My co-founder and I broke up. Here's a post-mortem.

My cofounder of a year and I recently parted ways, and I thought that I'd take some time to write down our experience, and how we came to the decision to stop working together. I think it's an under-talked about part of the startup journey for a lot of people, and while in our case the breakup was not at all acrimonious, the feelings of failure and disappointment are still there.

First off, a little background on the two of us: David and I met through mutual friends in New York City. He was an algorithmic trader who had studied math and CS at MIT, and I was a software engineer/freelance journalist who'd covered the technology industry for years for publications like the New York Times and Wired. On paper we were a really strong team. Not only were we technical and pedigreed, I was also clearly a hustler! It's hard to break into traditional publishing and if I could do that while also holding down a day job as a software developer, I could definitely start a startup in a bull market. In retrospect I was pretty arrogant.

I saw this tweet a while back that says that as a founder, your product is clarity, and I feel that perfectly sums up the biggest problem that David and I had, which was a lack of it. Coming out of the gate, we didn't have a clear idea of a problem we wanted to solve, or even a persona we wanted to target. We were would-be founders in search of an idea. And while that isn't necessarily damning, it was compounded by the fact that we didn't have obvious domain expertise which would guide our search — i.e. you're two ML PhD's, you should probably start an ML startup; you've spent 10 years in manufacturing, you build a solution for factories.

My work experience consisted of five years in infrastructure engineering, first in Observability at Uber, and then in local search infrastructure at Google. But these were big companies, where I was a line engineer, a cog in the machine. I didn't feel like I had visibility into the process of actually creating an infrastructure product, talking to customers, understanding their needs, architecting the whole system. David, meanwhile, had spent his entire career at Hudson River Trading, a super successful algorithmic trading firm, but he had left precisely because he was bored of trading and wasn't particularly interested in building financial infrastructure. He was also philosophically opposed to crypto. This meant that some of the areas where we had perhaps the most founder-market fit, we were not considering. (Side Note: the other area where I arguably had domain expertise was media/journalism, but we were forewarned that building a content business was not a great idea, especially if we were technical and could do something else.)

We ended up doing basically a brute-force search through the space of startup ideas. Most of our ideas were B2B, but a few were B2C, or in the creator space. (It's long, but if anyone is interested, I outline all our projects at the end of the post). The B2C projects typically arose from our own problems and we would do just a minimal amount of asking around in our networks or putting up a landing page to validate the idea before diving in and building. For B2B projects, we were running a bit more blind, since we typically did not have work experience ourselves in the area we were exploring, so we would reach out to certain personas on LinkedIn, for instance, account execs, ask about their workflow, develop hypotheses for specific problems they had, and then try to come up with a solution. We both read The Mom Test and were aware that people would regularly lie to you in an effort to be nice or encouraging, and we tried to stave this off by sticking to factual questions and asking people to screen-share their existing setups as much as possible.

One thing that we struggled with was when to switch from "we want to learn about the space" mode to "here's a solution, what do you think?" mode. The Mom Test seemed to imply that by presenting any idea at all, you were biasing the interview subject, but it also seemed impossible to get good feedback by perpetually speaking in generalities. This was further contradicted by studies that show that people are more truthful when they have concrete artifacts to respond to; in order to jog their judgement, it helps to show mocks and demos.

Another thing I struggled with, despite the admonitions of The Mom Test, was asking the tough questions. Because conviction was generally in short supply for us, I think I instinctively recoiled sometimes from asking questions that might further deplete it. And while we did prep for the calls sometimes, we should have prepped more. In particular, we didn't always go into the calls with certain hypotheses in mind that we wanted to test, or gather more data for. It was often a more dispersive conversation that, while enlightening, didn't necessarily get us closer to identifying either a problem or a solution.

The reality is that even so, a lot of the products we worked on probably would be viable (or we could at least raise money for), but we never stuck with one of them long enough to find out. We would work on something for a while, get bored of it and realize that we didn't want to spend the next ten years of our lives on it, and then want to move on to something new. There were a few exceptions, where we worked on something for longer and didn't see the traction we needed to, but by and large we were pivoting a lot, and for a lot time, between wildly disparate areas.

