Your Metrics Are Bad and Why “Data Driven” Isn’t Enough

Being “data driven” is all the rage these days.

We all — businesses, government entities, sensor-equipped individuals — have more and more data that can help with decisions. The era of Big Data is here, yada yada yada: you know the annoying cliches as well as I do.

There are more and better tools. Dozens of startups are working on better ways to collect data, process it, query it, visualize it.

I recently talked with an entrepreneur who, fresh off of raising a big round of funding, was told by his investors that he needed to make his company more data driven. He wasn’t sure what “more data driven” actually meant, and he told me he wasn’t sure his investors did either.

It sure sounds nice, though — doesn’t it?

Honestly, I don’t know how I’d define “data driven”, and I’m not sure I care enough about the term to really think it through. But I’m pretty sure I know what’s missing.

Very, very few companies know what questions to ask of their data. They have metrics that are beautifully plotted on their real-time data dashboards. They’re calculated in technologically scalable ways, using something that’s much simpler than SQL, and they’re accessible by everyone inside the company.

But more often than not, the metrics are superficial and poorly thought through. They’re not reflective of the health of the product or the business.

I’ve certainly been guilty of this: for months if not years at my last startup, anything other than new user registrations barely mattered to me. For Circle of Moms, getting new users was extremely important, but at times distracted us from more important long-term goals.

And I see this again and again with tech companies. There’s a focus on one or two superficial metrics, rather than a deep understanding of what it will take to build out the broader ecosystem necessary to make the company successful.

I don’t want to be too negative: the understanding of these ecosystems has significantly improved in the decade-plus I’ve been in Silicon Valley. Ten years ago, entrepreneurs building consumer startups barely thought about distribution (if we build a great product, people will come to us!). Five years ago, entrepreneurs (myself included) started to realize that distribution mattered, but rarely took the next step (Facebook is the notable exception). Today, more and more entrepreneurs understand that both distribution and engagement matter, even if they can’t get at all of the underlying components.

Today, a few of the strongest consumer companies — Facebook, LinkedIn, Twitter — have built out growth and data teams that collectively measure and understand the key dynamics.

But there are still huge areas of our society — small non-tech businesses, government at all levels, medicine, academic studies, many startups — where there are lots of data, but not much understanding of what the data actually mean.

And that’s a big problem: I’ve long felt that having bad metrics is often worse than having no metrics at all.

If I were trying to gauge a basketball player’s skill level, my top preference would be to use a well-structured metric incorporating an entire season’s worth of extremely detailed, second-by-second data, looking at his impact on all aspects of the game. My second choice would be a good coach’s purely qualitative assessment of his skill. And my last choice would be a simple stat — say points per game — that was state of the art in 1950.

Today, most businesses are using the equivalents of the coach’s qualitative assessment and points per game to make their decisions. And quite frequently, “data driven” effectively means “we’re using points per game.”

Most of the new “Big Data” companies are focused on the relatively simple stuff: speed of processing data, ease of accessing data, beauty of data presentation. Those are all valuable, but they aren’t enough.

So how will the “bad metric” problem be solved? Certainly with some mix of better data training for everyone, plus tools that automatically discover and surface the important metrics. Both are important and I’m not sure whether training trump technology or technology trumps training.

Either way, if we want these new data to improve our collective decision-making, the good metric-bad metric problem badly needs to be solved.

How to Build a Quant Venture Fund

We’re Moneyball for carpet cleaning!

It’s easy to mock the mindless business metaphors of the day, and many of the “Moneyball for _____” ideas deserve that mockery.

But in the field of venture capital, there’s a huge opportunity for data — used correctly — to help investors make better decisions. Last month TechCrunch published a story suggesting that it’s already happening.

My somewhat informed guess is that most of what’s in the article is unsophisticated and/or vaporware. The “deep data mines” Leena Rao mentions are probably more like spreadsheets filled in by interns and/or Python scripts. Such spreadsheets — complete with stats from app stores, Alexa, and AngelList — would certainly be useful, but hardly qualify as analytically brilliant.

So what could an innovative VC do instead? I will walk through a model, mainly tailored for an early stage fund.

Data Collection

Probably the most important piece of building a predictive model is the collection of a great set of features that are likely to be predictive.

In the context of a venture fund, you’d likely want to collect a bunch of data at the time of an investment. You’d then have to wait a while (probably a couple of years) to see how those data predicted winners and losers.

For the data to be predictive, you need (1) underlying data structure that doesn’t change much and (2) a world where the same factors are likely to predict success.

That’s relatively easy if you’re trying to figure out whether a restaurant’s second location will be successful: you can look at a bunch of common metrics like customers, revenue per customer, employees per customer, demographics of the old location and the new one, etc.

It’s a lot tougher in the startup world, where the rules of the game are constantly changing. A million users today is very different from a million users in 2003; most of today’s important platforms didn’t even exist ten years ago.

That then begs the question: what are the “metrics” that are most likely to be predictive in the world of 2018 or 2023? To create a model that survives for more than a year or two, one would have to look at variables that will look similar in five or ten years.

Perhaps surprisingly, that means ignoring (or at least de-emphasizing) the variables of 2013 being tracked in that Python spreadsheet.

Instead, the model would use scores generated by (hopefully insightful) VCs themselves. Those VCs — and possibly other reviewers — would score each startup on traits like the following:

– how charismatic is the best founder? (1-5)
– how business-smart is the best founder?
– how tech-smart is the best founder?
– how well do the founders seem to complement one another?
– do you think the best founder could be “world class” at something some day (even if not today)?
– what about the worst founder?
– how well do the founders understand the market?
– how scrappy and hard-working do the founders seem?
– how would you rate the potential size of the business (independent of founders)?
– how would you rate the company’s user traction?
– how would you rate the company’s revenue traction?

This is just a set of example questions.

If you got the same (say) 5-6 people to rate all of those companies, you’d have 50+ data points for each startup. It might be the case that averaging scores (e.g., the average founder charisma score) would wind up being key. Or it might turn out that one VC’s “how scrappy” score is incredibly predictive while another’s was not at all.

Ideally, you’d do the actual reviews in a very standardized way: reviewers would either always talk to people in person for a certain amount of time, always talk to them on Skype, etc.

You’d also compile the answers to more straightforward questions: how many founders were there, how much revenue did they have, how many months had they been working together, had they started companies before, what colleges did they attend, etc. Some of those are probably predictive.

You’d have to do this for a bunch of companies (over 100, ideally a lot more), but not actually do anything with the data. i.e., if a firm invests in a startup and VC #1 rates the founder’s charisma a 4, business smarts a 3, etc., you just stick that in a database and let it sit.

The Model

When you build a predictive model, you use those inputs to predict something. Sometimes, what you’re trying to predict — who will win a basketball game, whether a customer will pay for your service — is fairly straightforward. In this case, it’s not.

The obvious value to be predicted is the long-term valuation of the company (or the fraction thereof that an investment would capture): the money it returns to investors when it folds, its acquisition price, or its valuation at IPO. This would, after accounting for details like liquidation preferences, dilution, and taxes, reflect the return for a potential investor.

It’s not necessarily clear, however, that this would yield the best predictive model. As any VC knows, returns are shaped in large part by one or two exceptional wins: many funds have one very successful investment that provides the majority of returns. If there are only a few such companies a year, even an all-knowing VC is unlikely to have the data scale to make a good predictive model.

Instead, an investor might look at a few different scores:

1) How good did we think this investment would be when they first pitched us?
2) How good did we think they’d be 2 months after we invested? In two months working together, the VC should hopefully learn a lot to tell her if this was a sound investment. This is probably only a useful data point for actual investments, as opposed to rejected deals.
3) How good did we think they’d be 1 year after we invested?

etc., etc.

