This is a very long post for people who want a lot of details. If you want the shorter story of how I plan to avoid these mistakes at Otherbranch, look here.
Discussion on Hacker News here.
It's a long story
It's been a year and a half now since I was at the helm of a dying company.
I've had a lot of time to think.
My name is Rachel. I joined Triplebyte in 2018 as a random ops employee; after a few other roles I eventually became Triplebyte's head of product for its last two years. I'm also the founder here at Otherbranch.
I'm writing this retrospective for three reasons.
First, I’m trying to bring something like it back (if the website didn’t tip you off). And to try again, I personally need to understand why Triplebyte failed. I want to avoid those failure modes, and that means I need to know what they were and how to avoid them. Writing about them is a way to demand enough clarity and rigor from myself to publish something for the world.
Second, Triplebyte was, at one time, a company with an excellent brand delivering a well-liked product. Many people have asked me, both professionally and personally, why a product that had such a good reputation with engineers would abandon the product they loved - or how such an apparently-healthy product would fail. I'd like to answer that question as well as I can.
And third, I'd like to offer a peek into the realities and difficulties of business for readers who may never have faced such challenges themselves. I don’t mean to offer an excuse for our (or my) failures, nor for the many failures - both moral and tactical - made by businesses in general. But if you want to do business better, you need to understand business better, or at least as well as those who do it today. You have to know why those failures exist before you can avoid them. And I'd like to put some of that knowledge out into the public consciousness.
What was Triplebyte?
(I suggest reading this section even if you’re familiar with Triplebyte, since a lot of things changed over time.)
Triplebyte was a technical recruiting company for software engineers.
It came with an impressive founding pedigree. The founders were Harj Taggar, the first non-founding partner at Y Combinator, and Ammon Bartram and Guillaume Luccisano, both engineers themselves. Ammon and Guillaume had previously co-founded and sold social-video company SocialCam with Justin Kan, himself best known for founding gaming site Twitch.tv (its original name, justin.tv, is derived from his name). Triplebyte was funded by Y Combinator (and by their extended investment arm, YC Continuity) throughout its history, along with other major Silicon Valley investors.
Triplebyte sat at the intersection of two things that are central to tech culture: a(n at least nominally) meritocratic culture of testing and statistical rigor, and the problem of hiring good software engineers. It grew up alongside the rise of coding bootcamps, growing skepticism about higher education (I think we might have done a podcast with Bryan Caplan once? it would certainly be on brand), and a huge boom in Silicon Valley investment driven by low interest rates and the wild success of Uber, Airbnb, and others.
The idea was this: if higher-education and prestigious experience weren't reliable sources of quality, and if you could make an ML model that could identify it better through testing, you could find great engineers who'd be missed by the usual resume screens. Then, because top engineering talent is enormously competitive, companies would be willing to pay for this newly-unearthed talent in a way that would result in a viable business.
(You may notice the double-vision here between "this is a good business opportunity" and "this is a meritocratic moral good". This pairing would become a defining feature of Triplebyte's culture. Like most organizations, the former often won out, but unlike many organizations the latter was not merely an excuse for the former. We really did struggle with that a lot.)
Over the eight years it existed, Triplebyte operated four main products. Some of their names changed over time, but I’ll call them by these names:
FastTrack. Engineering candidates would take a quiz, followed by an interview with Triplebyte's engineers. If they passed both, we would introduce them to companies on the promise that those companies would skip them to onsite interviews. FastTrack emphasized speed, an elite cohort of engineers, and white-glove service. This was our main product from the founding of Triplebyte through 2019, then faded into a secondary role in 2020 before ultimately being shut down in 2021. (This is the product we're most heavily drawing on here at Otherbranch.)
Source. Source was described internally as “LinkedIn for engineers”. The idea was to take the candidate pool we’d built up through years of the FastTrack product and create a network of skills-verified engineers that we could sell access to as a SaaS product. Source emphasized volume and scalability. This was our main product in 2020-2021.
Screen. Screen began as a growth experiment. The idea was that we’d give companies access to our quizzes for free, as long as we could ask candidates to sign up for a Triplebyte account afterward. This proved very popular with companies. Screen was never monetized meaningfully (although we did occasionally charge for it to get companies to use it in blatant violation of all economic sense). instead, it was meant to bring in candidates to create a valuable candidate database. It became our primary growth avenue in early 2021 and remained so for the remainder of Triplebyte’s history.
