 The queue presents On the Ground. Here's your host, Jeff Rick. Hi, Jeff Rick here with the queue. We're On the Ground in New York City, a really special edition of Our On the Ground. We're here at the Lower East Side on the edge of Chinatown, a lot of action, and really excited to come to Fast Forward Labs. We see Hillary Mason all the time at the show, so to actually come out to Fast Forward and get a feel for what's going on here on the ground in New York. So we're joined by our next guest, Catherine Hume, the Director of Sales and Marketing, welcome. Thanks, Jeff. Glad to be here. So you're like out on the field. You're dealing with customers in marketing, big data. So what's kind of the vibe? There's a lot of excitement obviously in the tech world about big data, machine learning, but what's happening on the customer side? What do you see? Well, we're seeing all sorts of things. So at Fast Forward Labs, we work with both emerging very small startups that are trying to build products that have machine intelligence or data science oriented capabilities. And then we work with really large Fortune 500 enterprises who have sometimes been in business for hundreds of years and have tons of data that they've collected and are trying to figure out how they can leverage new machine learning products to make use of that and continue to grow in the future. So on the small side, there's a lot of popularity right now for virtual assistants that happen to combine a lot of the capabilities that we have focused our expertise at Fast Forward Labs on, so one of them is language generation, right? So thinking about having a set of data and then representing that data and making it easily understandable by people who aren't data scientists who aren't analysts and can orient and understand insights in just plain old human language. So we'll partner with these startups and they say we've got a great idea to have a conversational interface with the product we're developing for consumers, but we're just looking for some help in building out a feature that uses these capabilities. So we'll step in and as technical advisors we collaborate with their development team and help them realize those capabilities. And from a financial perspective, it's often a trade-off. They've gone through series A, it's a little bit of seed funding sometimes up to series B, and they decide whether or not they want to hire a new developer or they want to work with us to leverage some of the more emerging machine learning expertise to build their products. So they want to build the next series or they want to build a different way to communicate than say a tableau visualization or what do you mean exactly? What are they trying to do? So there's actually a huge range out there. It's a really exciting time, one resource to look at. There's an investor at Bloomberg Beta named Siobhan Zillis who has this whole landscape of the machine intelligence world right now. And she has created these little categories that describe things like synthesizers who take lots of data and put it all together in a way that can become actionable. Or different advisors say where we work with one of our startups who's building an assistant for sales and marketing teams who can just go on and look through the calendar and automatically schedule meetings with external participants on one's behalf. We have others, we're big users of Slack here as our internal communication chat channel and there's a company we're advising who's building a bot who you can ask questions to when you're a new employee in a company and say I'm trying to build out this analytics tool, this analysis of my data in Salesforce, not quite sure how to use a reporting capacity and it's like a help tool on steroids that automatically provides you your answer and can even go in and read articles on the internet and suggest tweets for marketers so that they're just automatically generated tweets related to longer documents on the web. And is it pulling data directly out of Salesforce? So that's the other thing that's different, right? That the API economy is so interesting that you can pull all these disparate data sources and really apply your own secret sauce, your own kind of new method or mode of an application by using all this stuff that's already out there. Yeah, absolutely. And I think from sort of an investment perspective where there's value, it's just adding on an additional layer of value, untapped value to some of the existing either data sets or existing applications that already have analytics on them. One example in the natural language generation space, in one of our, actually our first report, we talked about systems that can render data more accessible in natural language. There is a recent integration that a company called Narrative Science out of Chicago has made with the data visualization platform CLIC where the sort of the first generation of making data more accessible and democratized to non-analysts and companies was all visualizations, right? So let's turn it into pictures and let's render all of this information insights. And then the next stage is, okay, let's go beyond pictures and tell stories, right? So there's some people who are visual, some people who are more analytic and just that presentation can be modified to meet different user thought patterns and needs. Yeah, so Catherine held up the book, we're teasing her. So, but this is interesting. I mean, this is part of what the fast forward delivery is and these are not yet available on Amazon, but hopefully you guys are gonna change your business model. You should either sell them or I guess they're available if you're a client, right? You get them part of the subscription, but these are pretty interesting and in depth, really reviews of ways to do these things that they cover on the quarterly basis, whether it's the, what's this from the natural language generation or the picture recognition or the realistic stream. So these are really high value books you have to talk to fast forward to get one. But so that's good. So now let's shift gears to the big companies. So what's kind of the conversation that you have with these big companies that know they need to do something? They probably have different things all over the place. Yeah, so I really, they sort of fall into two categories. So there's the one category that has heard a lot of buzz about