It may be the case that VCs (et al) are pretty good at #2 but not so good at #1. To gauge that, they could build a predictive model to predict how they’ll feel a few months in, given the dozens of measurements they took pre-investment. Within a couple of months, they may be able to more effectively forecast the likelihood of intermediate success potential. This intermediate gauge is subjective but still more thorough than what’s being done: what predicts how I’ll assess this opportunity when I really understand it more deeply?

Because returns are driven by a few outliers, asserting an exact expected long term valuation for early stage companies is difficult. If a promising but young company has a 0.1% chance of some day having a $100 billion valuation, they’re ostensibly worth at least $100 million. If, however, those odds are only 0.001%, they may be worth as little as $1 million. And a venture firm has virtually no chance of having enough data to distinguish between 0.1% odds and 0.001% odds.

However, a venture firm should have enough data to distinguish between larger tiers of companies: companies at the 98th percentile or higher, companies between the 95th and 98th, companies between 90-95, companies between 75 and 90, and everyone else. Some of the companies in the top tier — defined by competence and potential more than ultimate outcome — will fail; others will be incredibly successful and go public.

The goal of this model would be to predict the tier, rather than the ultimate outcome.


Step one of this process is about recording data, not about changing decision making. And step two would be to take months or years of data and build a model.

Only after months or years, when the firm has actual outcome data and a predictive model, would it actually use the model to help with decisions.

An early stage fund that makes dozens of investments a year is best positioned to execute on this strategy. There are two reasons for this:

1) The metrics defined are more about people and potential, and less about business fundamentals; except in rare cases it would be irresponsible for a late-stage fund to give a company a $100 million valuation based only on those scores.
2) They have a lot more data.

All of the “human” factors that make an angel fund successful today — dealflow, partners’ skill at evaluating founders, helpfulness to entrepreneurs — would be equally important in this sort of fund. However, a formalized predictive decision-making process could improve returns significantly.

See also: Data Scale: Why Big Data Trumps Small Data

Why ‘A’ Players Should Collude

They should have been paying me more.

That was my thinking, at least. I was building predictive models for PayPal to detect fraud, and my models were effective at saving the company money. From 2002 through 2004, that work was likely saving PayPal at least ten million dollars a year.

Knowing how much I was helping the company, I figured I’d have some pretty sweet leverage if I ever wanted to try to negotiate a better pay package. My work was easily quantified, and it was making a big difference.

If my models were capturing an extra ten million bucks, surely the company could reward me with half that, or at least a quarter of it, right? Especially since PayPal was being valued relative to earnings; at a multiplier of 30x earnings, those models had created an extra $300 million in shareholder value. Even a few percent of $300 million would make my day!

(At the time, I was not the type to “lean in” and negotiate harder, so that never came to fruition — but that’s a separate story.)

The Non-Zero Baseline

Sadly, my thinking was flawed. My assumption was that the baseline — the business equivalent to the “replacement” player in baseball was a business that was exactly break-even. I could add $10 million in profits, four others could do the same, and the company could effectively divvy up those $50 million in profits among the five of us.

The problem is that the baseline for PayPal — what would have happened with average players at every position — wasn’t zero. The baseline was a business that was losing hundreds of millions of dollars a year.

So I was saving the company $10 million per year, so were perhaps nineteen of my colleagues. My statistical models were ‘A’ work and a huge improvement over the baseline, but so were our fraud policies, our viral growth channels, our eBay integration, our legal maneuverings, and many other areas.

Without all of those accomplishments, PayPal would have gone on losing hundreds of millions of dollars until we went out of business. If the baseline of mediocrity was a business that lost $180 million per year, adding twenty strong people (or teams) whose “above replacement” value was each $10 million per year could collectively improve our bottom line by $200 million — but that would still only lead to a business that made $20 million in profits.

In that sort of business — which is roughly what the pre-acquisition PayPal of 2002 looked like — my original reasoning made no sense.

Collusion Is Good

Recently, I’ve been evaluating a new startup idea. It’s something that’s as ambitious as PayPal, but also just as fraught with challenges: there are a lot of ways that it could lose money. And that’s encouraged me to revisit some of my PayPal memories.

One of my biggest lessons is the importance of team quality for such a difficult and ambitious company. Looking back, it’s as if a bunch of smart and hard-working people colluded and decided to work together on a business that would have failed if we hadn’t all worked on it. Without that collusion, PayPal wouldn’t even have a Wikipedia page, let alone a huge business or a mafia.

2002 was the ideal time for that sort of collusion, because there weren’t a whole lot of options in Silicon Valley. No one was starting their own company, and the list of hot private companies I was aware of had exactly two entries: PayPal and Google.

To solve the toughest problems — and building a slightly more elegant social networks doesn’t qualify — one still needs that type of collusion from ‘A’ players. The good news for founders is that Silicon Valley is attracting more talent than it was in 2002; the bad news is that starting a company has become cool again (I’m guilty too!) and the hot company list has grown from two to dozens.

Collusion generally has a negative connotation, but in this context it can be a very good thing. If, rather than spread themselves among ten mediocre companies, ten all-stars can be like LeBron (and Dwyane Wade and Chris Bosh) and collude, they can see better results and solve bigger problems. And unlike LeBron, they don’t have to do it in zero-sum games.

Happy 10th Birthday, LinkedIn!

Ten years ago today, LinkedIn was born. It’s radically changed my life over the past decade.

I had no knowledge of LinkedIn’s existence on the day it launched in 2003. But a few days later, I got an invitation from my former PayPal colleague Keith Rabois to join this new site, LinkedIn. I signed up (as user number 1400 or so), saw a few familiar names on it, and was immediately intrigued.

In the months to come, I used LinkedIn a handful of times, connecting to colleagues and meeting a couple of entrepreneurs who reached out to me for advice on fraud prevention.

In late 2003 or early 2004, my former PayPal colleague Lee Hower reached out to me to see if I “knew anyone” who might be interested in working on data-type problems for this company LinkedIn. Lee and I had lunch, and I learned that LinkedIn had built its network out to be a couple of hundred thousand users. This was pretty cool, and something I could certainly see working on.

A week or so later, I was sitting in a Mountain View conference room with Reid Hoffman — with whom I’d spoken exactly once when he was a PayPal exec — and Jean-Luc Vaillant, brainstorming cool stuff we could do with data.

Not long after that, in February 2004, Jean-Luc set me up with a Mac laptop, and I started delving into the data. For the next few months, I was essentially moonlighting at LinkedIn, coming into the office one or two days a week, while mostly still at PayPal.

When I found that I was consistently more excited to get out of bed on the LinkedIn days than on the PayPal days, I decided I would (after biking around France for a month!) join LinkedIn full-time.

Thus began my LinkedIn employment odyssey. I worked full-time at LinkedIn from October 2004 until January 2007, leading a small analytics team with two awesome hires, Jonathan Goldman and Shirley Xu.

I learned a ton at LinkedIn, worked on some interesting and important products, and got to collaborate with lots of great people I now consider friends (and of course, LinkedIn connections).

After I left, my team became the large, influential and highly-regarded (kudos to Jonathan and DJ Patil) Data Science team.

Though I was no longer employed at LinkedIn from 2007 on, the company has continued to play a crucial role in my life. The first two engineers we hired at Circle of Moms, Brian Leung and Louise Magno, came in through the same LinkedIn job post in 2008.

Looking back at the Inmails I’ve sent, I see a number of people I reached out to try to hire and now know well; in many cases we didn’t wind up working together, but they became friends and valuable connections.

LinkedIn has prepared me for meetings with hundreds and hundreds of people. I wish I still had database access so I could run the query to figure out just how many.

