May 23, 2018
60 word summary: I spent two and half years trying to start a startup I thought might do lots of good. It failed. I explain what happened, how it went wrong and try to set out some relevant lessons for others. Main lesson: be prepared for the fact you might find the experience mostly stressful and have nothing particularlly useful come of it.
What was Hippo
How did we get started? How did things fall apart?
What would I do differently?
Potential lessons for aspirant effective entrepreneurs
It’s often said EAs should be trying new things that probably won’t work but will pay off dramatically if they do. If we’re not failing, then we’re not trying enough new stuff. One thing that’s been mentioned quite a few times, although less so recently, is altruistic entrepreneurship: starting companies as a high-risk, high-reward strategy: you might build something that solves an important problem and acquire oodles of cash you can give away afterwards. I tried this and it didn’t really work out for me, so I thought I’d share the story. I’ve been meaning to write this up for a while – I called time on Hippo back in October 2017 – and seeing Joey Savoie’s rather optimistic post on the value of starting charities has prompted me to finally add my own less optimistic data point about starting companies.
Hippo is/was a happiness tracker-training app, which I used to try and describe as a ‘FitBit for the mind’. The idea was you could track what you were doing, thinking and feeling, get a better understanding of how your life was going and use that to change how you live. The way it worked – or rather, was supposed to work – was that it would randomly ping you in the day, you’d enter your data, the app would display your data and you could learn what was driving your emotions. We never got as far as selling it, but we were considering our potential markets as (1) consumers, particularly the sort of people into mindfulness and the quantified-self (2) an emotional support/employee satisfaction tool for companies (3) a mental health monitoring/self-care app paid for by health providers.
If you have an android phone, you can still download it.
I had grander plans than tracking happiness. The goal was to build an app that would ‘solve’ happiness. The idea was to create something that would monitor how you were feeling over time, let you understand that better, then analyse your data to give you the support, guidance and tools you needed when you needed them. In practice, this would mean combining subjective data (i.e. self reports) with objective data (location, movement, etc.) to learn what was associated with your mood scores, and then give you flexible, tailored guidance.
As an example of how the guidance might work, suppose you reported an unusually low happiness score. Hippo might then suggest you go for a walk, watch a TED talk, practise mindfulness, write down your greatest achievement and how you achieved it, phone a friend, etc and would try to intelligently offer the best option. I’d compiled a list of 40-odd evidence-based activities and re-thinking exercises, i.e. behavioural and cognitive interventions, that I’d come across during my research, and hoped we could use machine learning to give people ever more useful recommendations over time. From speaking to hundreds of people, I’d found there was a huge gap between the things that were shown to increase mood/support mental health and what people either knew about or did.
As I saw it, there was a neat and obvious way technology could help people live happier lives that no one else was trying. It was perhaps, in tech jargon, a 'moonshot': an ambitious, unlikely-to-succeed attempt to use new technology to solve a huge problem.
I became sad. This gave me the idea for the app. I found someone to help me build it. We worked on it part-time for 2 years. We never quite built anything that worked. Eventually everyone left.
I got the idea for Hippo back in Spring 2015. I was working on another business idea, which also never got off the ground (a happiness-behavioural science consultancy). I felt I was generally failing at life when I suddenly, unexpectedly discovered I was very depressed. At the time I looked round for an app to help me - this was back in 2015, when apps were new and the sexy solution to everything - and couldn’t find anything. I threw a bunch of things at my depression - antidepressants, CBT, mindfulness, gratitude journaling, etc. - which were, in my case, really successful. After I recovered I was left with the abiding frustrating that the technology that would have saved me could exist, it just didn’t exist yet. Armed with my own experiences and my utilitarian hat firmly on my head, I started thinking seriously about how to create a happiness app which could, in theory, end up spreading around the world and changing lives. It seemed like the most good I could be trying to do.
The early days (March to September 2015) are now a bit too long ago to remember. I started talking to lots of people about it to gather feedback, Lean Start-Up style. Through a friend’s introduction, I spoke to a major high street bank. They said they were interested in running a trial; the idea was we’d using anonymised mood tracking on employees then (1) provide a ‘temperature check’ of employee satisfaction and (2) analyse the data to tell them what they should do to have happier staff (and less absenteeism, churn, etc.). We were in!
