Greenfeld comes perilously close to independently discovering Hayek. Local, specialized knowledge is essential, and is one of the hardest features to scale, if it's possible at all.<p>Also, this reminds me of stories about Soviet manufacturing quotas[1]:<p>> The Soviet Union was the largest producer of shoes in the world. It was turning out 800 million pairs of shoes a year–twice as many as Italy, three times as many as the United States, four times as many as China. Production amounted to more than three pairs of shoes per year for every Soviet man, woman, and child.<p>> The problem with shoes, it turned out, was not an absolute shortage. It was a far more subtle malfunction. The comfort, the fit, the design, and the size mix of Soviet shoes were so out of sync with what people needed and wanted that they were willing to stand in line for hours to buy the occasional pair, usually imported, that they liked.<p>Or even more ridiculous[2]:<p>> But probabably the most outrageous malfunction of the system was of a factory that fabricated metal products. It had a quota for scrap metal for recycling. The factory was operating more efficiently than expected so it was not generating as much scrap as expected. Since the factory was not meeting its scrap metal quota the higher authorities were going to fine it. So the factor personnel did what to them was rational. They took perfectly good zinc metal sheets and converted them into scrap to fulfill their scrap metal quota.<p>When looking at large organizations through the lens of information problems, one realizes that Amazon, Google, Walmart <i>et al.</i> all exist on continuum along with the Soviet Union or even the US Government.<p>[1]<a href="https://www.econlib.org/archives/2009/09/soviet_shoes.html" rel="nofollow">https://www.econlib.org/archives/2009/09/soviet_shoes.html</a><p>[2]<a href="https://www.sjsu.edu/faculty/watkins/stalinmodel.htm" rel="nofollow">https://www.sjsu.edu/faculty/watkins/stalinmodel.htm</a>
"Frontline" ran an episode some years back where they investigated some dental clinics that did astonishing numbers of unnecessary major dental operations.<p>It turns out that the government had a program that paid for these operations when performed on poor people. So the clinics would convince their customers that they needed major dental work, then would do the work and bill the government. The clinics were totally specialized just for this.<p>"Frontline" pinned the blame on the clinics, sort of "how could the do that". Of course, they're right. But blaming the clinics doesn't work, because the actual problem was the government set things up so that the rise of clinics to exploit the rules was legal and inevitable.<p>You're now faced with the problem of crafting a bureaucratic rule to define a "necessary" major dental operation. Of course, it's impossible, and the same problem as "management by metrics".
A bit of a segue, but this management by metrics goes beyond just revenue.<p>KPI, KRI, KCI, etc. Truly a heap of metrics that eat out the time of middle management, and consequently of the people down the hierarchical pyramid. It over-allocates resources towards resource management itself rather than towards products themselves.<p>There is a drive to justify your existence, and that of your fellow workers, when your company becomes large enough that it has bureaucratized and, moreover, acquired some form of legacy.<p>David Graeber in Bullshit Jobs mentions that two categories of BS were the box-tickers and the taskmasters. They are evidence of this management by metrics.<p>It's all about budget: gotta spend that hard-earned budget to find ways to save budget on your budget-consuming core products... to make sure your budget allocation is the same next month or year.
We should do better than to be ignorant of national and global decisionmaking and responsibility in public health and pandemic epidemology.<p>It is a baseless claim that public health is systematically ignored in favor of pushing the pandemic’s reproduction number down. Do we really expect public health officers to be untrained, stupid, and dismissed? Is that what we really are seeing? Do we really think that public officials are ignorant of the metabolic and social implications of having to drive stakes into society?<p>It is a fact that in functioning states, public health decisions are taken within an ethical framework of how to do least harm and preserve the most freedom, considering resource availability and longer-term effects of public health decisions, including the interacting feedback loops of the economy and public health.<p>While overreliance on small metrics is clearly bad, this style of off-hand ignorance of public service is not what we need.