The other thing we did too much of was over-intellectualizing. It wasn't that we weren't aware this was bad. But it was really challenging to avoid because when we didn't over-intellectualize we often ran into issues like market size. The wisdom from VC's and other founders was that the market by far was the biggest contributor to a startup's success, and when a good team met a bad market, the market won. So we would see problems we thought were real and interesting, but then worry about whether the market was big enough, and lose conviction. The pressure to build a venture back-able startup often prevented us from building any business at all.

Tactical Errors

  • Raising Money — Some part of me thinks we would have been better off just picking an idea to start and raising money on a slide deck, and figuring it out from there. At least we would have had some external recognition and momentum. We never really tried to do this, even though we talked to a lot of VC's and some offered us money de novo — in part because I think we never got our story straight. These days you can raise money pre-revenue and pre-product, but I do think you need to have a convincing narrative, and we just never were able to tell one that convinced, at the very least, ourselves.
  • Build in Public — The other thing that I think we should have done more of was to build in public. This would have been very natural for me because I was already a journalist/writer, but we switched around so much between ideas that it felt slightly embarrassing to be shilling a new product every month or so.
  • Branding/Messaging — both David and I were backend engineers, and we did very little thinking around brand. For the B2B product ideas we had, this wasn't too much of a problem, but when we worked on a few consumer or consumer leaning ideas, I think it really hampered us. One disagreement we had consistently was over whether or not the site had to look good. My take was that it needed to, just because it's relatively easy to make a super slick looking site now, and because the tech itself is not that distinguished, and consumer expectations are so high. David disagreed. He was annoyed that I often cared more about the superficial details than the architecture, left technical debt in place, and in the course of trying to make the site look better, would sometimes introduce additional bugs that he then had to clean up. I felt bad about that, but I also felt that people wouldn't even have a chance to encounter these backend bugs if the site didn't look legit.

Things we could have vetted for better in each other

  • Goals Alignment — When we were vetting each other as potential co-founders, David and I asked each other all the questions you were supposed to, like what does success look like, and what are you motivated by. In hindsight I was either not completely honest or just didn't have enough knowledge of myself to answer those questions completely honestly. David was always clear that he wanted to build something really really big, and while I concurred with that at the time, I think I just really want to build something of my own. This sounds obvious, but I think it's easy to disregard in the dating stage when you often really want a cofounder relationship to work: when you're talking to potential co-founders, it's not sufficient that you answer the questions, you also have to answer them truthfully.
  • Passion Alignment — When we started out, I think David and I considered ourselves to be fairly pragmatic. We both grew up in Chinese immigrant families and more or less bought into the notion that passion follows excellence. We were willing to work in areas that weren't that sexy to us personally, if it would get us somewhere fast-growing and big, and we figured the excitement would follow. I still think that this isn't wrong — there are people who can get off on the building-a-fast-growing-business part of founding startup, but I don't think it's me. I do have to be pretty passionate about something in order to make it work. And over time, it became apparent that the things David and I were truly deeply passionate about were different — for him it was more knowledge, ML, text search, crunching and parsing large amounts of data and organizing it (sounds pretty similar to algo trading, heh :-p). For me it was creating content and bringing friends together to grow intellectually and socially, taking classes, exploring new fields. There were some products we worked on that sort of sat at the intersection of these passions, but a lot of times the intersection didn't really hit the sweet spot.
  • Complementary Skillset/Personalities — David and I were both backend engineers and neither of us was particularly experienced at marketing or design or operations, or even front-end product development. This lack of complementary skill sets, and perhaps more problematically, complementary personalities — neither of us were hype men who could, push comes to shove, really make the sale — I think ultimately hurt us. Selling made us both uncomfortable. Asking for favors and goodwill from our network without being able to give something back made us uncomfortable. But at the time, we just figured that since each of us had a broad enough base of competency, that it would be fine — we would just bite the bullet and each do a bit of everything. In retrospect this was pretty naive.
  • Communication Styles while David and I shared many values and qualities, we had very different communication styles and thinking styles. He is extremely process-driven and data-driven, while I am much more instinctive. It annoyed him when I made assertions that he didn't think were backed by anything more than feeling. (One of his pet peeves in particular was when I preceded a statement with "I have to imagine...") In turn, I often thought that he was too detail oriented, or pedantic about stuff that didn't matter when we didn't even have the big picture settled yet. I think my style of conversation is optimized for good dinner table interactions, and is more fun and breezy and socially aware. His, in turn, was optimized for information transfer. I respected that, but also felt like in a lot of conversations, especially with third parties, that he was just sullen and curmudgeonly.

