Is this a career?

The trouble with “working in data.”

The answer is blowin’ in the wind.

A while back, Dan Luu wrote a blog post that periodically makes the rounds in Silicon Valley. It begins:

Reaching 95%-ile [ in most skills and activities ] isn't very impressive because it's not that hard to do.

And continues:

Personally, in every activity I've participated in where it's possible to get a rough percentile ranking, people who are 95%-ile constantly make mistakes that seem like they should be easy to observe and correct. "Real world" activities typically can't be reduced to a percentile rating, but achieving what appears to be a similar level of proficiency seems similarly easy.

His point is not that nothing is hard. It’s that, to get reasonably good at most things, all you need to do is put in a moderate amount of intentional practice, and very few people do that. People might commit time to the thing—learning a game, giving talks, writing, whatever—but they’re typically passive players, fumbling through without stopping to ask what mistakes they’re making and how they can fix them. Often, according to Dan, the distance between a novice and an apparent expert isn’t a gulf of skill and 10,000 hours, but a few hours a week of effort and an active intention to improve.

I have a theory that there's also a startling amount of distance between people who are merely very good at something, and people who are truly great—except, this gap is vast. Staggeringly, arrestingly, hopelessly, inhumanly vast.

Nate, a former coworker of mine, is a horrifyingly good runner.¹ He ran the 2023 New York City marathon in a disgusting two hours and 37 minutes, running a six-minute-mile 26 miles in row. The winner, Tamirat Tola, ran it in under two hours and five minutes, pacing at 4:46 per mile, beating Nate by 74 seconds per mile.

Nate is a far, far better runner than just anybody any of us know. And Tola demolished him. If Nate, Tola, and the world’s fastest mile runner, Hicham El Guerrouj, had raced for the first mile of the marathon, Tola would’ve finished closer to El Guerrouj—who once ran a mile in 3:43—than Nate would’ve been to Tola. And Tola maintained that gap for 26 straight miles. It’s incomprehensible, what it takes to run that fast for that long.

Olympic decathletes run the 1500 meters in about four minutes and 15 seconds. Olympic 1500-meter runners run 1500 meters in three minutes and 27 seconds;² they would nearly lap the decathletes—who are not 95th percentile runners, but something wildly higher than that—on a 400 meter track. James Sprague, the current world Crossfit champion, a 210-pound barrel of muscle, can clean and jerk 340 pounds. That ties the world record for that lift—for women who weigh 155 pounds. In Sprague’s weight class, the men’s world record clean and jerk is 511 pounds.

World chess champion Magnus Carlsen beat ten good amateurs at the same time, while blindfolded; he took over a half-lost game against an opponent as skilled as an international master and came back and won, while drunk. Troy Smith was nearly unanimously voted the best college football player in the country in 2006, and only lasted 20 games in the NFL. Jimmer Fredette was voted best men’s college basketball player in the country in 2011, and was an NBA starter for seven games. Deeply experienced politicians who have the skill to win major elections routinely get humiliated when they run for more ambitious offices. Srinivasa Ramanujan solved problems that had completely defeated Cambridge mathematicians. Experienced magicians can palm cards and always keep your card at the top of the deck; the best ones can shuffle them exactly as they want to.

Obviously, this is imprecise, and the distribution of expertise can’t be quantified by any single number or plotted on a single graph. But if it could, it seems like it would be shaped like backwards L: The slope of the curve would be nearly flat, and then nearly vertical. For almost its entire span, it is a shallow hill that most of us can climb, if we put in a bit of effort. And then, it is a cliff, straight up. And lots of people are hopelessly stalled at the bottom, perhaps better than 95 percent of the world, but a mile below the world’s best. Because, even when great climbers think a cliff looks impossible to scale, there will be somebody who can go straight up it.

A career in science

So here’s my question, then:³ If you work in data, what cliff are you climbing?

Some people may want to be experts in tools. They want to know every data observability tool on the market; they want to get a Google alert when someone mentions “reinventing business intelligence” on Product Hunt. Their mountaintop is being the illuminati behind the Magic Quadrant, or to be Marques Brownlee in a blazer, a company t-shirt, and corporate dress sneakers.

Some people might want to be experts in technologies. They read ArXiv abstracts about vector databases, and dream of rebuilding Postgres in Rust, for Iceberg, in a box. For them, data is incidental; the interesting thing is the computer that processes it.

Some people are here for the techniques. They work in data to talk about Nick Felton’s Annual Report and how to represent uncertainty in a chart, because they love data visualizations; they work here to debate causal inference, because they are statisticians; they care about data modeling methods, because they are Ralph Kimball.

But most of us, I suspect, are some undecided fourth thing. We are practitioners of these other three things, but unpracticed in them; we aren’t, as Dan Luu instructs us to, actively trying to become great technologists or statisticians. Instead, we drift among them, and define ourselves as analytical generalists, or, even more atmospherically, as “data people.” We would probably say that we are good at critical reasoning, are rigorous and quantitative, are always curious and question the status quo. We are passable programmers and statisticians, and experts in “analytical thinking.”