Great article! I also found that the aim to be objective in everything and use metrics as truth and not just another input is pretty harmful over the long run in business and science. Things that are hard to measure are ignored most of the time.<p>The best example are apples. When you breed apples to be 3% larger each year but taste slightly worse, you can weigh them and see that they got slightly larger. As you cannot accurately measure taste you will notice that they basically taste the same. After years of breeding you have giant apples with bad taste.<p>The author's proposition of using as many metrics as possible and talking to real people is a good solution to make better decisions.<p>I would add that for some decisions you should actually just focus on the few facts that truly matter. The classic example being a potential business partner with a bad reputation. If somebody is not trustworthy everything else probably doesn't matter.
Marketer here and this is so spot on. So many companies that I have worked for have had founders that just didn't seem to understand metrics as well as they thought they did.<p>Huge spread sheets with % and pivot tables and week on week breakdowns BUT no one could tell me much about what all that data meant. Apart from the fact that everyone had their own explanation it was just useless information to be fair. Churn, CAC, amongst various other 'hot' metrics.<p>In the end, most of these companies took a couple of years to realize that these metrics which Unicorns write amazing blog posts on are either BS or just not made for them. So much frustration was felt because a manager read a TechCrunch article about how X (with 100's of millions) did this and achieved something spectacular.
The book "Seeing like a State" expounds on this topic significantly, using scientific forestry as an example. The "recipe for failure" [0]:<p>1. Look at a complex and confusing reality<p>2. Fail to understand all the subtleties of how the complex reality works<p>3. Attribute that failure to the irrationality of what you are looking at, rather than your own limitations<p>4. Come up with an idealized blank-slate vision of what that reality ought to look like<p>5. Argue that the relative simplicity and platonic orderliness of the vision represents rationality<p>6. Use authoritarian power to impose that vision, by demolishing the old reality if necessary<p>7. Watch your rational Utopia fail horribly<p>[0] <a href="https://smus.com/books/seeing-like-a-state-by-james-scott/" rel="nofollow">https://smus.com/books/seeing-like-a-state-by-james-scott/</a>
I am currently refining the metrics I want to present to my management. I work in information security so we do not exist until a issue arises.<p>I base the value of a metric presented to management on the "and so what?" basis. In other words: how it helps them to better run the business.<p>There is the "annual objectives" kind of report: how much we did and how good we are. Meaningless.<p>There is the "current events" one (nr of incidents, etc.). Meaningless.<p>There is the "these are systemic, company-wise issues I need you to arbitrate over", or "we need to do that, but it means changes and incomfort". Actual decision making data, the worst to present because they show problems and not celebration.<p>I guess that this is the case of these organizations that do not provide green money to the company (aka. "cost centers"). i hate these metrics.<p>(please note that there are metrics that are useful to *operational teams* so that they realize there is a trend, or that some performance upgrades will be needed, but they should stay there)
For those inclined, there's a nice article by Marilyn Strathern, The Tyranny of Transparency [0] that echoes the sentiments in this article. It builds upon Hari Tsoukas' The Tyranny of light: The temptations and the paradoxes of the information society [1], which I heartily recommend if interested in reflections on this topic.<p>To entice, here's a small excerpt from the abstract of [1]:<p>> The overabundance of information in late modernity makes the information society full of temptations. It tempts us into thinking that knowledge-as-information is objective and exists independently of human beings; that everything can be reduced into information; and that generating ever more amounts of information will increase the transparency of society and, thus, lead to the rational management of social problems. However, as argued in this paper, the information society is riddled with paradoxes that prevent it from satisfying the temptations it creates. More information may lead to less understanding; more information may undermine trust; and more information may make society less rationally governable.<p>[0] <a href="https://www.tandfonline.com/doi/abs/10.1080/713651562" rel="nofollow">https://www.tandfonline.com/doi/abs/10.1080/713651562</a> (alternative: <a href="https://sci-hub.se/https://doi.org/10.1080/713651562" rel="nofollow">https://sci-hub.se/https://doi.org/10.1080/713651562</a>)<p>[1] <a href="https://www.sciencedirect.com/science/article/abs/pii/S0016328797000359" rel="nofollow">https://www.sciencedirect.com/science/article/abs/pii/S00163...</a> (alternative: <a href="https://sci-hub.se/https://doi.org/10.1016/S0016-3287(97)00035-9" rel="nofollow">https://sci-hub.se/https://doi.org/10.1016/S0016-3287(97)000...</a>)