machine learning data. There has been a higher up executive who is interested in making a strategic shift to try to use data more effectively. But up to this point, they don't really have a research and development resources internally. Analytics, if it does exist, has existed with under the CFO. It's focused on BI. It's focused on analysis of transactional data for P and L reporting, et cetera. But they haven't really merged out into big data web analytics, et cetera. So in those instances, they'll call us and they'll say, we're keen to learn about this. Can you help us transfer your knowledge on what's emerging, what's recently possible? Give us a continuous stream of new ideas so that we can identify an ability to build a new product and shift around our data strategy. The other type of customer we work with do have advanced research and development teams, large financial services companies, consumer package goods, just some big Fortune 500 companies. And there, we're just an extension of their R&D team. They may have some contact with academics, but it takes a long time to read through all of the new papers that are coming out as this landscape changes so quickly. And they know that they can trust us as from an applied perspective to really give them insights that could take off and shorten their ability to use a new technique from a year's time to two months time. Yeah, it's interesting because you guys are pretty close to the financial district here in Manhattan. And obviously financial services, arguably, is really a data business, right? They don't have any money anymore, right? It's just numbers on a page when we trade money and it continues to advance and grow. So those guys and also a little differences, little shapes of portions of points here and there can have a huge impact. One of the things we see though, consistently, both in the large banks and then with the hedge funds in the area is that they're struggling right now to really understand what big data can do for their business. They all have a hunch that they can understand it and should understand it. They are interested in merging analysis, what used to be qualitative analysis. So hiring on some recent PhDs or recent undergrads that are gonna do research on a company and then use that for their capital market strategy. They really wanna incorporate analysis of Twitter or other social media vehicles into their trade strategy, but are still struggling to figure out how that might work. So they bring us in to parse a lot of the rhetoric that they hear from other vendors in the space to plan out how do we build the infrastructure, what's possible, what might this lead to in the future and actually do something that's meaningful as opposed to just data for data's sake. Right, right. And what about kind of the dark side of machine learning that people are afraid of? Basically, taking away everybody's jobs and you talk about virtual assistants and it seems like there's almost no limit where things can be automated and moved to a machine versus the flip side of the coin is having more people kind of empowered by the database decision making tools who can actually do a better job and continue to have still that contextually sensitive point of view that the machines don't necessarily have. It's just different. What do you see in the field and how is that kind of debate shaping up? There's really two totally different trains of thought on this question. So the one that's getting a lot of press because it pushes on our fear buttons and it's really spectacular is the one where a machine suddenly becomes smarter than we are and take over and exterminate the human race. And we of course respect that that's occurring. There's a lot of smart people to think that way but it's not really where we think our attention should be directed today. In contrast, there's a bunch of people that are trying to think about, A, from an employment perspective, to what extent do these systems actually replace middle class high skilled workers? And our opinion is that it's not gonna happen tomorrow and we try to encourage our customers to ask more practical questions like, for what types of processes internally are we willing to accept systems that give probabilistic results? So one of the things in using statistical data technologies is that unlike computer programming 15 years ago, these are not deterministic systems. They start with the data, they build models and those models will be improved over time. So there's confidence rates and whether or not the classification say is accurate and they're getting higher and higher and deep learning as an example as we did in the image object recognition reports has some fantastic success rates in terms of accuracy but it's still not exactly perfect. So the question becomes which business processes can I trust this, can I trust automating and just let the system do its job and for which other ones can we set up different workflows where system does the first pass, we look at those results, we then refine them and push some information back to the tech team so that they can develop them and that's a whole different business process than automation of a workflow using a step-by-step deterministic system. So we think that those questions are actually more important from the employment perspective and just getting the systems to work because it's really his new stuff and there's a lot of work to do. So before we think about the robots taking over, let's actually make robots that work. Right, it's funny you say that because we had Dr. Michael Jordan on from Berkeley at a GE event and he talked about the difference between computer science and data science and the big one being probability and confidence level. Totally. Traditional computer science, right, it's A or B, it's on or off, it's one or zero where in real data science there's always a confidence factor, there's always a probability but I want to shift gears one more time. You mentioned something off here, that was interesting. Obviously academic institutions are very involved in the development of data science and companies are very involved in the development of data science but then you mentioned you're out talking to artists which I didn't expect for you to say. So where do artists fit in this puzzle, in this development? Totally, so first just to talk about our company. So we were unique, we