The Opportunity Cost of Features

Someone Fought Hard For That Pixel!

Know someone who might like Capital Gains? Use the referral program to gain access to my database of book reviews (1), an invite to the Capital Gains Discord (2), stickers (10), and a mug (25). Scroll to the bottom of the email version of this edition or subscribe to get your referral link!

Sometimes, I like to go to a chain restaurant and spend some time appreciating just how much work and stress went into deciding what makes it onto the menu. Every menu item has an effect on the customer’s revenue and that revenue’s margin: add the right dish and they'll have a higher propensity to order an additional appetizer, it'll lead them to request 0.034 more Tiki Beach Party ’Ritas per meal1 , the colors will pop on the Instagram photo, a large party where one member has a dietary restriction won't choose a different restaurant, etc. The more your decision is driven by unconscious influences, like the order in which options are presented, which ones have a nice photo, or whether or not something is labeled "NEW," the more optimizations like this will pay off.2

At a restaurant, unconscious influences are just one element of their economics, it’s only when these restaurants are operating at scale (across the country) that make it worth watching. But for other companies, the top line is almost entirely determined by a succession of these choices.

Take consumer Internet companies as an example: they have very low marginal costs for handling a single user-session. The variability in their economics comes from fixed costs, and from the interface tweaks they make to drive different in-app behaviors. Consider Meta, and its two big revenue drivers, Instagram and the Big Blue App. Imagining that the typical user is accessing these apps on a full HD screen, with 2080 x 1080 pixels. Meta's trailing four quarters' ad revenue is $125.4bn. So their approximate ad revenue per pixel is about $28,000 per year.

When companies are young, they experiment, but they don't necessarily track metrics too closely. And in many ways they can't track those metrics closely. If you want to test customer elasticity with a 10% higher price, and you're only getting half a dozen transactions a day, it's going to be a long, long time before you have something significant. (Specifically, if you're measuring the gap between a 0.9% conversion rate and a 1.0% conversion rate, you need to show about 300,000 people your landing page in order to get 95% confidence.)

At a bigger company, though, especially at a company whose model is to get a tiny amount of revenue per minute from billions of users, a sample size of 300k is no big deal. So for simple tests that are examining the one-off effect of some variant in copy—like seeing whether "add new friend" or "add connection" leads to more friending—they can quickly get to optimal. But as they get closer to optimal for the established parts of the business, the opportunity cost of big changes only goes up.

A fun example of this comes from Chaos Monkeys, Antonio Garcia Martinez's wonderful memoir of working in tech in the early 2010s. At Facebook, he saw a source of easy revenue: put a big ad on the logout page! This was clearly found money: people who logged out were done with their session, so it wouldn't hit engagement. And there were a lot of people logging out of Facebook on any given day.

But it turned out to be an incredibly dangerous idea, with militant opposition from Facebook's growth team. As it turns out, one of the biggest sources of new users for Facebook was from people who went to cyber cafés in the developing world, sat down at a computer, saw the Facebook logout screen, and decided to check it out. It's a powerful endorsement when most of the monitors are showing exactly the same thing. Facebook didn't want to run somebody else's ads on their own free ad inventory, especially in places where the third-party ads wouldn't be at an especially high price point. The compromise was to run logout ads only in mature markets where there weren't major competitors.

The high cost of risking a pixel is a good explanation for a few phenomena:

  1. Why do big companies often launch a separate product and promote it through their main app rather than making it a first-class feature of the app? Because a button on the bottom navigation bar for Facebook, or the Google homepage, is some of the world's most valuable ad inventory.

  2. Why do they cut features? Because those features have a cost, not in compute or developer time, but in distracting people from more proven money- or engagement-makers.

  3. Why do big companies ship so little? Because as their margins expand, the downside from disrupting the main app's experience rises faster than the upside from adding a new one.

This makes big app changes, like Facebook promoting Marketplace (for a while) and adding Reels, or Google incorporating Bard, all the more impressive. The direct dollar cost of developing these is a fraction of the opportunity cost of drawing attention away from relentlessly A/B-tested organic content and high bid-density ads. The curse of success for a big consumer Internet company is exactly the same as the curse of success for a mid-career professional who is minting money at a job they don't particularly enjoy: after a while, the cost of doing anything else is intolerable.

Read More in The Diff

The Diff has alluded to this model many times:

1. Sodium: the most revenue-accretive element on the periodic table.

2. This works the other way, too. If you're making a big and quantifiable decision like refinancing a mortgage, the marketing materials get significantly less splashy as you get closer to the part where you calculate what the dollar return is. Considering a big financial decision might be impulsive, but making one is more deliberative. On a related point, it's impossible to track directly but I'd be willing to bet that decisions based on memos get much better outcomes than decisions based on powerpoints.

Share Capital Gains

Subscribed readers can participate in our referral program! If you're not already subscribed, click the button below and we'll email you your link; if you are already subscribed, you can find your referral link in the email version of this edition.

Join the discussion!

Join the conversation

or to participate.