Lots of companies fail because of co-founder conflict, but I actually think that under different circumstances David and I could have been successful together, i.e. if we had a clear sense of the problem we wanted to solve/the product we wanted to build and we were both really excited about it. The experience of being cofounders also hasn't, I think, fundamentally changed our view of each other. I still think that David is someone with incredibly strong principles who works extremely hard and is extremely smart.

But results speak for themselves and after a year of kind of going to circles on various ideas, it was clear that something wasn't clicking. We thought about bringing on a third co-founder, someone with more domain expertise, but our interests were sufficiently divergent that maybe the way forward wasn't together.

What's next for us:

David is taking a bit of time off, and he's looking at joining some other early stage projects. I recommend him wholeheartedly. You can drop him a line at itsdchen@gmail.com.

I am taking some time (probably through the rest of the summer) to focus on my writing. I have four or five big pieces on various parts of the technology industry that I've been itching to work on for a while, but felt like I couldn't because I should be 100% focused on the startup.

I am also doing some consulting work for tech and finance firms, mostly writing strategy-type memos.

Some other ideas I'm interested in exploring:

  • technical content creation — lots of companies need to create content like technical blog posts, explainer videos, documentation, etc. to educate their users on how to use their product. There's a dearth of technical people who can also create content to do this.
  • group therapy for break-ups — breakups are a great catalyst for personal growth and basically a universal experience. People mostly rely on friends right now for support while going through breakups, but 1) not everyone has friends and 2) friends generally are sympathetic for about a month, while breakups can often take years to recover from.
  • any sort of idea that can take advantage of TikTok as a distribution channel — advertising on TikTok is still new so I think there's probably more juice to be wrung out vs. advertising on Facebook and Google. It's also a fundamentally different format.

If you're interested in any of the areas above, DM me on Twitter or drop me a line at yirenlu92@gmail.com.

TL;DR Takeaways

  • Have at least a problem in mind to solve, or at least a slightly narrower area you want to explore.
  • You should be at least have enough conviction about what you're working on to want to talk about it all the time.
  • Having stuff look good matters, having stuff sound good matters.
  • Build in public.
  • Be very deliberate and careful about how you conduct user interviews and make sure you go in with certain hypotheses that you want to validate.
  • Find a cofounder with complementary skills and personalities. Just because you're good friends and both hardworking, smart, people doesn't mean that you would necessarily be good cofounders.