When I first started poking around LinkedIn in 2003, I had a couple dozen connections. I looked at the profiles of people like Lee and Reid, seeing well over 100 connections. I figured I was simply not the kind of person who’d ever amass that number of professional contacts.

Today, I have over 900 connections on LinkedIn; the vast majority of those are people I’d feel comfortable reaching out to for an important professional purpose. Part of that increase is a reflection of my evolution, but a lot of it is thanks to LinkedIn.

To Reid, Jean-Luc, Lee, Allen, Chris, Sarah, Matt, and the millions of others who have helped to build LinkedIn, thanks and happy birthday!

In Search of a Happy Medium Between Academic Papers and Shlocky Business Insider Top 10 Lists

Here’s the state of the art in home page design for an academic website:

And here’s the state of the art in home page design for a rapidly growing media site:

Suffice it to say that I don’t want to emulate either the blandness of the first interface or the tabloid feel of the second one.

Now that Numerate Choir is a year old, I’ve been thinking about these extremes of online content, and trying to figure out where sites like Numerate Choir fit.


In the world of virality, headline usually matters more than content: the headline is what people click on, and it strongly influences what people share.

In the world of virality, short, simple, and universally entertaining is key. How something is presented matters more than what is presented. That top 10 list can be utter crap, thrown together in five minutes, but hey, look at the pretty slides! And my friend is #8 on the list!

The world of virality is superficial, but it has one very valuable characteristic. It understands what normal people value, and it concerns itself with what they will respond to in the real world. It doesn’t concern itself with “could” or “should”; the key term is “actually does”.

Art and Academia

In the world of art and academia, depth and truth should matter above all else. Headline is superficial: what matters is who reads something, not how many people read it.

In the world of art and academia, entertainment is looked down upon. It’s far better to be profound — and full of jargon — than amusing or captivating.

The world of art and academia is also superficial, but in a different way: superficiality is manifested as the desire to prove one’s sophistication. Those in the academic and artistic sphere generally don’t understand how to create content for the masses, but they are far more likely to discover things that matter.

Numerate Choir and a Little of Each

As I wrote about in my post on the state of journalism, there’s a growing dichotomy between 1) pseudo-news that can be both popular and profitable, and 2) deeper content that often fits better into a non-profit model.

I have a love/hate relationship with both of these extremes. And my work on Numerate Choir has oscillated between the two.

Like an academic, I write blog posts that I — and anyone with editing experience — knows are way too long even for a geeky Silicon Valley audience. And then I float to the other side: how can I craft a headline that will maximize sharing on Twitter? And why can’t I easily a/b test the stupid thing?

Like most academics, I do a poor job marketing and selling my work. I naively hit the publish button and hope for the best: beyond a quick post on the major social networks, I do nothing to publicize my writings. But then like the growth-loving impatient exec I criticize, I keep a close eye on Twitter to see if any of the cool kids have shared my new post.

Like an academic, I hope that what I write will help people understand an issue and maybe think a little more deeply. Like a growth hacker, I do care about how many people read them (even if I recognize that they aren’t for everyone).

It’s amazing yet discouraging that my most read post — by a factor of two — was a relatively lightweight rant on not needing a real-time dashboard. That post was written in response to a friend’s emailed question; I wrote and published the whole thing in under two hours.

(Credit where it’s due: that distribution was helped in large part by Andrew Chen cross-publishing it on his blog — thanks, Andrew.)

I often write with a lofty goal: capture some of the truth and quality of academia done right, and reach a larger audience. Sometimes that’s worked: The Visionary and the Pivoter and A Founder’s Constant State of Rejection (also on were both read by lots of people. Two other posts I was especially proud of, Why Big Data Trumps Small Data and In a Data-Driven World, Honesty is the Fundamental Virtue, were far less successful metrics-wise.

My blog posts have generally followed one of two patterns:

  • I post something, then link to it on Facebook, Twitter, LinkedIn, etc. A few of my friends read it, and one or two of them might retweet the link. The post is seen by at most a couple thousand people.
  • I post something and link to it and a few people click through. That’s the “soft launch.” One or two of the clickers happens to be a friend who’s an order of magnitude better known than I am. When he tweets it out himself, we have a “real launch” (thanks Naval/Keith/Andrew/Eric/Jeremy/Dave). That tech celebrity share serves not just as a driver of traffic but as social proof to others: this is not a sleazy Business Insider post.

Hence Twitter is essentially an oligarchy: a handful of people have most of the power. While many could exercise it better, it winds up being an acceptable model to propagate this sort of semi-intellectual, semi-popular content.

Metrics for Success

What metrics define success for this blog?

The business model for most online media sites is pretty simple: visitors and page views translate directly into ad dollars. In most cases, the revenue per user is constant: it doesn’t matter if the viewer is President Obama or my one year old daughter banging on the keyboard.

For this blog, revenue is and will be zero, regardless of how many readers there are. I’m not sure what the most important metric is; I can think of at least five that matter to me.

Several times, I’ve written about topics I’ll almost surely never touch professionally. When I do this, it’s often with the hope that somehow it will reach someone far more influential than I, and affect a positive change on the world.

Other times, I’ll write to crystallize my own thoughts. If I get great feedback, that’s nice, but I’m writing more for myself.

Sometimes, my posts can be a good way to tell a bunch of my friends about my experiences; they then inspire deeper conversations about interesting topics.

A couple of times, I’ve been able to link people I know to a post I wrote explaining my thoughts on a particular issue. This means I don’t need to write it up again and again. As any engineer knows, re-usability is good.

Finally, I hope to share my personal learnings with others. I can think of only two places where a 35-year-old can be considered wise: a society where the life expectancy is 40, and Silicon Valley. I’ll drift off that sentiment if I can.

In a Data-Driven World, Honesty is the Fundamental Virtue

In times of war, there are no greater virtues than loyalty and bravery. A country with disloyal citizens is likely to lose every battle. Bravery is an essential trait to overcome the harshness of war.

Likewise, for most of human history, sexual purity was promoted as an essential virtue. Because contraceptives were not an option, a promiscuous society would be one with frequent unwanted pregnancies. As a result, many societies developed strong cultural norms to discourage physical relationships before marriage.

Today, the world is relatively free of wars and effective contraceptives are widely available, so these traits are valued less. Our culture constantly re-evaluates its norms.

Meanwhile, as more and more of our lives are recorded, the data we collect facilitate better decision-making. I’ve written about many aspects of that: data help journalists find better stories, help predict the future, and much more.

However, a data-driven society is only functional when people follow the right cultural norms. In a country at war, a culture of loyalty helps ensure that everyone is in line. In a capitalist society, a culture that discourages theft can allow small businesses to prosper without fear of losing property. In a data-driven society, we must stay intellectually honest.

Without intellectual honesty, the data are flawed and unreliable. Flawed data lead to poor decision-making; it’s usually better to use only your gut than to rely on a poorly formed data set. And unfortunately, many people are using data in intellectually dishonest ways.

Schools and Cheating

In a data-driven world, we must not cheat.

One of this year’s Goldsmith Award finalists is an astonishing data-driven series which uncovered high levels of school cheating. I’m proud that we honored that story, but I’ve been somewhat taken aback by some of the responses I’ve heard to it.

Many people I’ve spoken with, when told of this investigation, immediately blamed the reward structure. No Child Left Behind, they say, created an overly pressurized education system. The sort of large scale cheating exposed by the Atlanta Journal-Constitution was inevitable given the high stakes of the tests.

That is nonsense. One wouldn’t excuse a CEO stealing money from others because there was so much pressure on him to improve his company’s performance — even if the CEO thought the means of evaluating him were unfair. No Child Left Behind and other education policies aren’t perfect (my suggestions), but they’re a starting point.

