Scale By The Bay 2021 : Alyona Medelyan, Keynote: How to start a startup for AI Engineers

Scale By The Bay 2021 : Alyona Medelyan, Keynote: How to start a startup for AI Engineers

Recording: Scale By The Bay 2021 : Alyona Medelyan, Keynote: How to start a startup for AI Engineers

i would like to talk to you about how to start a startup um for if you are an ai engineer and i presume many of you here are and before i start i would love to connect with you on linkedin and twitter as alexa mentioned i do post things regularly and i but uh most importantly i would love to hear any feedback or questions that you have after this talk cool let's get started what do you think these companies have in common well first of all they all use ai as part of their solution and second they also employ some of the most amazing ai engineers in the world because they can but it turns out that they also buy some great ai solutions in fact i can i can't tell you and just due to um contractual obligation i can't tell you exactly which ones but i can tell you that four out of five of these have paid significant amounts for thematic the company the software that um my company makes and this is interesting because when i started thematic lots of people said well why why why bother google or amazon or ibm already have developed nlp apis and it's really powerful they can do what your solution does but it turns out that if you solve real customer problem um companies will buy from you so why should you care because i think you should start as well and an ai startup so what's an ai startup let's talk about this first [Music] when i say ai i mean machine learning deep learning natural language processing all this technology that that we are working with but since it's a powerpoint presentation um i will keep using ai to refer to it in my career i've learned that there are four different types of ai startups the first one is when ai is part is the core offering that's being offered so for example primer or hugging phase some of the recent startups that are very successful in this field the second type of ai startup where you need to use ai to solve a specific problem but um customers don't even care that ai is used to solve it it's really about the problem and not about the technology that's used to solve it thematic falls into this problem into sorry into this category and many other startups do and these two first types are the ones that i'll i will be focusing today uh so when i say why you should start an ai startup it would be for these first two types there is also startups that um solve an existing problem um it doesn't require ai but they kind of edit after um when they need to fundraise and they want to to look better or when they expand their offering and these are probably the least ai startups out of the four and finally the fourth type is where you basically have a group of phd students who are joining together working on cool stuff and uh most likely they will be acquired as a team by one of the giant companies and the reason why this happens is because it's very difficult to hire ai engineers and it's especially harder to find a team that works well together and ai researchers take advantage of this no most notably george uh jeff hinton when he sold his barely formed ai startup of three students for 40 million dollars to google if you haven't read this article it's such a fascinating story an amazing article i hardly encourage you to read it so a bit of background on on myself i have as alexey mentioned i have um computer science uh phd from waikato university in new zealand which is where i'm now i also did a master's in natural language processing always loved using natural language processing to solve real-world problems and it really motivates me to um to work with stigma with this technology so instead of staying in academia i joined i left after i graduated i left and joined a startup that raised 10 million dollars called finger unfortunately they failed to get an attraction so i left on my maternity leave and i was kind of cooking up what should i do next and to be honest i've never even considered that i will ever start an ai startup that would compete on the global stage this is me here on the right wearing ugly maternity pants pushing the pram and i was listening to podcasts all the time and one of these podcasts was y combinator podcasts and little did i know that a few years later my co-founder who is also my husband nathan and i we were invited to apply and were accepted into y combinator and we suddenly realized that what we're building can become a billion dollar startup well others believe we couldn't so we did as well so you know with a bit of hard hard work and a bit of luck so fast forward a few years and what we do is we use ai to analyze customer feedback that companies constantly are collecting through a variety of channels support tickets chats surveys reviews and we help them find answers in this data we aren't yet a billion dollar company but i'm so proud of what we've achieved of the types of companies we work earlier this year one of the oldest companies in nlp called clarabridge was acquired for one billion dollars so there's clearly demand for what we do so why should you start an ai startup um well first of all if an immigrant mom with two young kids on the other side of the world can do it then you can do it too here's what motivated me first of all it's not money you can earn significantly more if you work with you know at a one of the well-funded startups or one of those big fives but it's about learning learning new skills i realized that there are skills that i'm really good at and enjoy doing that unrelated to to ai and and that was very cool personal growth is huge so uh learning to learning to be a leader motivated team learning empathy and how other people think you just discover so much about yourself and and as a person as well but most importantly it's about impact and impact for me is basically seeing people use the technologies that i develop when you're an ai engineer and you start a startup you truly and start understanding what technology is actually needed in the world and how it fills the gap and what's important and you're in charge on making sure that it happens which is really exciting so um you might say that if you do research at google you will also get impact absolutely you might for me it was slightly different um i did a four-month internship at google and i was um implementing my phd work within their code base when somebody said hey um can you make it work in chinese our google chinese blog search is on the rise and we it could really be used as a feature in our product it was a long time ago obviously the blog search isn't a product anymore but um i was able to actually do it by using google's amazing nlp and infrastructure suddenly my code could understand blog posts and languages i don't speak including chinese but ultimately this feature never saw the light then when i joined somebody else's startup finger they raised tens of millions of dollars they even bought another company they had 50 staff but ultimately the product got very little traction