How Sigopt CEO Scott Clark grew to $576K revenue and 12 customers in 2017.

this is the top where I interview entrepreneurs to our number one our number two in their industry in terms of revenue or customer base you'll learn how much revenue they're making what their marketing funnel looks like and how many customers they have I'm now at $20,000 per Tov I haven't experienced kids have bent on global domination we just broke on a hundred thousand units Bowl mark and I'm your hood Nathan Laska hello everyone this episode's 705 come over tomorrow morning I talked to Panay he's a 31 year old who has raised 1.3 million dollars to help you be more efficient this question is and you get what he's helping you be more efficient with you have tuned in to find out hello everyone my guest today is Scott Clarke he's the co-founder and CEO of big opt a Y Combinator and and andreessen horowitz backed off of this optimization as a service startup Scott has been applying optimal learning technologies in industry and academia for years he holds a PhD in Applied Mathematics and an MS in computer science from Cornell University and a BS degree in mathematics physics and computational physics from Oregon State University he was chosen as one of Forbes 30 under 30 in 2016 Scott are you ready to take it to the top of course all right man bring it home for us so tell us first what does the company do and how with your revenue model how do you make money yeah so we're optimization as a service we help companies building different complex AI and machine learning pipelines if the most out of them by fine-tuning all the different knobs and levers that sometimes block people from getting peak performance where software is a service company people pay for a subscription to our API based off of the number of models that they tend each week okay mom and give us a general sense of kind of size there with the average customer paying you per month are we talking five bucks or 10,000 bucks yeah great question so our worker pricing that we publish on our website starts at 2,500 dollars a month on getting you a little more than a dozen models a month enterprise clients ramped up from there around the $10,000 a month mark is typical and so what we do to the averages four or five grand yeah somewhere in there so the 10k a month is the enterprise licenses where they start on so that's what we target mostly doing top-down enterprise sales but we have the other tier available for smaller startups kind of ramping into it when you see you're typically doing top gun enterprise sales what do you mean by that yeah so we usually engage at the executive level people who are really excited about using artificial intelligence or some of these more advanced machine learning techniques transform their business but often what they find is there's not a ton of experience in this field distant general it's become rapidly popular and so they're not able to necessarily hire large swaths of people with decades of experience tuning and tweaking and building these models so takeoff is a way to kind of make that just to move faster help them get more of their models and we go from there to the actual practitioners where we can bolt onto their systems and kind of amplify it but Prudential MIT Hotwire these are some of the guys you've won over tell us specifically how Prudential is using you if we can't use them for confidentiality reasons tell us about one of the other ones yeah so Prudential is really investing in machine learning and data science I think we see this across the board in a variety of different insurance companies where some of the more traditional models are being augmented by the amount of new data that's being able to be collected things are getting more sophisticated people are leveraging some of these newer techniques to kind of transform the industry as a part of that as the data gets bigger and bigger as people start to look towards these more sophisticated things like deep learning and artificial intelligence the need for efficiency and the need for kind of the best possible performance only increases and that's where the GOP's is able to bolt onto these strategies that they're developing and make them better and better give me a specific two examples so Prudential is obviously in the insurance business what data set have they given you and said hey throw your guns at this and try and figure something out for us yeah that's a good question so for confidentiality reasons I can't go to too much detail there one way that we differ though in terms of some more traditional machine learning as service companies where they take a dataset and then we throw a bunch of magic at it and then produce some model the way that Zig up works is we'll go into a company that already has something in place we work with different credit card companies on when you can talk about yeah so I can give you a specific example in the credit card base fraud detection so this is a problem that's been around for decades something that credit card companies have struggled with for a long time they have models in place to help solve this so you get a call from somebody and they say hey with this you it's just a real charge something like that there's a machine learning model behind the scenes deciding whether or not to do that obviously they want to minimize fraud but they also want to minimize how often they bug their customers with these false positives as well if it wasn't actually fraud so there's a model in place but even with a lot of domain expertise poured into that and decades of time and energy it's not perfect these companies still lose millions of dollars of fraud every year so what sue God is able to do is provide this optimization service that both on top of these models basically fine-tuning all the different knobs and levers the configuration parameters that make these machine learning models actually work in order to get boosted performance so instead of just taking a raw data set of decades of raw data and giving them some model to like rip and replace what they already have we sit on top of what they have and provide this additive boost by fine-tuning it you can pick a crew up for a car something like yeah so for like a pit that's a great analogy for a pit crew for a car you have to understand how the knobs of different industries in other words I mean I'd be shocked if you had kind of one system you apply it across all industries and if that's not the case you're spending a lot of time on professional services or custom engineering work per new industry client you sign up which one is true so yeah great question so the area that we focus on is what's called as black box optimization which is designed to be this general-purpose optimization framework where you only look at the inputs and the outputs of a system and the interior is this black box that you don't introspect so it turns out the way that people typically tune these systems are things like brute force random configuration search manual tuning turns out humans are pretty bad at doing ten dimensional optimization in their head so what we're able to do is without any domain expertise without making any assumptions about that underlying model I'll perform these standard techniques by providing this ensemble of black box Bayesian optimization strategies behind an API and that's