. There’s a bunch of use cases. There’s so many applications for this data, and that’s just scratching the surface.
Martin: Michael, imagine I’m a customer of yours and I’ve got a research question. Maybe what are people thinking about my brand? Is it then that I just can plug it into the platform and write a query in human language speaking and your machine will automatically filter out what it means and then look for patterns and you can get the result back? Or is it more that you have some kind of predefined analytical solutions which I then can tap into?
Michael: It’s a combination of both. So it’s very easy to use. Let me give you an interesting one. Let’s say you work for Dove. Is that a bird? Is that a chocolate? So we might have both of those as customers. So we’ve got to help our system understand which of those Doves you mean. And so that’s a process called brand disambiguation, and we provide you a wizard that helps step you through. We’re going to show you, “Here’s your brand, Dove.” You just type in Dove and we show you, “Okay, here’s words and phrases, hashtags, people, domains that are commonly mentioning your brand, Dove.” Let’s say you were the chocolate. You would quickly see people talking about shampoo, so you’d select shampoo as an exclude and then it updates, and now it’s a little bit cleaner.
One of the trickiest brands we’ve ever had to deal with was All. Another tricky one is Uber because Uber is always throwing up their arms. It’s uber delicious. Whatever. So some of these brands can be really tricky. Now, if you have a really tricky brand, as a service, NetBase can clean it up for you, can disambiguate it for you, and then you can focus on the analysis. So the analysis is fairly predefined but you can take it in any direction you want. What we’ll normally give you is we’ll tell you the sentiment about your brand. So that’s a score we would put from that sentiment. That ranges from -100 to +100. It’s a natural ratio. And we’re able to compute this ratio because the accuracy of the underlying analysis is so good we can even trend it for you. So we can give you very up to date information on the net sentiment so you don’t have to go do that customer satisfaction survey or this can help complement that. So we’re also computing things like passion. There’s a big difference between somebody who just likes Coke and somebody who really loves it. A lot of passion. So we’ll compute the ratio of the strong emotion, whether it’s positive or negative, to the weak. And that’s another metric.
Then we’re also different from our competition in addition to being very accurate on those analytics. We’ll take you deeper. So there’s three things that we’re going to tell you. First, we’re going to tell you what opinions do people have? What specifically do they like or dislike about Coca-Cola? Maybe they like the taste and maybe they hate that it’s harsh. How does it make them feel? So we do this emotion analysis. And it was really Coke that drove us in that direction because Coke, their marketing budget is all about make this product, make people happy, make them love it. So emotional analysis and then behavior analysis. So somebody might be switching phone carriers. That’s a behavior. Or somebody might recommend this particular movie. That’s a behavior. Or they intend to buy. Or they intend to shop. And so that’s a whole metric that we can compute. Net intent to shop, a score from -100 to +100, whether or not people are saying they’re going to shop at the store. So those are the three types of sentiment insights. And then we’ll do hashtag analysis, popular posts, popular authors, cloud score, just all kinds of different things that we can give you about it.
Martin: Michael, how do you nowadays acquire customers? So something like the direct sales force or your partnership network?
Michael: Both. We have sales through direct and then we have partnerships, particularly for international. It’s a great point of leverage. So we support over 40 different languages but most of the U.S. sales presence itself is in the United States and in Europe. So to cover Brazil and Japan and places like China, we’ve got a very, very good Chinese parser because Dr. Wei Li, our principal scientist, he is a Chinese national and NetBase is his third incarnation of a Chinese natural language parser. So we’re very strong in that technology as well. And so we cover these other geographies through partnerships.
Martin: Those are mainly consultants or just sales organizations?
Michael: So they would be maybe like agencies or third party software vendors where they’ll provide additional support on top of what we offer because they speak the native language and that kind of thing.
Martin: What has been the most innovative usage of your platform that you can think of?
Michael: Well, I think probably the most interesting thing that we’ve developed in the last few years in response to customers is this audience marketing idea. So I have maybe a million followers, I’ve got a big brand, and I want to know, “Well, maybe they mention me once, a couple of times a year, but what else are they talking about?” This is an opportunity to get a focus group of a million people and find out right now the types of things that I want to know, like what do they say they want to eat or drink or drive or watch or recommend? All sorts of questions that I would normally have to answer through a focus group that I can only afford to do once a year or maybe once every two years, and it’s got a very small sample. We’re talking millions of people. Very, very nice big sample that you can then take and do things like figure out where should I be advertising and what new products should I be developing? So insights about new product development and innovation.
But with audience 3D, audience marketing, once you have those insights, you can then turn around and target people. So let’s say that we have a customer that is a fast food restaurant and they are putting a new menu on their item, sushi, let’s say. So now you want to drive people who love sushi or who eat sushi in the door. So with NetBase, you can harvest an audience of people who love sushi, eat sushi, all kinds of things that they do with sushi because if I bridge behavior and emotional analysis, we can find those people. And then through advertising of tailored audiences or custom audiences on Twitter and Facebook respectively, you can target them with specific messages where you’ve come up with creative ideas for how to resonate with them based on the audience marketing, and then of course measurement. So this solution which really customers push us in this direction but it turned out better technology was a great fit, it allows them to discover new audiences and opportunities, target audiences with key messages or resonate, and then measure the effectiveness of those campaigns.
