Empirical Estimates of Golden Handcuffs

Last post, I argued against common explanations for Golden Handcuffs.

Before we dive deeper, it’s worth asking, are they even real? Are employees of elite companies actually pathologically unable to leave? Is it just a stereotype with no grounding in reality?

Here are a variety of attempts to estimate the Golden Handcuffs effect empirically.

YC Founders Join Elite Universities, but Not Elite Companies

Rather than examining the psychology of employees, let’s just work backwards and count how many founders have big company experience.

A while back, I published a dataset of the top Y Combinator founders and noticed something odd.

Of the 26 founders, only 1 had held a full-time role at a FAANG company. [2] Two others had worked at Facebook, but only as interns. Outside of FAANG, only 3 founders had worked at companies I recognized at all. [3]

One explanation is the founder-as-monomaniacal-hero. Founders are totally single minded in their devotion, and would never do something as stupid as get sidetracked by prestige or status.

Except that they absolutely do. Of the 26 founders, 21 attended elite universities, mostly MIT and Stanford [1]. So maybe those universities are just incredibly good at fostering entrepreneurs, but more likely, they’re just good at attracting and selecting them. I take this as evidence that founders are not allergic to jumping through conventional hoops, pursuing instrumental goods, sitting down to take the SAT and so on.

It gets weirder when you consider population sizes. Google has around 100,000 employees, whereas MIT and Stanford undergrad programs are just 11,500 combined! And since average tenure at Google is under 4 years, each unit of headcount produces more alumni than a full degree program. [4]

All else equal, we should expect to sample many more founders from Google than from elite universities, but this doesn’t seem to be the case.

What’s going on?

Golden Handcuffs as Selection Effects

One explanation is that there are no golden handcuffs and it’s all selection effects. The kinds of people who work for Google are the ones who never intended to leave in the first place. The kinds of people who want to start great companies don’t have any interest in working for somebody else.

This is a reasonable explanation, but applies to less than half of the founders I looked at.

Only 11 of the 26 founders started a company right out of school. 14 have confirmed work experience, and another 2 have scrubbed the employment history, but have long gaps between graduating from school and founding their company.

Maybe working for a small startup shows that you’re less risk averse than a Google employee, but I don’t totally buy this. Brandon Leonardo (Instacart) worked at Webs, a company I’ve never heard of that claims to make “small business marketing simple”. Dan Kan (Cruise) worked at UserVoice, “the leading product feedback management software”. What exactly do these experiences select for?

In contrast, if you do want to start a company, working at Google seems like a great first step! You can meet co-founders and potential early employees, save money to fund yourself, gain legitimacy for investors and so on.

Of course, it’s possible the kinds of people who think about “gaining legitimacy for investors” are not going to start great companies. Maybe “real founders” don’t pursue instrumental goods.

A Third Perspective: Ex-Google Founded Companies

We’ve been dancing around the issue, but why not just go straight to the source and look at outcomes for the population we care about?

Are Google employees actually bad at starting companies?

Looking directly at startups founded by ex-Google employees valued over a billion dollars, we get:

  • Nutanix $5.8B
  • Cohesity $2.5B
  • Asana $4.3B
  • Lucidchart $1B+
  • Rubrik $3.3B
  • Instagram (sold for $1B, valued at $100B in 2018)
  • Pinterest $43B
  • Flatiron Health $1.9B
  • Xiaomi $46B
  • Affirm $2.9B

That’s a super impressive list! It doesn’t seem like ex-Google employees are bad at starting companies and pathologically incapable of quitting their jobs. Instead, the oddity is merely that they don’t attend Y Combinator.

That’s a different skew, and much easier to explain. Unlike regular YC founders, ex-Google employees may just be:

  • So wealthy that they self-fund until they can raise a Series A and aren’t willing to sell 7% of their company for $125k.
  • So tired of bureaucracy that they refuse to join an accelerator.
  • So credentialed that they don’t feel a need to go through Y Combinator to gain further credibility.
  • So well connected that they raise a Series A without going through Y Combinator.

Whatever the explanation, the fact remains that ex-Google employees do in fact leave to start companies.