Problem: I didn’t have a product; it was just an idea. I told them I’d get back to them in a few months. I couldn’t code - I still can’t - so I considered paying someone to have an early version created. Fortunately, I didn’t. That would have been a huge waste of money given I had no idea what I was doing. I stumbled across an android developer, David, who was introduced to me through a friend. David agreed to start working for free in exchange for notional equity (‘sweat equity’), which was a huge coup. We started to try and design the app over summer 2015, which I found horrendously frustrating because I have no eye for design. But we were under way. Stardom and riches beckoned! I could practically see my face on Wired magazine.
In September 2015 I moved to Oxford to start a moral philosophy PhD. I was, and still am, researching ways to increase happiness, so there was some synergy. I’d been given a place in March 2015, just as I had about pulled myself out of depression. I didn’t find it a problem to work on a start up and do PhD research simultaneously. The work I was doing for Hippo – designing the product and talking to users – I was able to do much quicker than David was able to build the product (he had a normal developer job). I’d work out where I think we needed to go next, pass that on, and then get back to reading philosophy, psychology and economics papers. Switching between the two types of activity was hard because the skill sets were so radically different. The empirical work I did into happiness as part of the PhD did feed into the app, but nothing about start ups helped me write better philosophy. At the time I thought I could easily do both, but now I’m spending all my time doing philosophy I can really see how much more effective I am compared to when I was splitting my concentration between two things.
At the turn of 2016 we launched the first, extremely alpha version of the product to a few people. It wasn’t an instant success, but some people did use it and find it useful, which astonished me. I did a whole new set of user interviews and my co-founder and I slowly worked on improving the product. We iterated through this several times. I found the uncertainty of trying to work out what was the best thing to do next terrifying. By the time we launched the alpha it was 9 months since we’d had the expression of interest from the bank. We went back to them apologetically telling them “hey, we’ve built something” but they never replied. Too slow, clearly.
Around March that year started talked to a computer scientist from my college, who I’ll call ‘A’, about building a machine learning algorithm. Basic premise: you tell us how you’re feeling, we give you an evidence-based suggestions. You tell if it was good and it learns over time. Over the next couple of months, I designed a spreadsheet with all the inputs and outputs and an initial algorithm, which I handed it over to A who’d agreed to implement it. He was pretty enthusiastic but basically nothing happened for months and months. I’d get distracted with my work, then sometimes chase him and he’d always promise it would be just around the corner. I didn’t really have any bargaining power or other options, and I believed he would eventually produce something, so I chose to wait it out. In retrospect, there was no need to even try to use maching learning: we could have just randomly given people suggestions if they pressed a button; the suggestions were only text anyway. Ah, Captain Hindsight, the smartest man I know!
Throughout 2016, and doubtless to the frustration of my then-supervisor, I’d go through bursts of working on Hippo, then switching back to philosophy. The main problem with Hippo as a happiness tracker was that, fundamentally, it wasn’t that useful to people. Sounds obvious now, but took us ages to work out that was the problem. People seemed to like the idea, but didn’t find it useful and, where they said they found it useful, still didn’t stick with it. We ran into some key technical issues – it wouldn’t randomly send the notifications asking you for your data - and never solved them, which meant Hippo never actually did what we wanted it to.
Despite all this, and clearly in grip of self-delusion, I pitched Hippo to Y-combinator in Autumn 2016. It seemed like a big, crazy enough idea they might go for it. At the time, I thought Hippo had a higher expected value than my academic research (which wasn’t going terribly well, in part because I was neglecting it to work on the app) and was contemplating the process of dropping out of university if we found funding to allow David to work full-time. I thought the gap in the market(s) and the solution was so obvious I was convinced that, if I didn’t get there first, someone else would.
A consistent problem, with a single developer working in his spare time, was that we constantly struggled with the amount of development we could do. Late 2016 onwards I increasingly focused on trying find additional developers who would come and work for Hippo. This was a really hard sell: we weren’t growing, we didn’t have any money. Probably the most psychologically painful part for me was having to try and convince other people of my idea, an idea I was still struggling to articulate, to get them to join the project I was constantly frustrated with myself. Around the same point I started entering a few pitch competitions and going to entrepreneurship courses. I thought it would be useful to get feedback on the product and maybe find some guidance on what I should be doing. As far as I could tell this was almost entirely a waste of time. I did meet a lot of people who were slightly enthusiastic about helping, but almost no one who was genuinely useful, i.e. shared my dream and could tell me something I didn't already know. I was approached by an angel investor in early 2017; he’d heard about the project through someone I vaguely knew. This made the whole enterprise suddenly potentially serious: he was looking for a sort of HR-happiness-engagement product already but hadn’t developed anything yet himself. In the end, we couldn’t reach a deal. I wanted some investment so that David could work full time, whereas what the investor wanted was equity in Hippo in exchange for (what seemed to me) an extremely vague partnership arrangement and, possibly, some investment afterwards. To quote a current line from British politics, it seemed that ‘no deal is better than a bad deal’ and I turned it down.