Martin: Michael, when I look at the data that you are getting, you said before it’s basically public data, so virtually everybody will have access to the same data. Then my question is what makes you unique? Is it the technology in terms of the NLP analyzers or that you are really giving real time analytics, or is it that you found a cool way or efficient way to reach and acquire customers? What makes you unique and why is it so hard for other competitors to copy that?
Michael: Well, let’s first understand what we’re doing. So for the laymen, what NetBase does for you is we read. So again referring back to a brand like Coca-Cola, they might have thousands of posts per day. All right. Maybe they can hire some people to read that, but can they read the thousands of posts about their 7 to 10 competitors? And this is every day. And on top of that, you’ve got all of the people who are talking about these brands plus the category itself, soda. Now you’re talking millions of posts a day. And on top of that, all of those people, that’s the audience that you want to target to get them to consume your beverage. You need to really enrich your understanding of who that audience is, and to do that, you need to be able to track those million people. What are they talking about every single day? Millions of people. Now you’re talking about billions of conversations. It’s impossible to stay on top of that. So in order to solve this problem, NetBase has developed very accurate analytics that reads for you at a very, very fast pace. And we have a whole bunch of patents pending. We have some patents that have been issued on how to do this at scale and very, very quickly. And none of our competitors are going to be able to replicate it.
And that goes back to our founding moment. We delivered for customer value that we confirmed this is what you want, this is what the customers want, and then that became the basis for our product development, not only for our user interfaces but for our analytics as well. We hired Dr. Wei Li and his team of natural language processing experts, computational linguists who have been taking the results that we analyze for customers manually and they’ve been slowly whittling away at the amount of time that it takes to go through results, just automating more and more of it, as we’ve been doing for 11 years and we can do it now across 40 languages, all this kind of analytics. None of our competition made that investment early, and now it’s too hard for them to catch up.
ADVICE TO ENTREPRENEURS FROM MICHAEL OSOFSKY
Martin: Michael, this is your first company but you have been in the business for 11 years. What have been the major learnings that you can share with other people interested in entrepreneurship?
Michael: Major learnings for entrepreneurship? I’d say one learning that we had was you sometimes need to rethink who you’re targeting, because originally we were an innovation business. We were called Accelovation. We were accelerating innovation. So we took our product out to the market and we tried to find people who were innovating. Now, all companies will tell you, “Oh, we innovate.” But when you ask them, “Who is innovating in your company?” because you need to cold call to find out, very few companies had an innovation person. Now there are a few that have a Chief Innovation Officer but typically that person is running brainstorming competitions and that kind of thing. But who’s actually going and talking to customers, understanding technology landscapes, bridging them together, prototyping, developing new products, sizing markets, championing these ideas? That is often a very diffused process.
So what we had to realize is that too few companies had an innovation function. What we had to do is recognize how companies actually innovate is they have a marketing department and they have an R&D department and there’s lots of rifts between them and there’s lots of changeover, but these are two different departments and we have to have a product for marketers and we have to have a product for R&D people. So we dropped the name Accelovation, we came up with the name NetBase that would appeal to either one, and then we split our product into two but we ended up spinning out this one to Elsevier, and then we focused our own brand just on the marketer.
So that’s one of the takeaways is that sometimes you have to recognize that you may have built this elegant beautiful system but it may assume that it’s going to be used in a certain way that’s not realistic, so you’ve got to adapt and rebrand potentially, resize your market. And we were fortunate. We didn’t have to go like, “Oh, market is this smaller.” We went the opposite way. We said, “Oh, the market is too small. We’re going to broaden it.” So you might have to resize your market and you might have to repackage your product.
Martin: What other types of advice can you give to first time entrepreneurs?
Michael: So other advice? I’d say try to bootstrap your business as much as you can. And bootstrap doesn’t mean reach into your back pocket and pose dollars into your company. What that means is go see how can you deliver value to your customer right now before you have any kind of product? Is there some kind of service that you can offer that’s going to get money in the door, help give you validation so that you can bring on investors at a better valuation for yourself, and give you the learnings so that you can drive your product development process? So as much as possible, try to get cash in the door from the very first day.
Martin: Michael, thank you so much for sharing with us.
Michael: Thank you, Martin.
Martin: So if you are starting a company and you have developed an awesome product, and nobody wants to buy your product, maybe you need to redefine your target audience or maybe even your branding, like what NetBase did. Thank you so much. Great.
Michael: Thank you, Martin.
Martin: Thank you, Michael.
In San Mateo (CA), we meet CEO and Co-Founder of Neo4j, Emil Eifrem. Emil talks about his story how …