Magnet. Magnet was an automated sourcing tool built on Source’s infrastructure. We would set up campaigns on behalf of companies hiring through us, and Magnet would message candidates automatically (transparently; they knew they were automated) with potential leads. Magnet emphasized signal quality and lack of required effort from both employers and engineers. Magnet was our main product in 2022 as a last effort before we ran out of runway.
So. With the preliminaries out of the way, why did we fail?
We never solved candidate acquisition at scale
Companies usually don't fail for a single reason. But if I had to pick one, this would be it.
Recruiting firms work best when they're working with candidates that are hard to hire. Great engineers, in particular, are hard to hire (and were harder to acquire in 2018 than in 2024) because there are always fewer of them than the market wants - or rather, that's the case for the experienced, skilled, US-based engineers that most US companies are trying to hire. The candidates we most needed to acquire were, by nature, the hardest candidates to acquire.
Some companies (the companies I intend to focus on myself) do want smart people who don’t have much of a resume. But that list is not very long, and (as we’ll see in a moment), Triplebyte wasn’t and couldn’t be satisfied with such a small TAM. So in practice, seniority mattered a lot. This was especially true after we pivoted from FastTrack to Source, deemphasizing high-touch quality and emphasizing pay-for-access. No one wanted to pay for access to a pool that didn't come pre-filled with the people they already knew they wanted to hire (as opposed to the ones we could tell them they should consider).
During the 2010s free-money tech boom (senior, US-based, credentialed) engineers did not have trouble finding a job if they wanted one. The slow-motion tech implosion (nice name for a band) of 2022-2023 changed this dynamic, with a glut of engineers on the market after layoffs and company failures, but that wasn't the case for most of Triplebyte's history (and likely will not be the case in the future, at least not until/unless AI meaningfully replaces a large percentage of engineers). For Triplebyte, that meant that convincing sought-after engineers to use Triplebyte instead of their many alternatives was a core problem. And we never succeeded at doing that at scale.
Even though early Triplebyte had a well-known and well-liked brand (particularly in the early FastTrack era), and even though we could identify talent the market would otherwise ignore, neither of those things was bringing in enough candidates for Triplebyte to grow at the rate expected of a venture-backed startup.
Instead, the vast majority of candidates in the FastTrack era - roughly three-quarters - came in from ad campaigns. Any engineer active on Reddit from 2016-2018 is no doubt familiar with Triplebyte Mike (our growth-guy-turned-erstwhile-mascot). Our ad channels saturated to the point that organic (hopefully-tongue-in-cheek!) hate subreddits developed purely because people were sick of hearing about us. For all that people loved FastTrack as a product (and people really did), ad spend dominated word of mouth by almost an order of magnitude.
Worse, ad spend doesn't scale well: each marginal dollar you spend on ads gives you less value than the last. Harj, one of the founders and the CEO at the time, talked about it for YC as part of a series on mistakes startups make. We scaled ad spend to maintain a targeted growth rate, but this became less and less effective as ads ran into greater and greater diminishing returns. And when those diminishing returns came home to roost, we had no other means by which to scale.
In short, FastTrack was a viable product at moderate scale (on the order of $10M in revenue a year), but it didn’t have a path to growth. Or at least, it didn't have a path to the kind of growth expected of a VC-backed company in 2018.
Which brings us to…
The catch-22 of venture capital
In the days following the pivot from FastTrack to Source, a common lament among engineers was "ugh, another company ruining a viable and valuable business in favor of pursuing growth". The word "enshittification" hadn't been popularized yet, but if it had been, that’s the word people would have used. And it would have been (in a broad sense, at least) the right word to use.
But what I think might be hard to understand from the outside is that (in an appropriate sense) we didn't have a viable business. Yes, it was financially break-even, at times even profitable, even with little focus on optimizing ops costs. Yes, we had a pretty good brand. Yes, engineers loved us and companies liked us. But to understand why it wasn't viable at low-mid-scale, you have to consider how we got to that state in the first place.
Triplebyte was and is widely known among software engineers. But for all that I'd love to believe that that was because of the wonderful quality of our product, it realistically wasn't. It was ads. A great product can get you a great brand image, but it doesn't create that kind of brand awareness quickly. The overwhelming ad spend described in the previous section made sure it was almost impossible to avoid hearing about us.