We’re now in Spring-Summer 2017 and the final few months of Hippo’s life. A, the computer scientist, got conscripted into the Egyptian army, never ending up building the AI algorithm we’d been discussing, and disappeared. This turned out to be good thing in one sense. With him gone, I decided I should manually test the smart-suggestion system: I got people to send me messages on facebook, I asked them some standard questions and then sent them a suggestion for what they should do now. It actually worked pretty well and caused a brief burst of enthusiasm. After a philosophy talk I’d given, another developer, call him ‘V’, offered to work on the app and our fortunes suddenly looked up. But only briefly. We went our separate ways for summer, and when we came back V had found a job in San Francisco and was no longer interested. Around the same period, I saw CEA advertising a part-time role as Peter Singer’s research assistant. He was planning to write a book on overpopulation, a topic I’d accidentally ended up working on in my thesis. I applied, was offered and gleefully accepted the role (I know, odd twist for a start-up story...). I’d also done about 2 years of my PhD and it looked like finishing was both in sight and worthwhile. This series of events were significant because, for the first time (in my eyes) it seemed I would have a higher expected value by sticking in philosophy than continuing with Hippo. V’s departure or, rather, non-arrival, killed my enthusiasm for Hippo because I knew I wouldn’t be able to summon the energy to try and convince someone else to join David and myself. By contrast, having had a pretty shaky first two years of the PhD, getting the RA job made me much more confident I might find a position in academic philosophy afterwards.
I would, actually, have kept working on Hippo, even despite this. We were reasonably close to implementing the happiness-suggestions feature (‘instacheer’), at which point the app might have started being useful. David, my co-founder, ending up getting a different job and losing interest for his own reasons. Disappointingly, I never got to test my underlying hypothesis, namely: if we create a moodtracking app and use that info to give people people nudges on what to do, it will be useful enough that people and companies will pay for it.
And there I was, October 2017, two and half years later, the last one standing, and decided it was time to give up. So it ended. Not with a bang, but with a whimper.
As I see it, our failure was that we never built the thing we were trying to build. It’s not that we built a rocketship, launched the rocketship and saw the rocketship accidentally explode into a million pieces. Rather, we had a blueprint for a rocketship and there are still a million pieces for it sitting on the launchpad awaiting assembly when we walked away.
Why did we never get there? I think my particular mistake was probably being too optimistic about how much development work David and A would and could do. I should have been more critical of our development productivity and not given them so much of the benefit of the doubt. That might sound indulgent (“my main flaw is that I believe in people”), but I've come to realise this is a genuine flaw of mine: you can be too distrustful and you can be too trusting. I am the latter. I don't think I was a very good manager.
To explain, if I’d realised H, the guy who agreed to build the machine learning algorithm behind the smart-suggestions feature, wasn’t really doing anything, I would have tried much harder to find someone to replace him. Or, I would have done what I eventually did, which was test the prototype feature manually. Really, I should have run that test months earlier than I did. Equally, if I’d realised how slowly David would be able to make progress - one guy working part-time sometimes - I’d have done one or more of the following: (a) tried harder find an additional coder; (b) learnt to code myself - which I might have done if I’d not found David; (c) looked harder for investment; (d) given up earlier. As it was, we sort of plodded along, making enough progress that it seemed better to continue the current course. If we’d gone much slower we/I would have changed course. If we'd had the development resources to produce our product, we'd at least have tested the product we had in mind. We probably have failed later too, but that would be for a different reason requiring different explanation. A major problem was that, as a non-technical founder, I couldn't evaluate how much was happening. Of course, if we'd been building a building, it would have been obvious...
