Efficiency is Dumb
(or at least efficiency above all is dumb, the actual post is more nuanced than the title, but I'm angry)
I originally started drafting this back in around 2023 as the LinkedIn topic du jour was how tech companies (especially the, scare quotes for emphasis, “bloated” big tech companies like Google, Facebook, and Amazon) needed to react to the inflation spike and resulting interest rate spike with greater “efficiency.”
Now it’s obviously taken on an even more sinister meaning as it’s the buzzword for morons to try and illegally break the United States Government. But unfortunately the pedantic things I wanted to scream into the void 2 years ago are still relevant, even in our more horrific context (though this is still more business focused than sanctity-of-our-democracy focused, because, well, write what you know.)
That said, the one benefit of the new tenor “efficiency” is taking is that it frees me up to be clear-eyed in my thesis: if you’re a business leader who cares about “efficiency” above all else, then you’re probably bad at business.
You can be bad at business and still have a successful business! You may luck into success despite your best efforts, whether because your fundamental idea is so great it can support failures elsewhere (the argument of original-regime Twitter, some might argue is true of Google today) or because you have an effective monopoly (definitely true of Google today) or because you can convince people to behave irrationally and keep giving you money (WeWork pre-IPO, Tesla today) or myriad other ways to fail upward.
But having your primary mission statement be “efficiency” seems like a big way to waste a bunch of people’s time.
Alright, to step back from the ledge of the soapbox for a minute, I don’t mean efficiency is totally bad or worthless in and of itself. In the same way that, for most people, on the margin, you should aim to be a little kinder and exercise a little more and eat a little healthier and donate a little more money to people in need, most businesses could probably stand to be more efficient on the margins. And some businesses are really in dire, existential situations where efficiency becomes a top priority.
But I’m talking about the morons on LinkedIn and in the Oval Office who hold up efficiency as the ultimate good, devoid of any downside. My assumption is most don’t actually believe it and are just being rhetorically lazy, and the ones that actually do believe it are so dumb as to be beyond saving. But the point of this substack is to get anyone who is tempted to go down that path of laziness to remember that you have brains and critical thinking skills, and should use them whenever anyone trots out meaningless platitudes like “our number one goal is efficiency.”
First annoying thing about worshipping efficiency: what do you even mean by that???????
Efficiency is, definitionally, a ratio metric, meaning you define efficiency in terms of how much of a given input it takes to produce a given output. The smaller that input/output ratio is, the more efficient the system.
So just saying “we want to be more efficient” is literally meaningless without more context. Do you want to release more features per engineer?1 Do you define a feature as a Jira ticket, a pull request, or a customer-facing release that gets a release note, or what? Do you want to land more sales per salesperson? Do you mean that in terms of number of sales, or in terms of the total dollar amount of sales? Do you want more dollars of revenue per dollar of expenses? Do you care if that revenue is recurring or one-off?
Depending on how you define your inputs and outputs you’ll have different measurements of efficiency, and that will imply wildly different strategies to “improve” it.
Contrived example, but think about fuel efficiency in cars. That’s a ratio metric defined as how many miles a car can drive per gallon of gasoline. An electric car or a bicycle or walking is the ultimate fuel efficiency machine, per that definition, because the amount of gasoline they need in order to move is 0; the fuel efficiency is literally incomputable!
But one benefit of hydrocarbons like gasoline is they are incredibly space and weight efficient. The amount of pure energy you get out of combusting a tank of gasoline is almost 10 times as much as the amount of energy stored in an EV battery, despite the battery being way way way bigger and heavier. So if you want to know which car is more efficient in terms of miles driven per kilogram of fuel storage, a gas car is significantly more efficient than an electric car.
So again, to the LinkedIn example, “be more efficient” is meaningless unless you know what you consider to be your inputs and outputs. An engineer who cars about miles/kilogram of car would design a very different car than an engineer who cars about miles/gallon of gasoline.
And aren’t all these people who post about efficiency also people who claim to be “data driven” and say stuff like “what you measure you can manage”?2 Well how can you measure efficiency and know if you’re doing a good or bad job if you aren’t specific about what you’re measuring!
Alright, if you probably push them, they’ll come up with some sort of definition, probably something like “revenue/costs” or “revenue/employee” or something like that. So then just as important, efficiency is a measure of a system, so it’s very important you define what the boundaries of the system are.
For the car example again, a common disagreement is whether electric vehicles or hybrids are really more environmentally friendly than gas cars. If you look at it in terms of miles/gallon of gasoline burned, then it’s an easy win for electric/hybrid. And if you look instead at tailpipe emissions generated by the car while driving, it’s still an easy win for electric/hybrid (obviously.)
The critique then goes that just looking at tailpipe emissions is an incomplete picture: the electricity that powers the battery of the EV still has to come from somewhere, and depending on where you are in the world that somewhere may be primarily burning coal or natural gas. And sometimes people can do math that shows that when you net out all the emissions created through the charging lifecycle, that can be as bad or worse than the tailpipe emissions of burning the gasoline in the tank of the gas car. Same is true if you take into account the rare earth metals used in constructing batteries, or the overall emissions cost of building a new electric car vs buying a used gas car.