Here's a list of some of the areas we explored
  • Better management of FB + Google ad spend
    • We had a hypothesis that for e-commerce companies, that there might be a need for a more niche marketing analytics tool beyond Power BI, Tableau, Looker, etc.
    • After talking to a lot of DTC brands, there didn't seem to be a super obvious gap/wedge, although people were spending a ton of time custom building attribution models, setting up their marketing analytics stack, etc.
  • Mellow Groceries
    • We were drop-shipping Asian groceries from Chinatown in New York City at the height of the pandemic. David and I had some disagreements about whether it was ok to lose money on every transaction. He thought no. I thought yes.
  • Drag-and-drop SQL
    • Both of us hated writing SQL and were wondering whether there was a tool we could build to avoid having to do that.
    • We spent some time talking to people in BI and learning about the various ETL transformation tools, data warehouse solutions. People really love Snowflake.
    • This didn't really seem to be a need — there was a bit of "people should just suck it up and learn how to write SQL" attitude.
  • Jot — implicit document search
    • The idea was that these days there's tons of "implicit" documentation at companies that is happening through chat — i.e. you ping someone to figure out what's going on with a new feature, or how to debug something. But all this implicit documentation is lost in 1-1 chat silos. We built a Slack app to clip parts of Slack conversations into a notes repository, with the idea of eventually being able to pull structured instructions out of unstructured chat logs.
    • talked to some potential customers, but it turns out that while lots of companies acknowledge internal documentation is a huge mess, no one cares enough to want to solve it.
  • Medical Debt Collection
    • This was actually super interesting. Basically hundreds of billions of dollars of medical debt goes uncollected every year, and it ends up being sold on the secondary market for pennies on the dollar. There are charities that buy up whole tranches of it and forgive the debt in bulk. We had an idea to provide a platform for patients to give some proof of what they are actually able to pay (say by linking up their bank account), and then allow them negotiate a discount on their medical bill directly with the hospital or doctor.
    • We talked to a lawyer and it turns out that due to their contracts with insurance companies, doctors and hospitals actually have some legal obligation to attempt to collect on the full amount, several times, before offering a discount. The bigger issue we anticipated was that it's notoriously difficult to sell to doctors.
  • Social stock trading
    • We had an idea for social stock trading app that was sort of similar to Public.com, since David was part of a bunch of chat groups with friends who would talk about trading, and a lot of people we knew were quite passionate about it.
    • We vetted this idea with some of our friends, and they weren't really convinced that they would actually use an app like this — as in, beyond periods of extreme market volatility like last March, how often were people actually posting in these chat groups? And how often were people actually posting new stocks/new ideas in these chat groups? Not that often. As it turns out, the market was way more volatile for way longer than we had previously anticipated.
  • Robinhood for Corporate Bonds
    • There's no good way for consumers to access corporate bonds right now
    • David talked to one of his old co-workers from HRT about the idea and was laughed out of the room. I think supply — who would be willing to sell small amounts of corporate bonds — was considered an issue.
    • Another issue was that corporate bond yields are maybe 7% and not particularly exciting when compared to Dogecoin
  • RPA for Insurance
    • We thought there might be some way to build a RPA tool specifically for the insurance industry that would help agencies do data entry into their legacy systems.
    • Insurance is incredibly complex — we spent three weeks just learning about P&C insurance and the different players in the agency software vs. carrier software market.
    • We were unable to identify an actual strong pain-point that could be addressed by RPA. A bit of a solution in search of a problem. Also had trouble getting access to insurance agents.
  • Sailor — research notes on sales prospects for personalized outreach
    • Account executives spend a lot of time doing research on prospects so that they can send them tailored messages. This includes looking at their LinkedIn, watching videos and testimonials, reading articles, looking at 10k's, etc.
    • If you can automatically generate "highlights" about a prospect, and maybe then be able to stick those highlights into a template for a personalized message, you're saving the sales person a ton of time.
    • We were scraping LinkedIn, Bing search, YouTube, podcasts, Twitter and cleaning up the information and turning it into structured metadata.
    • Ran into some legal concerns re scraping LinkedIn.
    • one learning we took away from here was that you should really mock everything up and demo that before building anything. We spent too long building the data parsing/notes portion before demoing it to people, and we could have gotten the same feedback by just showing mocks in terms of what types of information were actually useful.
  • Alexa for SaaS apps — we got a YC interview with Jot, but ended up switching to this one by the time we were interviewing.
    • we built a chrome extension that you could issue voice commands to like "Create a new Github issue" or "List all my Asana tickets"
    • the idea is that context switching takes up a lot of time, and nothing beats voice when it comes to ease and speed.
    • competition here is stuff like Slashdot/command-e
    • We had Michael Seibel for our interview, who was very nice, but he was very skeptical about people's desire to use voice, and I don't feel like I handled the pushback particularly well.
  • BookTime — basically a paid calendly tool that would allow creators/domain experts to offer 1-1 paid micro-consultations.
    • We spent nearly 3 months on this, and this was probably the project that we got furthest along on.
    • Started with authors and writers who could offer advice on pitching/publishing to other writers, got our first bookings that way.
    • We struggled to onboard more supply side people. Either people were already using something like Calendly or Acuity, and didn't want to switch, or they weren't, at which point they probably weren't into commercializing their expertise.
    • Even among people who did really well on the platform, they ended up churning because 1) synchronous 1-1 chats are hard to make a lot of money off of and it wasn't worth their time or 2) they wanted to consolidate this payment platform with their website and other social presence.
    • Never really decided whether we wanted to stay focused on writers, or move into other verticals like YouTubers, life coaches, etc. We definitely ended up being a bit of a solution in search of a problem.