Why were we able to buy those ads? Because investors gave us a lot of money.
And why do investors give companies a lot of money? On the expectation that they will pursue aggressive growth, not stop at small to middling scale. (This is somewhat less true today than it was in 2015, but it is still mostly true.)
In short: if you needed venture capital to produce a viable small-scale business, you didn't really have a viable small-scale business to begin with. Triplebyte's break-even status was an illusion, predicated on funding that would probably never have existed if that scale were the founders' only ambition.
I’ll be honest here and say that I don’t know what the ratio of investor pressure to founder ambition was in the decision to pivot to a new product. My personal belief, based on my knowledge of the people involved, is that there was almost certainly at least a little of both. And it's a little hard to separate them - a VC doesn't have to exert any deliberate pressure for an ambitious founder to know that their future chances of funding depend on their success.
This isn't a shot at venture capitalists. I have plenty of grievances with venture capital and the anticompetitive or robber-baron-ish practices it often creates. There's a reason this section starts with a link to Cory Doctorow. But expecting founders to deliver what they implicitly or explicitly promised is not one of those grievances.
None of Triplebyte's founders were fools. Harj was actually from Y Combinator itself; Ammon and Guillaume had founded a company before. All three were very familiar with the way venture-backed business is done. They knew exactly what they were getting into and went into it with open eyes, and I do not place any blame on any of them for that. Their ambition was to build a large and disruptive company that would change the way an entire industry is run. And ambition is not evil, even though it often justifies things that are.
There is nothing inherently wrong with shooting for the moon, even if your rocket blows up halfway. That is literally how we got to the moon.
But regardless, Triplebyte as defined by FastTrack sat in that catch-22: we needed venture funds to create a viable company, but that viable company did not satisfy investors' growth targets. And in that sense, we did not have a viable company in the first place.
That meant a pivot to a direction with better growth potential. Unfortunately…
We pivoted to a product no one wanted
"LinkedIn For Engineers" - that was the by-phrase within Triplebyte for most of 2020 as we shifted into the Source era.
Job searching on LinkedIn sucks! (True.) Engineers hate LinkedIn! (Often true.) So if we just make a LinkedIn that doesn't suck (uh-oh), everyone should want to use us instead!
I present it here in a somewhat comedic tone, but this wasn't a ridiculous idea on the face of it. We didn't need to worry about the cold-start problem (because we already had a bunch of users on both sides of the hiring process) and we were competing against an incumbent people don't like. Again, none of Triplebyte's leadership were stupid, and they didn’t pick that direction arbitrarily.
Conventional wisdom, and wisdom within the company at the time, was that if you want to disrupt an incumbent, you need to be a step function better. The claim was that our skill assessments and our engineer-specific functions could accomplish that. And, again, this wasn't crazy. Our assessments were very, very good. That part wasn’t wrong.
But the problem was that we couldn’t just be a step function better at something. That can work for a company just starting out (and in fact it’s standard advice for making a great startup), but we were a growth-stage Series B company with a nine-figure valuation. We needed to be a step function better at the core value proposition of our space. And the core value proposition of LinkedIn isn't "we make finding a job easy and pleasant". It’s "we have all the jobs and all the candidates". One of the other members of Triplebyte's leadership later on understood this very well, and spent a good chunk of a year trying to get it through my thick skull.
No one wanted another LinkedIn, because LinkedIn had already perfected its we-have-all-the-jobs-and-all-the-candidates value prop. FastTrack-era Triplebyte had niches LinkedIn couldn't match, but Source-era post-pivot Triplebyte largely lost them.
Nor was there any particular reason for engineers to seek us out. We’d lost the claim to unlock opportunity, the claim to expedience and convenience for senior engineers, and the claim of human help from a team of friendly and supportive people. We’d lost all the things that (insofar as they ever had) made engineers seek us out. So they didn't. For the remainder of Triplebyte's lifetime, activity on the platform, and value delivered to both candidates and company clients, would be heavily dominated by candidates who had used Triplebyte during the early FastTrack days. This would remain true even years later, when FastTrack candidates were outnumbered three-to-one and were years removed from signing up. Candidates from the FastTrack era were more than an order of magnitude more active up until the very last time I looked at our analytics.