and they run out of money everything that my team developed was sold to another to new management to start a new product from scratch again i felt like i couldn't get the impact compare this to thematic when before we even built the product and barely registered the company there were three customers who were paying us for the prototype which was a single python file plus a powerpoint presentation ins it really felt like instant impact so if similar things motivate you then i think you should really consider starting a startup why now well there wasn't ever a better time you can right now you can use a pre-seed on just the idea alone just having amazing credential and an idea um will help you there are giant seed funds announced by all the major vc firms um for this you need a little bit of customer validation some interest but not much again um and you can raise millions um if you don't want to do it that's totally fine as well we we didn't we were bootstrapping in the beginning and bootstrapping right now it's even easier so much advice out there lots of people are building in the open posting us their advice is they go on twitter and their podcasts and the but the best part is the new change that covet brought that kind of leveled the field of how people sell um and and buy and take meetings um you can really reach out to anybody in the world and see if you can validate your idea with them so you can start a startup from anywhere in the world so how do you start a startup that solves a real business problem using ai so kind of focusing on this there are three key steps in my experience the first one make sure you solve a real business problem second learn to sell learn to market and three focus relentlessly on finding product market fit and i'll explain each one of these in detail the first one is probably the hardest if you are an ai engineer because it's so easy to build cool technology but what you really need is a hair on fire kind of situation so michael seibo is the was our mentor and the c he's also the ceo y combinator and i loved the analogy that he told us so imagine you're in a room and it's an empty room there's no water nothing um no furniture just an empty room and your hair is literally on fire suddenly you notice this brick lying on the floor you grab it and that's the only thing here so you start hitting yourself on the head trying to put out this fire it kind of works um you still use it there and um so the types of pro the type of problem that you need to find is is the one where people are using bricks it doesn't matter if you have ai or or don't but you if you solve that problem itself with something better than a brick they will buy from you second you need to learn how to sell and learning to sell is basically understanding um you know at first you you might think sales people are cringy why you know why should i even um consider this i don't want to do selling um but um in what happened to me was that i had two friends who both had computer science degrees who worked themselves so i respected them so when i had the situation where i developed um first the first thing that i tried to commercialize was my phd research and it just felt like it wasn't going anywhere what was happening um we had a website and a cool demo we had enquiries one day even nato has contacted us and said we're interested in buying your technology but ultimately nobody was buying so when my friends told me why don't you do a sales course i really listened to them and it really changed everything before so sales felt like magic or something cringy but in reality it's all about asking the right kinds of questions so for example what questions do you need to ask to understand whether it's a hair on fire kind of problem this question is when do you need this by if they say sometimes next year or next quarter it's not a hair and fire problem if they say oh ideally yesterday then it is a clear and fire kind of problem the second part is kind of leading this uh this person down the sales funnel or in other words figuring out whether your solution that you develop is the right one for this problem and again it's all about the right kinds of questions once you have a demo you realize that it's a problem we're solving you build a demo you show it to them you ask when would you like to see a proposal or would you like to see a proposal and just knowing these two questions really changed everything for from me we were able to find the problem and validate that we have the right solution the third part is audience and this is um a lot of people overlook this and it's really about how others will learn that your solution exists so you can use different methods to build your audience you can write content talk about the problem that you're solving um there are speak at conferences reach out to people by email partnerships you can do ads or work with influencers or become an influencer yourself the moment you know that you found a problem worth solving and you have a solution you at the same time you need to start building the audience now the third step is focus on finding product market fit and again it's really um difficult for ai engineers because it's so easy to get distracted with cool technology cool new stuff that's constantly being invented in this amazing field um but product market fit is the most important thing if you want us to succeed at starting an ai startup so what is product market fit it's not about paying customers it's not about profitability it's not about raising lots of money series a b usc in reality it's about having a product that solves a real customer need at a price that that's fair and customers are beating your door to buy that your that product another way these three components you need to find these perfect relationships between them um i'll explain you how it feels when you don't have a product market fit this is when everything feels hard especially finding customers um it's like pushing a boater up the hill this is what it feels like once you reach it it's about it's a bit like a boater is rolling down the hill you are inside that bowl and you're trying to assemble things quick bits and pieces quickly catch your breath basically i realized it's like absorbing which is a fun activity we do here on weekends in new zealand which is the picture on the right so as an ai engineer you basically always have to ask yourself what's in what's in it for the customer will it help us find product market fit so when i was preparing this presentation i shared shared it with an ai engineer at thematic rob dumbleton and i was worried that now he will start a starter but he said um you know what you need to add a slide on why you shouldn't start a startup so here's the slide um you can see the little poop emoji on a plate well basically being a founder is really really hard and another founder once told me that it's a bit like somebody is bringing your on a plate every single day and you have to eat it this is just a selection of all the tough things that i had to have to deal with the number one is hearing no from customers from investors from employees you're basically hearing your your