To improve, we’ll need to refine our testing system and get better at measuring progress. We’ll also need a culture of integrity from teachers and administrators. Without that integrity, our system will consist of results we can’t trust — and a terrible example for students.

Guns and Intellectual Curiosity

In a data-driven world, we must approach major issues with an open-minded, intellectually curious approach.

Of course people use data dishonestly for political arguments. But it’s not just sleazy politicians: I see intelligent friends on both sides completely misrepresenting the data on gun violence in the US. Anti-gun advocates point out that the U.S. has more gun ownership and more gun deaths than other Western countries and jump to the “obvious” conclusion that more guns means more violence. Pro-gun advocates point out that the U.S. has far higher gun ownership rates than many (non-Western) countries that are much more violent than they U.S.; they jump to the also “obvious” conclusion that criminals will find a way to purchase guns regardless of gun policy, meaning decreases in gun ownership would have no impact on gun violence.

It’s likely that each side is at least partially correct. Millions more guns in Americans’ hands mean at least a few more deaths; many of the most violent criminals will find a way to kill regardless of gun laws.

Yet my friends who post these stats do so with a lack of intellectual curiosity. In most cases, they haven’t looked at the numbers with an open mind, and they don’t really understand how gun dynamics work. I’ve never seen, for instance, someone point out that gun ownership in states is highly correlated with suicide rates but minimally correlated with murder rates. That fact — which implies that less restrictive gun laws may lead to suicides but not terrible crimes — doesn’t fit neatly into anyone’s pro- or anti-gun view of the world.

I’m realistic: I don’t expect that everyone is going to gather data sets on gun violence on their own. However, because the data are out there, I ask smart people to raise their bar: if you haven’t looked at the data closely enough to have an informed, nuanced opinion, please keep quiet. Either do some real research, or don’t spread your uninformed perspective.

Pitching Investors and Misleading

In a data-driven world, we must not mislead or be misled.

Working with many early stage companies, I see a lot of investor pitches.

Startups have gotten better at crafting an appealing pitch, by throwing out numbers like these:

  • a) We’ve increased revenue by 25%, month over month
  • b) Our user base is growing: we had 500,000 users last July and now have 800,000 (often with an attached graph showing total users at the end of each month)
  • c) Our monthly retention rate is 75%

Most of the time, these stats aim to mislead.

(a) doesn’t have a baseline or a time horizon. It might mean that revenue was $12 last month and is $15 this month. Does that sound as impressive?

(b) looks at cumulative signups rather than monthly signups. Cumulative signups are always going up, and it takes a lot more effort for the viewer to see whether the second derivative (signups this month versus last month) was positive or negative. We did this in investor presentations for Circle of Moms, because we knew it would obfuscate some of our negative trends. I advocated that approach, and I don’t feel great about it.

(c) may be a useful stat, but it’s almost always calculated in an obfuscated, company-friendly way.

These kind of pitches are “just the way it is” — as is the case for misleading political data analysis. But in a data-driven world, we need to aim higher.

Entrepreneurs should assume their audience is intelligent and mature, cognizant that not all numbers go up.

And investors should understand what they’re looking at, and should call BS on entrepreneurs who surface their numbers like this.


Access to data is, on the whole, a very good thing. Deep, data-driven knowledge allows us to make better decisions and preserve resources. With the right data, we can better reward the best teachers, fund the top companies, and create better public policy.

However, for that to work, our society needs to create a stronger culture of honesty around data. We can’t cheat to get around failures. We must seek out all of the facts, and not promote only those that fit into a narrow ideology. And we must use data to inform rather than mislead. If we don’t do those things, we’ll make decisions that are driven by flawed data — lies — and many will suffer.

The good news is that we’re still early in an age of data-driven decision-making. Our collective culture has developed to better discourage practices like stealing and killing. In this wonderful age of data and better decision making, can we become more honest?

Journalism, Through the Eyes of a Data-Focused Entrepreneur

Fueled by wine and delicious food, the table was full of energy. The journalists and politicians at my table were eager to outdo one another. They exchanged candid personal stories about famous TV newsmen and potential presidential candidates. They recounted tales of the shocking political corruption they’d uncovered. They told their colleagues what had really gone on at that recent big event. Unable to compete with their stories, I nodded politely.

Somehow, I’d found myself at the upscale Rialto restaurant in Harvard Square, feeling like a fly who’d landed on the wrong wall. I was twenty-four, a quiet and unassuming fraud R&D scientist at a small Silicon Valley startup called PayPal. The eight or nine people at my table were quite unlike me: all big names from journalism and public policy, all far more extroverted than I, all at least twice my age.

It was January 2002, and I was sure that two worlds couldn’t be any more different. Here in the journalism sphere were politicians and journalists, extroverted and boisterous, who told great stories; back at PayPal, analytical, introverted nerds whose skills were mostly technical. At the dinner table in Cambridge were those who could talk to sources and get the scoop; back in Silicon Valley, engineers who automated processes and crunched data.

Within about a decade, all of that would completely flip. News organizations would embrace the data-heavy, analytical approach more common to tech companies. Many of 2013′s top stories would use data at a level that was unfathomable in 2002.

The Goldsmith Awards

My first journey to Cambridge was set in motion only a few days before that dinner, when I got an urgent call from my grandfather.

Can you be in Boston on Saturday to help select the winners of the Goldsmith Awards?

I had only a vague sense of what the Goldsmith Awards were; why exactly should I fly across the country on a few days’ notice?

The Goldsmith Awards, he explained to me, were something he’d worked with the Shorenstein Center at Harvard to set up, using assets from the estate of the late Berda Goldsmith (his legal client). The awards honored great journalism, but their true goal was to foster better public policy. He wanted to reward journalism that shines a light on government, highlighting bad regulations and bad policymakers for the benefit of ordinary citizens.

Being a lawyer who was both thoughtful and crafty, Bob put in place a contract that would maintain close ties between the foundation and the Shorenstein Center. A key clause in the contract stated that one of the award’s judges must be a foundation representative.

Bob had called me because he wanted me to take over as the foundation representative. He hoped that I could go to Cambridge on Friday to see how everything worked. Then the following year, I could represent the Greenfield Foundation’s on the Goldsmith selection panel.

Of course, I said. I’d long been interested in public policy — I minored in Political Science at Stanford — and this would be a real honor.

2002: Reporting

At the judging session — before the dinner at Rialto — I got my first jolt of culture shock. The Goldsmith panel of judges evaluated dozens of newspaper submissions, and their criteria were often a surprise. I’d long been a consumer of the news; that day I got to see a news professional’s perspective for the first time.

Some stories were very impressive on the surface: engaging, in-depth, surprising reports on a policy topic I knew nothing about. It turned out, however, that they closely resembled another story told earlier by someone else. The first story a journalist told about toxins in the local drinking water was probably very impressive; the twelfth such story — reported using the same template as the first one — doesn’t deserve an award.

More stories were dinged for reasons I wouldn’t have fathomed as a mere news consumer. Some were largely the product of a single leak: they came from an insider who wanted his story told, rather than from sleuthing by the reporter. Others were impressive investigative feats, but pointed to flaws in public policies which had virtually no chance of being changed.

The best pieces that year were original, impressive in the depth of their investigations, and had substantial impact on policy. The winner, about hospital care at the Hutch in Seattle, stood out for the sheer amount of manual work it required: the reporters had to wade through “100 interviews and 10,000 pages of documents” to tell their story. The story was amazing, and was notable to me for the set of skills used by the reporters: their methods were a far cry from the algorithm coding I was doing at PayPal.