Saying the problem was misunderstanding how slowly development was occuring might also sound indolent ("why didn't you just build it yourself?"). Hardcore techies will probably be thinking I should have just dropped the PhD. I thought about that a lot, but as no point did that seem like the right choice. My objective was and is to increase worlwide happiness. Start a start up seemed like a means to achieving that end. But so did my PhD research, which is on the topic 'how do we maximise world happiness?' Given I had a technical co-founder but no programming skill myself, it never seemed like my comparative advantage would be to drop philosophy research in order to become an novice coder.
This is a bit tricky, because I don’t know how much I learnt that is also useful. My main reflection on Hippo was that I felt that I spent a huge amount of time to achieve not very much. I would consider that the cautionary tale. Here’s my list:
Startups do not necessarily teach you useful skills, particularly if your alternative option is to work in a different field. My expectation was that, whatever happened, I would learn loads to take to what I did next. As an aspirant philosopher, almost nothing I learnt from about entrepreneurship is directly applicable.
Failing unremarkably can happen and is something you should consider as a possibility. If I were to claim I had built career capital as a result of my 2.5 (mis)adventure, I would be deceiving myself. As it is I walk away with no cool story, no CV points. At least if I’d blown some money (mine or someone else's) and failed spectacularly, then I would have a story and I’d have felt that I’d really tried and failed.
You will likely learn things about yourself you didn’t expect to learn. In my case, I think this was (a) I put too much trust that people will do the work they say they will(/I have no idea how to manage people); (b) I don’t like working under great uncertainty; (c) I cannot stand trying to sell or ‘pitch’ commercial ideas (pitching philosophical ideas are fine); (d) I’m not very good at practical details (conceptual details are fine); (e) probably, most usefully of all, I improved my ability to be self-critical. Start up methodology (e.g. Lean Start-up, the Mom test) teaches you how to discover what people really think when they’re being too nice to tell you your product sucks. (e) does have psychological downsides (see below).
It’s not enough to try to build something you think the world needs. I started Hippo because I thought it might do an enormous amount of good. Yet, to make a successful product, you need to build something which is useful to people, not just something you think would to be useful to them. I potential danger with an EA start-up is that you might, as I did, find a hypothetical solution and try to retro-fit a business case.
If you try to start something, loads of people try to help you; unfortunately, many of them will be useless. I was honestly impressed at the fact people (well, mostly 1 other person) were willing to spend their time working on my crazy project and do so for free. Loads of people offered to help out: they gave advice, were willing to let me interview them to talk about the app, etc. Admittedly, lots of the help isn’t very useful, and a relevant challenge is to sift through all the different perspectives people give you and make your own decision.
The uncertainty can be terrifying. The constant experience of not knowing what to do in start-up life was, for me, anxiety-inducing and something that didn’t seem to get much better over time (if only there was an app to help people with their moods, oh wait…).
While failing is very common, it can still be psychologically very hard. In theory, I should feel glad that I dedicated a considerable amount of time to something I thought would have the highest expected value. As it happens, my abiding psychological response, when I think about Hippo, and which is probably irrational, is one of frustration and shame with only a glimmer of pride. Frustration that we never built what I wanted to test. Shame that I feel, in some sense, I wasted my time and should have known better. A glimmer of pride that I, at least, stuck with it for so long. The tragedy, as I feel it, is that I never found out if my hypothesis about an ideal piece of happiness-inducing tech was correct. I haven’t seen anyone else build it (yet) and I feel sufficiently burnt out by my experience I lack the motivation to try again.
Although I billed this a something of cautionary tale of failure, to be clear, I’m not suggesting EAs shouldn’t try to start start-ups. If you have a good idea, you should still go for it. In fact, I’m glad to see there are three other mental health apps being created by EAs - Mind Ease (Peter Brietbart, stress and anxiety), Inuka (Robin van Dalen, support chat app) and UpLift.us (Eddie Liu, depression) - and I wish them the best of luck. I'm also not suggesting I didn't learn anything at all. Rather, because stories of start-up failures are so often accompanied be tales of later start-up glory and vital life-lessons learnt, I wanted to add anecdata: it’s possible to fail in a deeply uninteresting way and not learn much. Further, it’s possible there’s something particularly galling about failing to start a start-up where your aim is to help other people and do good, rather than just make money, but I don't know any other failed asipirant effective entrepreneurs to ask them.
I hope my contribution here has been (1) to provide an example of someone who did try to start a company with the intention of doing the most good and failed and (2) to share what happened and what I did (or didn’t) learn along the way, and how I felt about it.
If anyone would like to chat to me about this privately, you can find me on firstname.lastname@example.org