If you want one reason I'm starting a company inspired by FastTrack: that's it. That's what product-market fit feels like. We broke the thing they loved, pissed them off with anti-privacy decisions, failed to deliver much of any value for years, and they still had so much loyalty to what we'd done for them in the past that they were still our best users.
Ultimately, the late-Triplebyte product that did partially succeed - Magnet - succeeded by breaking away from the LinkedIn-for-engineers mold and trying to do something fundamentally different, something that LinkedIn does not and cannot do.
(I won’t talk about Magnet much here, because the proximate reasons for its failure are pretty simple: we ran out of money and the market fell out from under it. It might have - smart money says probably did have - other means of failure, but we didn’t last long enough to find them in any great detail.)
Still, if we’d had enough of a candidate pool, it’s possible that we could’ve survived as a secondary, auxiliary tool alongside LinkedIn. To do that, we needed a way to build that pool.
Our bet on Screen didn't work
In 2020, we began to aggressively experiment with a product we called Screen.
Screen took Triplebyte's assessments (which, again, were as good as we claimed they were) and made them available to companies to use to screen their own candidates. Candidates would apply to a company and the company would use our assessments to screen those candidates, with the understanding that we would present candidates with a chance to sign up for Triplebyte to reuse their assessment scores in the future.
The idea here was that we could solve candidate acquisition by having companies do it for us. We could just let them use our assessments, at almost zero marginal cost to us, to bring in a constant flow of candidates. And we could do so in a way that served our more moralistic goal of more meritocratic hiring.
On paper, this worked pretty well. At its peak, Screen brought in tens of thousands of candidates a month, at almost zero marginal cost. By the time Triplebyte closed its doors, our candidate database was heavily dominated by candidates sourced this way. Of our approximately 250,000 accounts at the end, somewhere between 2/3 and 3/4 were Screen-derived.
And companies loved it! Screen was the only thing that did have market fit for much of Triplebyte's late history. It had a good growth curve, solid NPS, and seemed largely inoffensive to candidates going through it as part of a company's hiring process. As a product, Screen was wildly successful (granted, it helps to not be charging anything). If you’d been in any of our company leads meetings during this period, you’d have seen a wall of red “not on pace” quarterly goals with some yellow/green Screen goals being the only thing that came anywhere close).
But Screen had a problem, one that was not obvious at first glance.
Remember that fundamental problem I mentioned about forty paragraphs ago? The one where companies came to us (particularly post-pivot) to hire the candidates (senior engineers in the US, especially ones from top schools or with FAANG-like experience) that are hardest to acquire? Well, it turns out that companies also don't feel much need to pre-screen such candidates and want to avoid putting barriers (like, say, an online quiz) in their way.
The candidates that companies wanted to send through an automated screening step were the candidates that they considered both plentiful and of dubious quality. For most companies, that meant inbound applicants (if you've ever hired before, you know how awful the average inbound candidate is), it meant junior candidates, and it meant candidates from abroad.
For those use cases, Screen was an excellent fit. It offered a way to trim down the firehose of inbound candidates in a way that actually meaningfully measured knowledge and ability. In fact, usage was so high that a user in India actually crashed our whole web app on Christmas Eve by sending a class of some 15,000 students through it at once. (From a coworker who read over a draft of this post: “I remember this! [Our recruiter/recruiter-liaison] and I were debugging it. He was drunk at a bar, I think.”)
For a product meant as a growth channel, “we brought in so many candidates it crashed our website” sounds like as clear a win as we could get.
But remember, our goal wasn't get people to use Screen - our goal was acquire valuable candidates. And unfortunately, the candidate groups brought in by Screen were mostly mutually-exclusive with the US-based senior candidates that our clients were buying Triplebyte subscriptions to find. We built a database - but not a valuable one.
Our core metrics didn't take this into account until too late. Our main metric for Screen was simply usage, and that metric looked great. Conversion to accounts was lower, but at around 30-50% it literally could not have been more than a factor of 2 or 3 better. It was only when you looked at downstream results that you'd find trouble brewing. We realized what was going on about seven months into this strategy, in late 2022, but by then it was too late: we'd committed to a path, didn't have runway to pick another, and were already facing much larger and more immediate crises (which, in my anti-defense, were at least partially a result of my own inexperience as a then-new head of product - I don't think I was doing a very good job at the time).