2013: Reporting and Data

Serving on the Goldsmith panel soon became a tradition for me. Last week, for the twelfth time, I found my heavy winter jacket in the back of the closet (it’s useless in the Bay Area) and packed up for a January weekend in Boston. I’m now the veteran; this year I served with several people who had never judged a competition like the Goldsmith Awards. I still have yet to judge with a panelist younger than me, but I no longer elicit the “what’s that little kid doing at the table?” stares I saw a decade ago.

But that change is predictable: I knew I wouldn’t stay twenty-four forever.

The big surprise is that investigative journalism, so different from my PayPal day job in 2002, now feels like a natural project for a Silicon Valley startup data guy like me.

Journalism has changed a lot in the past decade. In 2002, almost all investigative stories were anecdotal. A story about ineffective education started and ended with interviews of teachers, parents and students. The investigation about medical treatment told stories of the travesties patients had endured, without using terms like “probability” or “percentage”, let alone “false positive”. Occasionally, a Goldsmith submission would talk about the painstaking work that reporters had done to piece together hundreds of paper records to assemble some basic statistics.

Today, by contrast, data analysis plays a huge role in many of the top stories. Of this year’s six finalists, at least three would have been unlikely or impossible ten years ago:

  • Cheating our Children, from the Atlanta Journal-Constitution, is a story about cheating by teachers and schools on standardized tests. The team looked at thousands of districts across the country for highly suspicious anomalies, like every student in a class (supposedly) erasing an incorrect answer to question #27 and then filling in the correct answer. They found several hundred patterns of student improvements that were most likely the result of fraud.
  • State Integrity Investigation, from the Center for Public Integrity, looks at the laws of each of the fifty states and grades each on their risks for corruption. To do this, a reporter in each state perused that state’s practices and regulations — a far more manual approach than Cheating our Children — and assembled a database of information about that state. The end result is both a great way of pressuring states (Utah, don’t you want to improve your D?) and an incredible Wikipedia-like online resource for others (especially journalists) interested in tackling related topics in the future.
  • The Shame of the Boy Scouts, from the LA Times, is the sad story of thousands of incidents of child sexual abuse records in the Boy Scouts. The Times pulled together thousands of newly released Boy Scout child records, using them to tell many unbelievable and sad stories about children who were molested. But the series complemented those stories with a feature that would have been unlikely a decade ago: they posted all of the documents online, for anyone to search and see.

These new data-centered stories are distinguished by three new attributes. The first relates to how a story was uncovered: many stories today are initially found not from a tip, but from a database search. In Cheating our Children, cheating was uncovered not because of a tip from a parent or a teacher, but because of a search for suspicious trends in the data. The steps to get the story were (at a high level) similar to what I was doing at PayPal in 2002: using algorithms to identify a handful of likely fraudsters.

The second data attribute is quantifiability. Historically, journalism has not been a quantitative field, relying instead on an anecdote or two, along with an assumption that “there are many others like them”. And while quantification would be silly for many stories — either Nixon’s people broke into Watergate or they didn’t — it’s an important part of any broader societal story. In 2011, the Goldsmith winner informed the public of local hospital practices that were quantifiably worse than others out there. This year, there were some great not-quite-finalist stories that found and measured the effect of cops speeding and explained just how harmful overly prevalent pain medications can be.

Finally, many of the top stories today are complemented by a structured, searchable database. Each of the three stories above features an interactive tool allowing anyone to find the information most useful to them. On, I can look at my local school district for evidence of cheating; on I can see how my state fares with respect to corruption risk factors; on I can see whether there were any reports of sexual abuse at a specific Boy Scout troop.

Though the world of journalism has its challenges, these are three great developments. They widen the range of stories journalists can tell, they raise the bar on their quality, and they make them individually relevant to the reader.

The Great Bifurcation

The landscape of news and other content is bifurcating, with increasing separation between work that aims to be high-traffic and work that aims to be high-impact. On one side is entertaining content, aimed at driving page views. That content may be news, opinion, or something else, but its goal is very simple: to be part of a traffic machine that underlies an ad-supported online operation.

Traffic machine content is most successful if it arouses curiosity (yes, I do want to check out the six ways that olive oil can help me lose weight!) and can be even more so if it’s also something the reader identifies with and wants to share (this is why people who voted for my political candidate are smart!). That high-traffic story, while cheap to produce, is usually not especially deep or insightful, and it may not even be true. Thus it has little or no positive impact on our institutions.

The other side of the coin, high-impact journalism, is a very different animal. It takes a lot of work and money to produce, but often doesn’t generate a lot of traffic. It may have a great impact on society, but it’s tough to justify for a business. And that’s why, increasingly, it’s the domain of non-profit entities and organizations that are only nominally for profit.

Journalism via a non-profit can be a good thing: those organizations — who today operate at both local and national scale — can focus on the highest impact work rather than try to mix unpopular high-impact stories with popular low-impact ones.

2023: Reporting, Data, and Software

In the new non-profit news organization, there is a simple question to ask: “how can we do work that will have the largest positive impact on public policy?” That is essentially the same question my grandfather asked when he set up the Goldsmith Awards over two decades ago.

To understand how that question will be answered a decade from now, one must first understand the roles played by three different people in today’s professional world:

  • The investigative reporter skillfully combs through documents and asks the right people the right questions to find information. He then turns that information into a compelling story for his audience.
  • The data scientist takes the data available to her and mines it to quantifiably understand a subject. With words, numbers, and data visualization, she shares — usually with less verbal skill than the journalist — that understanding with others.
  • The software developer takes a process that works manually, and figures out how to first generalize it and then automate it. For instance, if you have a meeting in your calendar with an accompanying address, how can software automatically send you directions at the appropriate time? A human can do it manually; the developer writes the software that will automate the process many times over.

A decade ago, the stories I read for the Goldsmith Awards were solely the work of reporters from the first group. They were executed by skilled journalists who knew how to comb through documents, convince insiders to give them secret information, and write stories elegantly.

Today, the data scientist is a key part of journalism: data skills are nearly as important for producing Goldsmith-caliber work as classic investigative skills. Data skills help both at the early phases of a story in finding anomalies worth writing about, and in moving beyond anecdotes to show that trends can be quantified. That anomaly-finding helps increase the range of stories that can be told; quantification makes the stories better.

Still, today’s journalism has a one-off quality that would frustrate a typical software developer. Sure, I can read a story about cheating in schools — or even look at how it affects my hometown — but will the story be automatically updated in three years so it’s still relevant?

In the next decade, it’s likely that we’ll see investigative reporting evolve and improve in several ways:

  • More and more journalism will be automated and updated regularly. District scores will be mined every week; state corruption will be automatically assessed monthly. In some cases, there will be written stories that complement the new data; in other cases the automated jobs will simply feed into an interactive database available to readers.
  • Investigative reporters will get better at soliciting information from their readers and viewers. It’s become a lot easier for readers to contact reporters with tips than it was a few decades ago, but there’s still a lot of room for improvement. Facebook, LinkedIn, Quora, and Twitter make it easier to find and contact the person likely to know a specific piece of information, but they’re not ideal. One could, for instance, imagine a world where citizens could record any suspicious or unacceptable government actions in a form that could be searched by reporters in the future; this would markedly improve many investigative stories.
  • The number of journalists with data skills is increasing rapidly, and that isn’t going to change any time soon: my Twitter feed is filled with data+government+journalism enthusiasts from many different backgrounds. They’re offering online courses, pushing for open data, and a lot more.
  • More and more data — particularly from governments — will come online. The picture today is awful: most government documents are still posted in unstructured form as PDFs and Word docs, making data analysis a lot tougher. That will change.