all have sort of cross-disciplinary backgrounds. I did a doctorate in comparative literature. We have a couple of people on the team that had done graduate work in physics. So we like weird stuff and we try to explore things that are on the fringe and use that to influence creative applications in companies. So in doing that work, there's a couple of artists based here in New York City, a guy named Kyle McDonald and another named Jean Kogan who are very actively using new deep learning techniques, the stuff that's used to identify objects and pictures to, as an example, do what they call style transfers. So you take the Mona Lisa, you take Van Gogh's Starry Night, these famous paintings that you can really recognize as, oh yeah, that was Da Vinci and that was Van Gogh, and then they take a picture, it could be a picture of you, Jeff, and they re-envision your photo as if Van Gogh painted it or as if it were a Roman fresco. And these are ways in which, because these artists are not like companies that are trying to make money, determined by the commercial needs and the market needs that end up shaping how products developed or not like the academics that are sort of working to find out the theoretical applicability and generalizability of a given technique, they get to go and do applied research without any constraints and end up doing a great job just pushing the boundaries of a technique and also showing its limitations, which for us is immensely valuable when we go to consult with our customers. Because you made interesting comment that once you're in a commercial application or even an academic application that's tied to some other objective, you start making choices, you start working down a path even though at the beginning of the path, you start at point zero, you start having to make decisions and you start picking whether to go left or whether to go right. And the artists don't have that really kind of limitation, their ability to explore and continue to explore and change directions and back up. It's very different and opens up lots of different kind of new discovery. Well, it also for these kind of new emerging technologies, we do a lot of work with our large clients on should I build this technique or should I buy an existing vendor solution. And it's precisely, when anyone in product management knows this, there's gonna be a trade-off when you're deciding, you can't do everything, have to pick which feature you're gonna focus on for your next dev sprint. And that feature should hopefully, if the product marketing and engineering team are tightly aligned, end up serving as large of a customer base as possible so that you can make revenue numbers. And in doing so often, we saw this with the natural language generation as we've studied the trajectory of the technology over the last year. The trend right now is towards what they call self-service tools. So it's a lot of rules that are baked in and it's sort of removing the creativity and data science and making it just more rules that will link together the way in which someone wants to phrase something in some sort of relationship in a data set. So then for our clients, the questions become, all right, what are you gonna use this for? And do you wanna have the ability to have the scale and be flexible in the future? In that case, it might be better to go for, to build it internally so you can keep that creativity and take a little more time, but you can really customize it or do you wanna just go for the sort of more plug and play solution? Yeah. So last question, you're out in the field just talking to people, what's kind of your most common objection that you're running into that people that aren't familiar with that's more labs or really just aren't that familiar with the space or they kind of heard about it? What is kind of the most common objection and how are you answering that? I actually think the most common objection is we have a pretty unique business model. So it is something that we, it's a subscription model. You get access to the research and then you get four hours of on retainer consulting in addition to that and we'll scope out projects, but customers are used to buying, they have different buying patterns and they allocate budget per different departments. So they're used to buying either consulting or research. And when they do consulting, they wanna do a scope of work, think about the project, is it time and materials, it's a, you know, how things are scoped out and we have to do a little bit of education in the beginning of the sales cycle and rendering them comfortable with the fact that we're not a consulting shop, we're not a forester or a gardener, we're somewhere in between. And they just, it takes a little bit of time for them to accept that and once they do, it's normally okay. And you know, we talk with the lawyers and we say, you can license our content, you can use it internally, but that doesn't mean you can send it externally. So it's sort of this, it's the challenge associated with the new business model. But that's really business mechanics. It's not really an objection to data science. It's not an objection to, when you said that we have a unique business model, I thought you meant that the customer across the table was saying we have a unique business model, of course, you know, no, the stuff can apply to us, which of course, everybody had things, they have a unique business model until you dig under the covers and it's mostly the same as everybody else, maybe a little tweak here and there. Yeah, I find that, I think that's, I mean, in terms of the actual content, so of course, you know, we're, our product is most naturally aligned with early adopters. There's a lot of pragmatists that exist in the large companies that we serve. So there is the, you know, trying to find the person who really is excited about what's possible as opposed to the person who wants a time-tested, non-disruptive, controlled, comfortable solution that they can easily integrate into their existing processes. We're gonna turn into petroleum pretty soon, those people. Yeah. All right, Catherine. Well, thank you for taking a few minutes with us and hosting us here at Fast Forward Live. Yeah, I'm delighted to have you. Absolutely. Jeff Frick here, we're on the ground at Fast Forward Labs in Manhattan, New York City. Thanks for watching, we'll catch you next time.