My understanding is that’s mostly overblown and/or cherrypicked, but still: depending on how you define the total system, you can get very different results in your understanding of a system’s efficiency.
In the business context I think this is mostly relevant in terms of whether you measure efficiency at the employee, team, or organizational level. As a toy example, say you want to focus exclusively on efficiency in terms of number of sales deals closed per salesperson, and that’s the number one goal for the organization. Obviously you’ll need to have performance management for the sales team to drive towards your goal, but what does that mean for marketing? How much of a sale is attributed to them? What about engineering? They don’t directly sell, but they build the things that then get sold? What about legal or HR?
I used to work in product management for a major international deals platform3, and on the consumer-facing team (that owned our websites and mobile apps) our north star goal was to improve conversion, which is a type of ratio metric like efficiency. And the team and company did a fantastic job at bullet #1 in that it was precisely defined: conversion for us meant number of units sold per unique visitor to the platform.4 And so we got pretty good at figuring out what changes you could make to a marketplace to improve Units/UV.
The unfortunate problem though was that no matter what tweaks we made or levers we pulled, time and time again analysis showed that the two things that had the biggest impact on conversion were things outside of our team: supply (did we have enough high quality merchants on the platform?) and pricing. Making up numbers but if a tweak of how we display a deal would have say 2% impact on conversion, aggressive pricing changes could have like 10%+. But we weren’t the pricing team or the supply team! So constant energy had to be spent by leadership and analytics to try and figure out how to appropriately measure the impact of our work on conversion net of supply and pricing changes, because otherwise we would have no way to know whether we were doing our jobs well.
Similarly, you have to consider time horizons in your system definitions. In our case, there was always the concern that we may be creating short-term conversion benefits, but at the cost of long-term consumer lifetime value. Extreme examples (which we generally avoided!) would be things like advertising bait-and-switch or disingenuous pricing (which may get the initial click or initial sale, but leaves a customer with a gross feeling and they won’t come back later) or having stricter or more cumbersome return and refund policies (which would save money in the short term by preventing having to pay for a refund, but would turn off the customer long-term.)
Related, you can have things like safety policies or regulation that appear to impose an inefficient up-front cost, but pay for themselves in the long term. Right now there are crypto industry lobbyists who oppose some of the current regulatory rollbacks on the crypto industry, in part because they (imo, correctly) believe that if you want normal people to buy and use crypto products5, you need to convince them over the long term that it’s not just a free-for-all of scams, and safety regulations can help do that convincing6.
I see this especially being relevant for companies (or governments!!) that are large enough to have research functions, because if in the name of efficiency you only ever research the things you think will pay back immediately, then you will miss out on a massive amount of potential discoveries. There’s the old saw about how penicillin was developed by accident and whatnot (or the more modern version of how the thing that’s enabled the whole latest major AI breakthrough was a research project from a search engine.) There’s certainly risk to letting research go on without a concrete short-term payback, but if you don’t risk anything you’re trading the possibility that you’ll have wasted your time for the guarantee that you’ll never get the potential benefits that may pay back all the “wasted” research. That might be a reasonable tradeoff, but just know that you’re making it!
Okay so if you’ve actually defined what efficiency means, both in terms of the inputs and outputs, and in terms of the system boundaries, now you run into one of the real core problems of efficiency: it only measures a ratio, and says nothing about magnitude.
Going back to the conversion example, where conversion is defined as units sold per unique visitor (or Units/UV), almost all new product managers and analysts would, within their first say 6 months, run into an experiment that looked like an exciting winner because it was increasing our Units/UV! And then someone would ask the dreaded question: “What do UVs look like on their own?”
This is because the absolute easiest guaranteed way to totally knock Units/UV out of the ballpark is to have way fewer UVs. It’s a ratio metric, so if you tank the denominator and somehow not lose sales in the numerator, then voila, you’ve wildly improved conversion! You don’t even need Units to stay flat, you can still get a winner as long as your Units are falling by less then UVs.
Now even though technically we would hit our goals as long as we hit our conversion numbers…. Everyone was an adult with a functioning brain who realized it is obviously worse for the business to increase conversion by having fewer visitors. So we had to define our own little sub-goal of “Yeah increase conversion, but don’t do it the cheap way.”
Ratios are important for businesses; profit is a ratio metric! Earnings per share is a ratio metric! But magnitude also matters. You can have a sandwich shop that makes $10 in profit per sandwich but only sells 5 sandwiches a year, and any reasonable person would rather take the shop that makes $0.50 in profit per sandwich but sells 10,000, even if the former is more “profitable” if you just go by revenue/cost.7
So this is where just focusing on efficiency as its own good is incredibly shortsighted. You can win all the efficiency points you want, but if you do it by shrinking the output of the business, you may end up in a worse place than where you started.