These changes will allow journalists to more quickly find important stories and tell them more accurately. At a time when some news organizations are slashing budgets and others are defining themselves, that’s important.

Merging Worlds

When I went to Harvard eleven years ago, I couldn’t help feeling like I didn’t quite belong. It was an honor to be part of the Goldsmith Awards, but I was there because I happened to be the grandson of the awards’ founder.

This year, I flew to Boston a day early and spent time with Alex Jones, the longtime Director of the Shorenstein Center. As always, I learned a few things. Alex told me about Journalist’s Resource, a great online tool which lets journalists freely access research on complex topics. He highlighted the increasing role of data in journalism and among many of the top Goldsmith Awards contenders.

While there, I also chatted with Nicco Mele and John Wihbey, both staff members at Shorenstein. Nicco lectures on technology and simultaneously runs a web consultancy; John is the developer behind Journalists’ Resource. Both were full of ideas on how data, journalism, and technology can come together to improve public policy, telling me about cool projects like Journalist’s Resource and Nearby FYI. It was inspiring, and in my conversations I saw a new take on my grandfather’s vision for journalistic impact.

That Saturday night, after a full day selecting the Goldsmith finalists, seven of us met for dinner at Rialto. I was still the introverted techie, and I still didn’t come armed with personal stories about Clintons or Bushes. But having just discussed such a strong set of data-heavy stories, I knew something was different. The landscape has shifted, and journalists have caught on to many of the skills my friends and I value in Silicon Valley. Once just a fly on the wall, the data geek is now an important part of the story.

Thanks to Ben Greenfield for his great feedback on this post.

Six Steps to Growth: What I Learned as 500 Startups’ Growth Hacker In Residence

I just finished up a six month stint as Growth Hacker In Residence at 500 Startups. I want to share a few of my learnings.

But before I do that, a couple of things to get out of the way.

First of all, 500 Startups is a strong operation and an amazing story. If you’d just arrived in Silicon Valley yesterday, you’d probably have no idea that it’s only a couple of years old. In public, Dave McClure is a little zany/off the wall/all over the place, but that’s far from the entire story: he understands the venture world, he has good ideas on how to shake things up, and he’s built up a good team that complements his skills.

500 Startups has two major unfair advantages: a huge network of talented mentors helping portfolio companies (with many others clamoring to mentor) and impressive ties abroad — incredibly valuable for deal flow, fundraising, and partnerships. 500 acts like a growing startup rather than a greedy venture firm: with a small fund, Dave opts to hire more employees who will help startups, rather than keep extra money for him and the other partners. Venture capital is going to get shaken up over the next decade or two, and 500 Startups’ strong network and ungreedy, open-minded approach will put them in good position to thrive.

Second, any thoughtful person who understands how to grow an Internet business cringes at least a little bit at the term “growth hacker.” The hype-to-substance ratio has become high: I suspect that a high percentage of the 338 people who show up in a LinkedIn search for “growth hacker” don’t have any idea what they’re doing.

So it was with a tiny bit of reluctance that I accepted the “Growth Hacker In Residence” title. For better or worse, 500 is the ultimate anti-corporate firm, so I don’t have any business cards to implicate me. But it all worked out well.

Why Growth is Important

I didn’t push back on the title because there is in fact a lot of substance behind the term growth hacking, and hype isn’t always a bad thing. The underlying skill it describes — increasing a product’s usage and distribution — is incredibly valuable in consumer technology companies. And contrary to the complaints of some, growth is not just product marketing by another name: skills in product design, engineering, and data analysis are just as relevant.

To see why growth is so important, one need only look at the big successes of my former colleagues at PayPal (aka the “PayPal Mafia”): YouTube, Yelp, Yammer, and LinkedIn. All succeeded in large part because they were designed for growth. The ability to embed videos on MySpace separated YouTube from the pack; Yelp mastered SEO for traffic and retained a small core of “elite” reviewers to build content; Yammer made it exceptionally easy to invite colleagues; LinkedIn made connecting with professional contacts seem like an obvious and important need.

I don’t think this is a coincidence. PayPal had to scrap and claw to grow and survive; its alumni figured out how to grow their next companies the same way. By contrast, early Google was a company with more impressive technology that didn’t have to scramble to grow. Its employees likely learned less about growth while at Google, and those alumni have since had far less success with distribution and startups in general.

In that respect, most entrepreneurs are more like the Google non-mafia than the PayPal Mafia: they don’t bring to their startup a strong understanding of how to grow and maintain a large base of users. That lack of understanding is true of most 500 Startups companies — as it is for the companies of every investor. Fortunately, that skill is — to an extent, at least — teachable.

Helping Companies

So how could I help the teams of 500 Startups?

For one, I could help by setting the right context: growth is at least as much discipline and execution as creativity.

It turns out that there aren’t usually easy solutions to growth challenges. This is where the hype around the term “growth hacker” can be dangerous. Hacking implies something weird and unpredictable. It evokes a sense that the perpetrators are more crazy artists than precise scientists, that their methods are unusual and tough to replicate. And it implies that there’s some sort of obscure code that can be cracked to yield the magic growth solution.

If that were true, there might be a simple but hard-to-find switch to flip. This is almost never true. “Change the button color to green, and you’ll immediately go from 100 users to a million!” is probably not advice to bet your business on.

The Steps

A consumer startup with good product can become proficient at growth via a fairly straightforward series of steps. As a founder, I informally understood that series; working with the companies of 500 Startups allowed me to formalize it as a six step process:

1. Track: Figure out what needs to be tracked. Track it.

2. Understand: Delve into the data to understand how people are using the product.

3. Prioritize: Evaluate and prioritize the areas most likely to yield growth. Sometimes they’ll be tweaks, sometimes they’ll be re-architected features, sometimes they’ll be completely new features.

4. Design/Write: In the top area or two, design a few features that are likely to yield growth. I emphasize writing because the words describing a product often matter at least as much as any other characteristics.

5. Build: Code it up, push it out.

6. Measure: Gauge success of new features. GOTO 1, 2, or 3, adjusting strategy based on the results.

The Process

Much like a signup flow that sees significant dropoff upon asking for the first piece of information, companies often get sidetracked before finishing the first step. Tracking and storing data — so you can actually go back later and see what went on — requires discipline and is low on glamour and creativity. You know that tracking won’t help you tomorrow: isn’t it more important to build out that cool feature that might actually help right away, or fix a bug that a user is complaining about?

Usually, no. Those who try to grow by relying only on designing and building often wind up spinning their wheels: they work on the wrong area of their product, again and again; they never completely understand what worked and what didn’t.

In my stint at 500, there were a few times when a company I worked with didn’t get past step one. Needless to say, that was frustrating for me, and it forced me to come up with a rule: if you want me to spend my time helping you grow, you need to set up the basics of tracking before we start working together. That rule worked well.

The technical details of what needs to be tracked vary; a more mature company will usually need to store most user actions in their own database. Pretty much anything the company cares about improving — the number of users coming in via certain channels, views of certain pages, clicks via email, invitation responses — needs to be stored.

Once the data are in place, understanding the big picture comes to the fore. How are people finding out about the product? How and how frequently are old users coming back in to the product? What, if any, are the social dynamics: are people coming back because someone commented on something or because someone followed them?

When that’s in place, it becomes relatively easy to come up with some estimates of the metrics impact of potential product improvements. To be sure, those estimates are more likely to be accurate when done by someone who’s done them before, but anyone can and should make them.

Those estimates facilitate an ordered list of possible product changes. These changes might be minor — changing “Next” to “Continue” or “Submit” — or major, like building out a completely new signup process. And again, designing or writing by someone who’s skilled and knows what’s likely to be effective, is valuable.

Building out those designs is the obvious next step; I generally didn’t get directly involved with that at 500.