baby is ugly a million times over and over and you really have to have thick skin and have your feet on the ground the selling is basically hearing no over and over when your customers leave not because you know because of mistakes you've made it's painful but it's even more painful when some of your best employees move move on to other things it's heartbreaking um the other thing is that you all you are as a founder always have to maintain the morale for the rest of the company so you no matter what's happening and all of the happens ups and downs you have to be the upbeat one the happy one the encouraging one so even as i was writing these i felt very lonely so this is your counter argument if you you know might might be on the fence so ultimately i want to share just a few words about how all this advice translated into a thematic journey so when we started the problem that we were solving is helping companies understand what's really driving customer loyalty and customer loyalty was has been measured using the symmetry called nps and we basically tied the feedback to this metric and created visuals that understand what's impacting nbs and we still do that and there's still lots of value in that um the our solution used unsupervised or self self or semi supervised discovery of themes in in customer feedback and a visualization platform similar like to tableau but for unstructured data and we were building our audience straight away by writing content and also going to lots of customer experience conference where these people hang out and very quickly we started getting traction but then what started happening is that nps really wasn't um is not as in passion anymore especially now and also because of kovit we couldn't travel to conferences anymore and in general we realized that penetrating this market is very very difficult there are there are a lot of solutions and um a lot of a lot of players and we decided that we needed to pivot so today we hope we still analyze feedback but our main proposition is to help product teams prioritize their roadmap by helping them understand all of the qualitative feedback that comes through chat support tickets all the different feedback channels and the types of solution that we develop also has had to change so now we have to do chat analytics which is different from survey analytics or we need a special theme summarization that doesn't aggregate the what's happening in the feedback as much but is more specific and more precise which is important for product managers and of course travel is really isn't really possible although it's opening up again so we also needed to adjust our um approach to building an audience so as you can see finding product market fit is not done and dusted you are continuously trying to evolve your company your approach so if you start an ai startup you will never run out of opportunities to learn in conclusion if you're considering doing it consider all the options but do it it's easier than ever to start today and you will learn so much even though it will be tough the three things that i'd like you to remember is find a problem that feels like hair on fire type of problem where people are using a brick don't forget to build your audience from day one no matter how choose a choose your approach become an influencer twitter um write really good quality content on medium whatever that is find find your thing and continuously focus on finding product market fake fit and if you do that your impact will be significant well i hope it was useful and i will jump over to space spacial chat to answer any questions but thanks for having me alana before i go there uh there is a question from the audience uh what is worse what is better uh customers paying for your product and not using it or customers using your product and not paying for it what a great question um i think it's worse when they pay for your product but don't use it if they use you know if they use it but don't pay for it you can try all the different things to um to kind of gauge that they would what will make them pay how much and if they don't then there might be other ways to monetize the solution right and people don't use for twitter but they're still making money um where is the other situation where they pay but not use it you basically have a pro a product that you sold somebody bought because your pitch was great they've considered um using it but they don't so there's the need isn't there so you aren't actually solving a real problem got it got it i agree i agree and one more question for me uh so i know you made several moves you you know you really kind of happened to be there in the middle of this global pandemic and obviously you're you move back to new zealand which has its own story right like a fortress against covey so i just wonder if you have any insights obviously you have been you know you've been a remote company then like you moved to silicon valley to be kind of a regular probably you know uh centrally located company and then you were kind of forced by the circumstances to again disperse so any lessons from this any observations how you guys survived it what made it easy what made it better like and again of course people used to think that they need to come to silicon valley and do this and you kind of you know follow this to an extent uh but now you kind of do it from around the world any observations i think there are two elements to this one is um the company and how your team works and the second one is access to customers especially when you're early on having people you can talk to in terms of you know customers getting feedback from working closely uh is really important so for us being in in silicon valley where there are a lot of customers a lot of potential customers and and being in the us and being able to travel to all of these conferences throughout the us um was was important and we weren't planning to go back to new zealand anytime soon um and i'm actually traveling to us next month because i still feel like it's very important to kind of be in the us now that we we have that and contact with customers is really is really critical but um in terms of the team that's it's more important that you meet them at some point um we were lucky to build some relationship when we had an actual physical office in san francisco we also did team retreats where everybody from the team would meet for a week we met up in hawaii so half of the team from new zealand flew in and the us team flew to hawaii so that was amazing to kind of build those relationships and then it's easy to work remotely we can we can't do it now as easily so where we do we connect in other ways so for example we have friday sessions where each person is running it and it's not about work it's about something fun we had a tarot reading yesterday in one of the sessions and we also do we also celebrate by basically um once we hit our goals we give people time off and so everybody celebrates in their own way but then shares on slack how they celebrate so like it's very difficult to connect on a personal level as a remote team so that's the most important learning i would say you