No doubt, that is a potentially useful intersection of talents.⁴ You can build a career out of this sort of resume, and lots of people have. But I’m starting to think—which, shocker—that it’s a career with a low ceiling.

Pursuing a career in analytics feels like pursuing a career in “science”—it’s not specific enough to go anywhere. Yes, there is a loose set of attributes that all scientists have; they are empirical, skeptical, observant, and structured.⁵ And yes, there are noble jobs that are generically scientific, like teachers and educators, and there are approximate parallels in analytics, like team managers and mentors. There are also mercenary jobs in both—“influencers,” mostly, who seem more interested in talking about the thing than doing the thing. But there are not experts in science. Science is what you take in fifth grade, not what you win a Nobel Prize for.

The analytics discipline may well be the same: A general set of skills, and a survey course that you’re supposed to start in but not the career you’re supposed to finish in.

Everything nowhere all at once

Whenever we raised money for Mode, I used to have this bit about how the work that analysts do is inherently viral. You don’t create charts or do analysis for its own sake, I’d say; you do it because you want to share it with someone else.⁶

That is the optimistic way to look at analytics, I suppose—as a nerve center, infiltrating its way into everything. But the pitch has another implication, which is that analysts can’t stand on their own. To paraphrase an analogy a reader sent me, analytics is a dust devil. It is invisible on its own; we only know it’s there when it blows something around. You cannot see the wind without something for it to disturb; you cannot be an analyst without something to analyze. Otherwise, it is just a verb in search of a subject.

Another reader put it this way:

…the domain you're talking about is the analytics industry. I don’t know what that is… I have a clue, though—the framing necessarily means that the domain is secondary to the tool. Given that, I wonder if it can be defined as "analysis of data in fields where it's not in the critical path to revenue/purpose". Because when it is, it's usually baked into our understanding of the domain—econ, bio, transportation planning, policy research, etc.

It’s a subtly searing criticism of what so many of us do. It suggests that there are lots of analysts out there who don’t have to constantly elbow people to get their attention, but they call themselves something else—like quant traders, pollsters, CFOs, and general managers. And they got there by specializing: They committed themselves to finance, or political opinion polling, or baseball, and found their principle identity in that instead of under the vague umbrella of “data.”

Without that, we seem doomed to have almost definitionally peripheral careers. We are fact checkers, not authors; we are logical mechanics, not drivers. We are junior executives, bouncing from one director or VP role to the next, always given the same task of “building a team and a culture.” A few years ago, I thought that the way to get off this treadmill was to craft a better role for the executive analyst. But now I’m not so sure. If we want to do something more important, we have to do something more important.

Which means, I suspect, choosing to a cliff to climb. We can’t just be analysts or analytics engineers; we have to decide that we want to be true experts in understanding how to build consumer software first, and product analysts second. Or define ourselves as working in finance, then become an analytics engineer at a fintech company. Because there’s a corollary to Dan Luu’s theory of expertise: While it implies that we can become pretty good at stuff pretty quickly, it also implies that other people can become pretty good analysts. And in almost every field, that combination—a domain expert, and 95th percentile analyst—is almost always better than the inverse.

That’s ok, if that’s we’re happy on the edges. But if not—if people are antsy, reaching middle age in an industry entering its adolescence—we probably can’t get away with being good at “asking questions.” We need to know some specific things too. Otherwise, we risk being like the wind—everywhere, but nowhere in particular.

---

Page 2

Since I’ve started this blog, I’ve asked people to answer six questions:

1. Choose a robot?

2. Choose a number?

3. Choose a company?

4. Do you know what it’s like to discover something?

5. Do you know what it’s like to make money?

6. Will Microsoft Excel outlast money?

Lots of you filled out these surveys, but I only told you what one of them said (the fifth one, about making money). So, at some point in the next couple weeks, this blog is going to turn into a 2004 Bill Simmons column, and there will be a mailbag episode about the other five. If you want to manipulate an election, go nuts; submit some more votes; the L.L.Bean slippers are still up for grabs.

And, in the spirit of a true mailbag post, if you have any other questions you want to ask, the mailbox is open.⁷

1. And also horrifyingly good at data stuff (if, of course, that is a thing).

2. If you haven’t watched this race, stop reading whatever this ridiculous blog is, and watch it. The early lead, the break from Kerr, the little shove, the lunge from Nuguse—it’s like one of those contrived CGI-animated races they put on the jumbotron between innings at baseball games that are scripted to make it look like everyone has a chance to win, except it’s real, for a gold medal, in the Olympics.

3. Asking for a friend, obviously.

4. Obligatory Venn diagram reference.

5. Is this true? I don’t know, the only science class I took in college was about aliens and Atlantis. (It was a freshman year seminar on pseudoscience. Or, you know, my first analytics class. Ba-zing.)

6. And therefore, the pitch went, we could sell Mode to analytics teams, have them send reports and dashboards to everyone else, and eventually sell it to the whole company.

7. A normal URL, if you prefer.