This problem wasn't the fault of Screen as a product. For quite some time it was literally the only part of Triplebyte that was achieving anything. Users liked it, it delivered value, and it accomplished the product goals it set out to accomplish. It just didn't accomplish its business goals, because it fundamentally did not work for the kind of growth we needed.
The tactical execution of Screen was fine, but the strategic decision to focus on it was wrong. (Granted, I do not really have a better idea for what we should have done instead. There’s a reason candidate acquisition remains my #1 bullet point here. It's possible that by this point we were simply playing a losing hand with no path to victory.)
The way it could succeed at its metrics and fail at its business objectives is an example of a classic business lesson: make sure your metrics actually reflect your strategic goals.
We tried to change our users
The last thing I want to talk about is more abstract. It's a fundamental philosophical difference between Triplebyte and our major competitors.
Triplebyte was, at its heart, a technical organization. Two of the three founders were engineers themselves, and the company culture bore that DNA. We were a geeky, technical, quantitative, introverted, opinionated company, dominated for better and for worse by that particular kind of intellectual adventurousness that makes the Bay Area a fascinating place to live.
When we presented an ethos of background-blindness and meritocracy, that wasn't just a brand. It really was the way we thought about things, a spirit that everyone in the company cared about, and that was deeply embedded in company culture. I myself was hired with no tech credentials almost literally off the street. I would eventually myself hire a fantastic employee with no tech background out of an ersatz Home Depot in rural New York thanks to that same ethos. My predecessor on my original team is now a journalist at Vox. We were a weird bunch.
(This is not to say that that ethos was never violated. It was. Strategic decisions sometimes overrode it, sometimes in ways I agreed with and sometimes not. But the ethos was still embedded in company culture so strongly that these strategic calls sometimes led to people leaving the company out of loyalty to it.)
In short, we believed, with the specific sort of arrogance that is another thing that makes the Bay Area a fascinating place to live, that we knew how engineering hiring should be done. Our goal was to change what companies wanted to do, not enable them to do they already wanted to do.
And I don't think this was necessarily wrong. Our users often behaved in ways that I thought - and still think - weren't to their benefit, ways that create painful externalities that make job hunting awful for everyone.
But this post is called Why Triplebyte Failed, not Where Triplebyte Did Something Wrong.
Changing user behavior is fundamentally harder than just giving people what they want. And we were always fighting an uphill battle against established hiring patterns. It was a battle our ethos pushed us to fight, but it was often a losing one. We began with the premise that we were right, and that the goal was to convince users that we were right. That’s almost always an uphill battle.
After Triplebyte closed down, I spent a short time at one of our competitors, helping them get some of the assets they bought from us integrated into their product. I saw a very different approach there. They were an extremely user-, revenue-, and customer-driven organization, as much to a fault in that direction as Triplebyte was in our own direction. They begin with the premise of "what does the user want?" and work backward to get it for them, while Triplebyte began with the premise "what is the right way to do this in the abstract?" and worked backward to convince users to try it.
One of the key questions in my mind at the time: why were they alive when Triplebyte wasn't? It seemed important to answer that question, and to take philosophies very different from ours with deadly seriousness. And this difference in philosophy is one of the theories I came away with. (The other two are (a) that they simply got luckier with fundraising timing and (b) that, by nature of being an interviewing-as-a-service product, they avoided the candidate-acquisition problem entirely. These theories are not mutually exclusive, and all three may have played a part.)
In any case, "changing user behavior is extremely hard" is a pretty fundamental piece of product wisdom, and it is what we were trying to do.
What you should take from this
Doing business is hard. Some of these decisions were mine. Some involved me. Some did not; some of which I agreed with and some of which I hated. And I'm not, for the most part, going to say which is which, because the only thing worse than backseat driving is retroactive-seat driving. There is no shame in trying something hard and coming up short, and that's as true in business as it is on the job hunt.
I had more than my share of confident, well-justified, and wrong positions both as a lower-level employee and as a leader. And ultimately, I was a member of the leadership team of a company that failed. The buck has to stop with us, all of us, including me. None of this is about blame, and insofar as there is blame, that blame lies on me as much as anyone.
But what I hope is that all of this serves as a useful lesson. A bit of fluoride in the water supply of tech, to prevent the decay of the next company that tries to do what we did. A lesson in how a well-beloved company can fail not through any specific cataclysm but by a sequence of factors not obvious in any given moment.
And for me, a chance to try again.