Measuring the effects of those changes — for companies with data scale, via A/B testing — is the vital last step. It’s amazing to me how many companies will spend weeks working on a new feature or design, and then never really know if it was effective. If you’re tiny and just trying to get some kind of win, that’s okay. If you have an audience of hundreds of thousands of people or more, it’s negligent.

Succeeding As an Adviser

I effectively served as an adviser to the 500 Startups companies I worked with (though not financially), and I’ve come to realize that it’s not easy to add value as an adviser. Many advisers get a bunch of equity in a company, but don’t add a lot of value. If I agree to advise a company, I know there’s some chance that my advice won’t end up being relevant, but I want to maximize the odds that the time I spend with founders and employees has a measurable impact.

As a founder, I intuited those six steps, and then spent hundreds or thousands of hours building everything out with a team of people I saw every day. As an adviser, I’m spending hours or tens of hours, so my thought process has had to be tighter and more formal.

The six steps outlined above can work pretty well regardless of the scale of involvement. There are a few companies with whom I spent a few hours a week over several months. With those companies, we spent a bunch of time upfront to make sure everything was being tracked. Then every week or two, we’d run through steps 2-6: finding a problem area, brainstorming and designing possible improvements, building, and measuring the results.

With most 500 companies, my interaction was lighter: a 30-minute conversation every month or two. That may seem like too little time to go through six steps, but an abbreviated version can still work well. The conversation would usually look like this:

  • 1 and 2. Tell me about your business. How do people find out about your product, how do they get back in to the product, etc. This works better if the founders get answers in advance.
  • 3. Let’s figure out (in part by looking at the existing product) which areas to prioritize.
  • 4. Let’s look at the product and try to narrow down (almost) exactly how the changes will look.

Then the startup does the details of step 4, and all of step 5 on their own — actually building it out — and does the measurement step (6) with a little bit of guidance.

That process isn’t always going to be effective, in large part because most product changes don’t work as hoped. But by understanding the problem and betting on the changes with the highest expected returns, I’ve seen many wins. Here are a few:

  • One company I worked with, TradeBriefs, gets most of their return traffic from email newsletters. That became obvious when we walked through the basic metrics of the business to understand how everything fits together. The content of those newsletters had always been hand-curated, which means it consistently looked professional but wasn’t optimized.

    In previous lives, I’d seen how similar types of email content can have vastly different clickthrough rates. To better discover which content will yield maximum click rates, the team created a system that allows them to test different versions of similar emails each day. Their volume was large enough to allow them to randomize content for a small batch of users early in the day, wait, pick a winner, then send the winning (best) content to the majority of their users later in the day. They implemented that and saw a 30% improvement in their email click rates.

  • Another company I worked with is a community where users are creating all of the content and interacting with one another. They were seeing lower usage among people who had been on the site for more than a few months, and they wanted to address that.

    We spent a lot of time upfront on metrics, tracking activity levels and looking at how users transition across differing levels of activity. That digging led us to realize that people who became moderately active by day five were far more likely to stay active than those who had not. So we focused our efforts there, testing lots of little tweaks in the new user experience to get people to activate in their first few days.
    Cumulatively, those tests led to a 30% increase in initial activation; we expect that to pay similar dividends long-term.

  • A third company is largely a commerce company that makes money when people buy physical goods. Most of their traffic — and purchases — were coming from advertising on Facebook and Google. The founder discovered that she could get more traffic from those same channels much more cheaply if she sent users to a more fun — but less commerce-focused — area of the site. However, only 4% of that cheaper traffic converted into leads likely to monetize, making it not worthwhile.

    Digging into the flow, we found a number of places to optimize the user interface — largely by making it intuitive to click through to the next step — and increased that 4% conversion rate to 16%. That difference was a huge one, immediately making the new stream of traffic immensely profitable.

Those three examples are in the minority: for most of the companies I spent time with, the immediate value I created for them was little or nil.

Yet I come away feeling successful. In six months, I’ve been part of a few real wins and found a template for how I — and hopefully others at 500 Startups going forward — can help startups.

Thanks to Dave, Christine, the staff at 500, and the many, many founders and employees I’ve worked with over the past six months. I look forward to continuing to work with you all as a 500 Startups mentor and friend.

And now it’s time for me to go and build something…

Goodbye to a Patriarch

This morning, my grandfather passed away.

Robert K. Greenfield was ninety-seven. He lived a long and interesting life, and he had a real impact on the world in which he lived.

When I was in sixth grade, I was given an assignment to write about a modern day Renaissance Man. I wrote about Bob and his knowledge and interest in countless areas — law, tennis, wines, public policy. At the time, he was seventy-four and seemingly winding down his life. Little did I realize that he still had more to accomplish than most of us do in our lifetimes.

In 2008, Bob had a health scare, and it looked like he might not have much time left. I wrote him the letter below, to let him know what he’d meant to me and my family:

I’d like to share a few thoughts with you, and hopefully give a little something back. You have played a considerable role in shaping who I am, and you will continue to do so, no matter what happens with your current condition.

Most people give up at some point or another. They try unsuccessfully to change things, and become cynical. They get worn down by the grind of life around them, and become indifferent. They succumb to greed, and become corrupted. They lose the discipline to take care of their bodies, and become slothful. They lack the intellectual rigor to seek truth, and become mentally lazy.

You aren’t like most people.

I am in an industry and a world that’s very different from the one you inhabited when you were my age. I sit in front of a computer all day; I solve algorithmic and mathematical problems, trying to construct a product that helps to bring people together; I try to stay as far from the legal world as possible! I am very much a product of my generation and my background: an Internet geek who lives in San Francisco, has traveled around the world, has no kids at age 30, and spends at least ten times as much money on food as on gasoline.

And yet, I cannot help but look to your example as a guiding force in my own life.

Elaine and I see your 71 years of loving marriage with Mommom as amazing, and something we aspire to (if we’re lucky enough to live that long!). We were so glad you made it to our wedding, and honored by your kind words there.

(An aside: a Chinese colleague of mine mentioned that she and her husband were very impressed by what you said; she said that your brief speech made it clear that you were the patriarch and leader of the family. Apparently these things cross cultural lines.)

I am continually impressed by the strong foundation you’ve built with the Goldsmith Prize. The more I learn, the more I realize how difficult it is to set up something like that, and the Goldsmith Awards have been an unbelievable success. Your combination of creativity, foresight, diligence, and personal relationships have uniquely crafted something special.

I will work hard to build on the Goldsmith successes, but it goes beyond that. I also use the Goldsmith model every day in my life: today I try to create a project of similar magnitude in the Internet world; tomorrow I may look for the next big thing in the world of philanthropy. Thanks to you, I’ve gained great insight into the process of creating something substantial, and a big boost in the inspiration needed to actually do it.

There aren’t a whole lot of people who can make me feel guilty for not working hard enough, but my 93-year-old grandfather is one of them. Your discipline, hard work, professionalism, attention to detail — I could go on and on — are astounding. I didn’t realize that when I was a ten year-old kid, but at a certain point I realized that your successes — academic, professional, family, athletic — were not the result of happenstance.

Many successful people get ahead by cutting corners, by sacrificing their principles or the common good. But that’s also not you: your honesty, your sense of fairness, your (stubborn!) desire to stick to the rules keep you grounded, and are yet another reason you serve as an inspiration to me and the rest of our large family. The ongoing trend in your life has been to look beyond what’s easy, and instead do what’s right.

My words are a mediocre expression of my sentiments and appreciation for all you have done for me and my family, but I hope they give some small sense of the effect you’ve had on my life. That effect will last as long as I will — thank you.