Adjustments and Proportionality

Of course, we now have to ask if they do so proportionally. Even if Google has produced the founders of nearly a dozen billion dollar companies, we shouldn’t be impressed until we’re confident that they’re actually hitting above their weight.

Wikipedia lists 495 startups worth over a billion dollars, putting Google at 2%. But if you only include US based startups, Google is at 9 out of 122, or 7%.

To figure out the appropriate reference class, we’ll start with the number of software engineers in the US (4 million), then identify how many ex-Google employees there were 10 years ago when most of these founders started their companies.

A partner at google has compiled and released historical headcount data. This is a good starting point, but remember that we’re actually interested in the total alumni population rathern than point-in-time headcount.

To make this estimate, I assume 3 year average tenure and run a simplified simulation where 33% of the workforce quits at the end of each year, then Google rehires up to next year’s headcount.

Running this simulation from 2000 to 2010, we get that there were 50,000 ex-Google employees in 2010. They’ve previously said the workforce is 40% technical, so that’s equivalent to 20,000 engineers. [5]

Pitting that against the 4 million software engineers in the US [6], we get that ex-Google engineers were around 0.5% of the relevant population, but started 7% of the US based billion dollar startups. That’s a mulitple of 14x, and fairly good evidence that Google employees are not particularly bad at starting companies. [7]

Conclusion

Taking a step back, recall that what we’re actually curious about is not whether Google engineers are disproportionately successful at starting companies, but whether they even try in the first place. In other words, do they appear irrationally bound by Golden Handcuffs?

Compared to a randomly selected US based engineer, Google engineers have all sorts of benefits. They’re well credentialed, well connected, largely based in Silicon Valley, and supposedly smarter than average.

It’s entirely possible that their success rate is 140x that of a typical engineer, but they only try one tenth as often. The fact that Google engineers start successful companies doesn’t preclude the possibility that Golden Handcuffs are holding them back.

Further, consider that “rationality” in this case is really a function of both odds of success and opportunity cost. Even if Google engineers are less likely to start companies, it could be a perfectly reasonable choice given their relatively high opportunity cost.

So at the end of the day, the empirical data tells us a few interesting things, but can’t present a decisive conclusion.


[1] The other universities I count as “elite” are:

  • RISD
  • Joint RISD/MIT
  • 2 x Harvard
  • University of Waterloo
  • Berkeley, Columbia MBA

The ones I don’t include are:

  • USD
  • San Jose State University
  • Rice
  • Duke
  • Claremont McKenna College

I do include 4 founders who attended Stanford for grad school, but did not attend an elite undergraduate institution.

[2] Apoorva Mehta of Instacart spent 2 years at Amazon.

[3] Tony Xu of DoorDash was an intern at Square, and worked full time at McKinsey for 2 years, and eBay for another 2. Fred Ehrsam of Coinbase had spent 2 years at Goldman Sachs. Brian Armstrong of same had interned at IBM and Deloitte, then spent a year at Airbnb. Though notably, this was in May 2011, back when Airbnb was a small Series A startup with under $10 million in total funding.

[4] This ends up being a bit complicated. Reportedly, Google has an average tenure of 3.2 years with a median of 1.1. This is possible, but odd, and I’m not sure how they got these numbers. It’s tricky because at any given time, some population of employees has not left and you don’t know how long they’ll stay. If you count their tenure as their tenure-to-date, you’re undercounting how long they’ll actually end up staying. If you only count employees who have left, you’re skewed toward employees with short tenure. Also, Google has grown over the years, and we’re looking at founders who are successful now but started out 10 years ago when Google was around 25,000 employees.

[5] Data here. And yes, it’s possible the proportion of engineers has changed over time and this analysis is off.

This also helps explain why there are relatively so many founders from top universities. In the same 10 year time span, Stanford and MIT undergrad graduated around 115,000 alumni.

[6] Maybe we should be looking at the number of people who have been software engineers from 2000 to 2010 including retirements, and this number is actually higher. Assuming 40 year careers, and a constantly total number of engineers, it should be something like 25% higher, and Google is proportionately 25% better than it appears in the main text.

[7] There are lots of over corrections you could apply here, so don’t take this too literally. Maybe the relevant population is all people, not just engineers. In that case, Google ends up looking way better since the general population is much less than 40% engineers.