Finally, you could argue this is a natural extrapolation of the earlier points8, but I think it’s worth calling out separately: if you focus too much on efficiency and take too narrow a view of what you care about, efficiency can end up being directly counter to redundancy and reliability.
“Just in time” manufacturing, or the idea that you only have as much material at each step in the manufacturing process as needed to complete the next step, is notoriously way more efficient in terms of output/costs than traditional manufacturing approaches that tolerate having a lot of surplus resources build up along the supply chain. If you have surpluses, then by definition you’re spending resources on things that don’t directly contribute to your output, and you’re less efficient.
However, that’s only true as long as everything is running smoothly. As soon as there’s a disruption to one part of the chain, the just-in-time approach can grind to a halt, while the traditional manufacturing with surpluses can dip into those surpluses and smooth out the overall process until the disruption gets resolved. We saw this a bunch with supply chains during COVID, where the more efficient a supply chain was pre-disruption, the higher likelihood of shortages and dramatic stoppages once the disruption happened! Same was true with things like the big Southwest Airlines meltdown a few years ago: Southwest is traditionally cheaper than other airlines because they fly their planes point-to-point, rather than having planes fly back to hubs. But when too many flights get cancelled, a point-to-point schedule completely blows up and grinds to a halt, while hub-based carriers could retreat to hubs and smooth out their schedules more easily. More efficient (and therefore cheaper) but less reliable!
Similarly, it’s way more cost-efficient to not have to spend hundreds of dollars every month on car insurance…. Until the moment that you get into a car crash.
And in the business and startup world, it’s super tempting to make sure that every person is irreplaceably valuable…. Until the first day they get sick, want to go on vacation, have a kid, or quit. It’s way more efficient to have a bare minimum staff and have them be on-call to handle support in off-hours rather than staff appropriate support coverage…. Until an incident happens and you burn out (and burn through) your employees and they either get nothing done for the rest of the week or eventually quit. It’s more efficienct to run your servers right up to their capacity rather than over provisioning…. Until you get a traffic spike and everything crashes as you get overloaded.
Customers want to use products that work, and if they stop working they won’t give you brownie points for saying “well it wasn’t efficient for us to invest in reliability or redundancy.” They’re not paying for your efficiency, they’re paying for your product.9
And again, in all of these cases, it doesn’t mean that it’s always bad to be efficient. But efficiency does come with tradeoffs, and those trade-offs can run counter to the actual things that matter the most in most businesses: what you’re actually doing, the magnitude of it, and the reliability of it. After all the only perfectly efficient company is a dead one; sure it produces nothing, but at least it costs nothing! So don’t aim for perfection!
Most of these business examples will probably default to being tech-industry focused, sorry about that
Though—and this is such a “turns-out” I’m waiting for the double-triple-turns-out—it turns out that quote is actually saying “What gets measured gets managed — even when it’s pointless to measure and manage it, and even if it harms the purpose of the organisation to do so”
Opinions my own, &c.
Though even that was contentious and would change over time; focusing on units (aka number of SKUs sold) vs say orders (number of times a user makes it through the checkout flow) vs say revenue (total $ per user, regardless of how it’s split into orders or units) will suggest very different types of features. But then orders has the benefit of being able to be more easily modeled as binary “purchased, yes or no?”, while units and especially revenue are continuous metrics, and due to fancy statistics math it takes longer to measure statistically significant results on a feature experiment if you’re using a continuous metric vs a binary one, etc etc etc etc….
I leave to the reader whether that’s a good thing lol
There’s also the regulatory capture argument, saying that crypto lobbyists really want regulations not to protect customers but to make it so only the already dominant companies can afford to pay for the regulations, locking out newcomers and competition. Sure, but also…. that’s still an example where what looks like an “inefficiency” in the short term is actually in your best long-term interest! A more evil example, but still a lucrative one!
As a side bar, this is also something interesting in discussion around AI tooling and automation more broadly, in that in a purely economic lens automation is something that increases efficiency in terms of output per some input, say output/worker. And I think a lot of people’s economic hopes and fears around automation seem to correlate with their hopes and fears for the magnitude side of the equation. Do you think AI and automation will allow us to make more with the resources we already have? Then I see people being generally economically on board. But if you think it will enable us to make more with less or the same with less, then whether you’re economically pro- or anti- will depend on which side of that owner/labor table you see yourself aligning with. (I’m focusing on economic vibes here because there’s also pro- and anti- AI arguments that have nothing to do with the core economic question of automation and efficiency.)
Particularly the time horizons one
Though again, like in all of these, there’s a tradeoff! Customers do like when Southwest flights are cheaper, and if the shift to become a more traditional airline with traditional hubs increases prices, then there is something to be lost in balancing out reliability and efficiency. But I hope it’s obvious this is a screed for nuance and balance, not a screed for “the opposite of the linkedin bros but just as totalizing and dumb.”