where our domain expertise is got it you know one of the things I always struggle with and someone comes on and mentions a buzz word right like a like artificial intelligence or machine learning is like if my jobs like figure out what's and like where there's actually uh new technology because a lot of people will be like this machine learning but it's like not really machine learning so yeah I mean are you you know I'm giving you out you know Sal the fraud detection the stares the system I'm giving you literally millions of pieces of output data are you you're putting this output data in your kind of black box solution right that kind of works anywhere how do you know that what trends to look for to increase performance for a specific industry yeah great question so once again how we differ from maybe a traditional machine learning as a service company is instead of taking those millions of outputs of just the raw data which fraud what's not what we're doing is we're relying on the domain expertise of the person at that firm to build up let's say a deep learning model the payment the thousands of data points and like actually go through and decide how to classify fraud or not that being said that expert is left with this arduous task of how do i define this model what's the architecture of my neural network what's still learning right what are all these kind of high-level configuration parameters that I have to set in order for this model to be functional at all and typically this is an extremely trial and error based process where the expert could know everything in the world about fraud everything about the context of how they want to apply it within their platform but then they just need to sit there and try a big 10 hidden layer should be twelve thirteen very versatile and they're not a lot of intuition there so what we do is apply this ensemble of Bayesian and global optimization techniques to the problem so that we can efficiently configure this system they can focus on the domain expertise the outputs that we see is just we gave you a configuration to try how well did it perform so not the underlying data not the model but just these higher-level configure parameters so if we think about the data person inside of Prudential as the guy that's building a waterslide and you're the person that's giving them five curved pieces three pieces are ten feet long and he that has to use the domain export she has to use a domain expertise to kind of build the slide the water comes in the water goes out but they're using that domain expertise to build the important parts of their business is that mean is that an appropriate analogy yeah so in this case it might be things where we would like suggest different curvatures and then they would build that up and be able to see well how fun was the ride we're going to do so user surveys or something like that given that I'm going to suggest different curvatures different lengths etc but instead of doing that by kind of brute force trying all different options or manually trying to tweak this item in space we're providing these this best-in-class optimizer in order to do it so you only have to build ten slides instead of ten million in order to get to the best one so let me ask you another well first off I want to keep getting down here what I'm curious on which might not be what my audience is curious about so let me just call chrome real quick have you raised capital if so how much we have about eight point eight million dollars today with the Y Combinator and winter fifteen andreessen horowitz let our seed round immediately following that they also let our Series A last July great so we'll talk more about that in a second I want to go back though to the domain stuff Prudential uses you you hope to discover something that significantly increases their bottom line how do they make sure and how do you make sure that secret sauce now isn't passed on to Geico and an Affleck and everyone else in the industry yeah and that's kind of the joy of this black box optimization approach because we never see their underlying data because the model itself stays proprietary and within their systems all we're doing is fine-tuning these different configuration parameters whatever domain expertise they apply to differentiate themselves from the competitors remain within their system come on god look at me you really there's you figure out a way to really discuss eight yourself with whatever learning they pick up so there's not even a chance of you accidentally disseminating that information to a competitor so as we get more and more customers we can use that information to make our black optimization framework better and better that being said the entire system is designed to be very hands office allows us to work with some of the most secretive algorithmic trading firms in the world where their domain expertise in their models are literally how they make their billions of dollars but they can still use stick ops because all we're tuning are these configuration parameters these thresholds these pull and fast moving window sizes etc but at the end of the day their data and their IP never touches a stick out got it you've raised eight point eight million bucks you went through YC andreessen horowitz company in at the end of 2014 okay 2014 and then bring us forward to today how many customers you have paying you about a dozen customers around the world guys been ok this is very much an enterprise sale then exactly so that enterprise sales software as-a-service working with different fortune 500 and global 2000 companies you mentioned Prudential and Hotwire always another customer we have a handful that we can't talk about but major credit card companies algorithmic trading firms banks etc it's got I mean can I take the twelve customers and multiply times again that average monthly price of around 4 grand assume you guys are doing what some around 50 grand per month right now oh that's a fair assumption yeah ok are you making folks sign annual deals or can they pay monthly we offer both but we definitely prefer the annual deals and try to incentivize that what do you do a 10% discount or something uh yeah exactly that yeah almost exactly that awesome every deal is slightly different that's the joy of being a bras ya know 100 percent of to joy being a startup you can change every deal and then figure out we want to put up on your website right all right okay cool so so that's helpful to understand how are you has anyone started paying you and then stopped any churn yet no that's the nice thing is once customers start putting this into their system to replace it they have to go back to one of these previous techniques like trying to brute force the problem or something like that and not only can we do things faster and cheaper but we also get better results at the end of the day so it tends to be fairly sticky like that and what's your team size today 13 people all in San Francisco we have one remote person in Spain one of the nice things about this is we're able to bring together a lot of the world's experts in this so this is someone who's a professor and has ten years of experience in this field but everybody else is in San Francisco and why did you raise the money where are you spending most of it so building the team that was really important to us also the enterprise