Fortunately for all of us, he lived another four and a half years. This past June, with dozens of family members, we celebrated 75 years of marriage between him and my grandmother. He got to meet my daughter a few months after she was born last fall, and then again before he died this month.

Amazingly, he was an avid reader of this blog. In his typical self-deprecating fashion, he explained how difficult it was for a “twentieth century mind” like his to understand the new things I was writing about. I know that wasn’t true.

The last time I saw him, less than two weeks ago, he was physically tired but mentally sharp. He was trying to figure out how policies around the fiscal cliff might affect his family, he was doing what he could to help with intra-family disagreements, and he recounted countless stories from nearly a century on earth.

Aside from the combination of physical frailty with mental sharpness, two memories from my recent trip stand out.

One memory is my grandfather’s awe at the possibilities that exist in the modern world. I shared some of my own startup experiences, and he explained how he wished he’d been able to do something similar. Bob certainly had an entrepreneurial mindset — after he retired he set up the Goldsmith Prize — but his day job as a lawyer was much more cut and dry. I’m incredibly fortunate: I live in a time and place where I can dream big dreams and run with them. Every day, I try to bring to my projects some of my grandfather’s intelligence and tenacity.

The second memory is a more personal one: my grandparents spent much of their time on the couch next to one another, savoring their final days together. My grandmother asked my grandfather if he still remembered their first kiss; he said he did, but amazingly their stories diverged. I feel sad for my grandmother: she’s only 95, with years of sneaky tennis drop shots ahead of her. But it was amazing to witness that kind of love after more than 75 years, and I’m glad that their four kids, fourteen grandchildren, and seventeen great grandchildren saw it and benefited from it too.

Thank you, Bob.

After the Thrill Is Gone

A growing startup can be bliss.

When Circle of Friends was growing rapidly, I’d wake up suddenly at three AM, my heart jumping with a mix of excitement and nervousness. Because our technology was brittle, I’d walk out to the kitchen and look at my laptop to make sure the site hadn’t crashed. And I’d notice that, amazingly, our traffic for the period between midnight and three AM was our highest ever.

Our highest ever? Our highest ever!?!? How could I go back to sleep now? It was three AM and my day had started: I boiled some water to make a pot of tea, and I got to work.

That was the startup honeymoon phase, and it was exhilarating.


Circle of Friends (which became Circle of Moms) had its first office in our little house in downtown Palo Alto. We’d bought the house in 2006, knowing it needed work. Elaine, being an architect, saw this as a positive: it was a house we could truly make our own.

After a few months where I was both living in the Palo Alto house (with Elaine) and working in its kitchen (with Ephraim during the day — but alone at 3 AM), the center of my world moved to San Francisco. Elaine and I decided we’d rent out our house and find an apartment in the city, and Ephraim and I would get an office in SoMa.

That basic arrangement — living near Dolores Park in SF, working South of Market) — was my life for nearly four years.

When, in early 2011, we found out Elaine was pregnant, renovations on our house in Palo Alto were nearly finished. At that point, we hadn’t decided what would happen after the house was ready to live in: we were open to selling it, renting it, or living in it ourselves.

The fact that we’d soon have a little daughter made living in it — and a move to Palo Alto — more appealing. Our own new place, with a nice room for baby, lots of natural light, a beautiful kitchen (nice work, Elaine!), and an edible garden seemed like a great option.

But I also knew myself and my work patterns. It was hard to imagine having a baby, a startup, and an hour-plus commute. Dealing with two of those three would be challenging; dealing with all of them would just be stupid.

The more I thought about it, the more I realized what a move to Palo Alto would mean: the beginning of the end of my time at Circle of Moms.

Elaine and I soon decided we’d move back to the house when our daughter was born, and that meant my time at Circle of Moms would be limited. Certainly not “I’m giving two weeks notice” limited, but also not another five years.

I told Ephraim of my intentions, assuring him that my goal was to get Circle of Moms to a place where we’d both feel comfortable telling the team that my role would change. We agreed that it would not yet make sense to share those plans with the rest of the Circle of Moms team.

That marked the beginning of a very different phase from the one where I’d wake up at three AM and excitedly sit down to work.


On a Monday morning in July 2011, Elaine was now six months pregnant, and we were at the doctor’s office. She had gone to the bathroom, and my phone buzzed with an IM from a colleague.

hey mike – will you be in the office today?

“Yes”, I replied, “am at a doctors appt now, will be in around 10″. A moment later, I got my colleague’s response:

oh okay – can i grab you for a chat sometime before 11?

Before 11? Crap. She’s going to tell me she’s leaving the company.

I was right, it turned out. And though my co-founder and I did what we could that week to try to convince her to stick around, we knew our odds were slim. We were ultimately unsuccessful, and she left Circle of Moms a few weeks later.

The Responsible Founder

At the time, there was a part of me that wanted to check out and move to the next thing, but as a responsible founder, that wasn’t a viable option.

In each of my three jobs before becoming a founder, I’d decided at a certain point that I’d had enough. Within weeks of coming to that conclusion, I’d told my boss I was leaving; by a few weeks after that, I’d formally moved on.

On that July Monday in 2011, I’d just heard my employee’s plans to leave and join another company, and I knew she’d soon be out the door. Meanwhile, my writing was on the wall and I was ready for something new — but acting on it now was simply not an option. I had to pull myself together, convince the team that everything would be fine, and get back to work.

Fortunately, the process went about as smoothly as it could. The team took things well, another employee immediately stepped into the shoes of the woman who departed, and we continued to make progress as a company.

But that summer was still a challenge for me personally. Weekends, which Elaine and I spent in Palo Alto, were awesome. We were in our beautiful new home, getting everything ready for ourselves and for the baby, cooking in our spacious kitchen, and enjoying the warm sun. During the week, we’d stay in cold, foggy San Francisco, and slowly pack up everything from our small dark kitchen.

Several other people left the company over the next few months. Our traffic was growing and we were executing well on the product, but we certainly didn’t have rocketship growth and our talented employees had plenty of other options. Again, I was tempted to follow my employees and also move on to the next thing; again it was not an option.

With all of that going on, I’d become more and more open to being acquired. That was a big change from a year or two prior, when the acquisition offers that came our way held little appeal for me: earlier, I would have been disappointed if another entity had ultimate control over the product and the company. But now, I’d been through the startup building process, realized I was unlikely to be the visionary of this business, and was open to a “strategic partnership.”

I spoke with another founder who had recently been through the acquisition of a similar-sized company. At a certain point, he told me, he’d realized that his company wasn’t likely to be a huge success, and that an acquisition was the way to go. He spent an entire year getting the company to that stage, in the process building relationships with his eventual acquirer, and said it was easily the toughest year of his life.


Our experience was much the same: it was just under a year from the time I knew I was ready to move on until February 2012, when Sugar acquired Circle of Moms. It wasn’t the most fun year, but it was a productive one in many ways.

And I was fortunate. With a combination of luck and skill, we’d already built a real business, a strong team, a solid product. We wound up making money for our investors — even if it wasn’t a billion dollar sale — but the process still took us a year. Many founders spend a year or more on the less fun parts of running a company, without seeing a positive exit.

Christine Tsai, with whom I now work at 500 Startups, wrote a great essay comparing parenthood to building a startup. I agree with everything she says: in both domains, you work hard, it’s hard to figure out who to listen to, and it’s terrible… and fantastic.

There’s one other parallel I’ll add. As both a founder and a parent, you’re responsible, and you can’t just pick up and leave when you feel like it.

A mother’s work is never done. A founder’s work most likely will eventually be done, but it certainly won’t be today at 5 PM, and he can’t know in advance exactly when it will be.

And that means a founder can’t just give his two week notice and disappear. Even after the thrill is gone.