sales effort is time-consuming and expensive in itself so we wanted to make sure that we could fully invest in that give this the best job we possibly could of the 13 how many folks are sales related whether it's account managers or inside sales reps or STRs about a third if you include myself okay about three or four and then and then what are you you're probably still trying to figure this out but what's your gut tell you right now that you're spending to acquire one new customer well that's a great question hard to tell because we have so many different things in flight right now and the pipeline is kind of very dynamic but we have three or four sales people trying to close deals so are they flying are you have to make in-person sales typically or no we do visit our customers I was out in New York the week before that Houston the week before that so we do some of that as part of kind of enterprise sales necessitated but the joy is once we get people up and running the software is a service kind of takes over and it's a very plug-and-play easy to use yep that's great now do you I mean when you raise an increase in lead I mean did they ask you questions about you know L to be in a lifetime but I have a customer relative to what you're paying to acquire them and how you're going to manage that cash gap so one of the things that they really like to do is make these big bets on these new technologies that are highly differentiated from what people do today so I think a lot of what they've been able to do is at the seed round it was very speculative a lot of investing and just kind of the idea and that the founding team by seeing how we performed over the year and a half or so before they've been led that series a and by seeing how customers were responding to us and being a put back channel they were able to see that this is definitely a large market that were well poised to be to capture so I think a lot of it still is being really excited about a new kind of transformative idea and the goal with the series any money is to make this repeatable business process out of it is that a sexy way of staying mating of a about LTV to CAC ratio they just want to be an exciting space and be in the hype um I mean they do incredible diligence oh they definitely care about everything and there are the smartest people that we've ever been at the fortune to work with that is the right that is the right answer you must be trying to get a series B lead right now that's what's happening I can feel it they're not actively raising money right now that's anim about the AG gave us a nice cushion and we're building a very sustainable business like when was that that was in July last year okay got it so yeah bout about a year you're napping and most of your I mean I'm just doing back and and Matthew with a team of 13 again if you're paying in San Fran and conservative salary of called eight grand per month right that comes up to about 100 grand per year and then add on you know 20 percent on top of that is about ten grand per month across thirteen folks you know your call it headcount as 130 something like that that's the majority of your expenses right each month headcount definitely is a majority yeah many of you know I am buying companies that I really really like and there's no quicker way for me to get to the bottom of what is happening on that website and using this tool called NATO Makkah comm forward slash hot jar hoc jay-ar it basically will give me a recording okay when anybody lands on the website or give me a recording of where that we were scrolling and obviously doesn't basic stuff like heat maps too but I learned so much about where the users are scrolling and clicking on my site using that tool it helps me increase conversion rate make more money and grow the businesses faster and we'll have to see what happens with those businesses but I'm buying them on buying them very quick and I'm using Nathan lock accom ford slash hot jar for all of my website analytics you can't do I work with them it's totally free you go to Nathan latke dot-com ford slash hot jar no credit card required again use it as much as you want Nathan Wonka calm for Josh hot jar I'll see you there awesome and let's wrap up here with the famous five number one what's your favorite business book I really like the hard thing about hard things buy up and Horowitz I read it but we started working with them and reread it sense and it was a lot of good insights there number two is their CEO you're following or studying when I sing about a 16 D is that they're all CEO operators of the all the general partners are so I learn a ton from them I also recently read shoe dog by so might and found that really insightful de is resilience is motivating number three is their favorite online tool you have like acuity scheduling um I mean I use Gmail every minute of every day read it's probably my actual favorite website though but we can own tools wise slack probably number number four how many hours think you get every night I try to get a sometimes it's hard with the jet lag in the travel but I think sustainability is an important part doing this for the long run and what's your current situation married single yup kids married no kids no kids okay good and how old are you I'm thirty years old I had a thing about that one last last question sky take us back ten years what he was your 20 year old self knew but it doesn't get easier so set up habits and processes to make things sustainable when you have the time and ability to do so because that'll definitely help you once things continue to kind of wrap up there you guys have it from Scott founder of cig off back in 2014 he since gone through Y Combinator had andreessen horowitz lead his first round eight point eight million raised team of thirteen again making it easier for these large companies with big data sets actually makes sense of their data right inputs/outputs counting up twelve clients right now paying on average four grand per month for somewhere around 50 grand per month and monthly recurring revenue team mainly based in san francisco god thank you for taking us to the top thank you if you enjoyed today's episode with scott go back and listen to Eric yesterday he created his products because it was necessary to save his mom from dying literally it's a brand new health tech product that recognizes beaters it would mean the world to me if you guys got any value from this episode if you would go leave a review on iTunes right now and then subscribe you know I have to like have to get these episodes every freaking day for you guys and struck me I love it I would do with no listeners but boy oh boy makes my day and make my King Day when we see great reviews and get your feedback so thanks so much okay so I'll Drive I love giving away free money I feel like for giving away cars I have something special for you today how many of you have heard our super short guest talked about success they've had with Facebook and Google Apps well all of you listening right now if you're listening you get a hundred dollars in free AdWords here's how you get it thanks for listening it's a free hundred dollars from Google right when you sign up with my website closed provider Hostgator go sign up now to get your free money hostgator.com forward slash Nathan again that's hostgator.com forward slash Nathan