Great article. My take on it:<p>- Metrics give everyone a false sense of objectivity. If you actually look at any stats in depth, you will end up making decisions about what to measure, and what the numbers mean. These words are actually the most important part of any study, but are always left out in favor of the graph. Want to know how many people died of covid? You need to consider things like population adjusted deaths. And adjustment for age. And a a whole host of other things to do with how the data was collected, and how it was presented. The economic side is even more messy. How much would GDP have been if we'd done this or that? There's no number that's the answer, but we do need numbers.<p>- You'll always care about more metrics than one. There's no generally objective way to decide how to weight those. You might have a utility function that decides this, thus reducing the problem to a simple optimization, but the guns/butter balance has still got to come from some kind of value that you express.<p>- The big question is rules vs discretion. You set up an elaborate measuring system (eg GDP) and you imagine that you have certain levers that you can move in response to whatever situation occurs. Part of this is that actors expect you to behave a certain way. But what happens if something you thought would work mechanically doesn't behave like you thought? Should you move your levers contrary to your previous commitment?<p>- Finally, there are the unmeasurable sensations. You have a team, it feels wrong somehow. You talk to people, get opinions, you get a feel for what is wrong. None of it is something you found in your LOC/user summary, or CI pipeline fails, or whatever.
"Management by reliance ONLY on metrics leads us astray"<p>Gut vs Metrics don't have to be either/or. Gut also has its own limitations - Daniel Kahnemann has documented this very well in Nobel Prize winning research. So what should one do? An organisation has to be self aware that there are no "perfect" metrics. Therefore, one should clearly state the limitations of a metric(s). Then, make a qualitative judgment based on the metrics and their limitations. You make the limitations, assumptions transparent and communicate the risks, uncertainties. Metrics are an input to decision making, they aren't decision makers themselves.
My favourite illustration of this is in The Wire. The city wants to crack down on crime and increases the target number of arrests. The police chief sends out the word and police officers begin making a bunch of spurious arrests. Do you blame the politicians for picking crappy metrics and for their poor understanding of human nature? Or do you blame the individuals who choose to exploit the metric out of self-interest? Plenty of blame to go around.
It's the same result when one tries to run an economy through regulation, for the same reasons. A better way would be by incentives.<p>For example, there are regulations that 100 ppm of pollution X is illegal, while 99 ppm of X is legal. Emit 99 ppm, and you're good. Emit 100 ppm and your plant gets shut down. Even worse, you can emit 99 ppm for 10 hours and you're legal, but emit 100 ppm for 1 minute and you're illegal.<p>A far better approach is to tax total X emitted, not the rate, and not binary legal/illegal:<p><pre><code> 100 ppm for 1 minute times $1/ppm = $100 tax
99 ppm for 10 hours times $1/ppm = $59,400 tax</code></pre>
It's not the metrics per se - but the metrics not being totally aligned with the required outcome.<p>Good project management means being better at finding the right metrics - or at least recognising that our metrics aren't good enough, often because we haven't all of the information - why are we hiring a new developer, why is this piece of software being developed etc.
I've experienced this first hand so many times. It is so difficult to steer the boat when the management depends heavily on metrics they don't understand the need of. Just because others are doing it. And as the flow of pressure goes from top to bottom, engineers end up working towards achieving a metric which burns out the good ones and they leave.<p>I've spent sometimes hours in talking to clients on how they should first focus on metrics that help them ask the right questions.<p>This is prevalent everywhere, Scrum-masters holding meetings to discuss the scrum velocity.
While I agree with what I read as the premise of the article (that metrics for the sake of metrics is harmful and that qualitative data conversation is a great supplement to quantitative metrics) I wish the post was written with less hyperbole, especially the around big companies and politicians sections.<p>My take is that quantitative metrics are incredibly useful when paired with strong leadership, a clear mission, and vision statement that give human context to the metrics. That way whenever “the numbers” are brought up, the context and focus are around goals that really matter (are we creating a product that is meaningful and making people’s lives better?) versus cherry-picking and gaming the system to meet KPIs. Quantitative metrics are a dangerous tool since numbers are so easily digestible, so special care should always be given to understand exactly what the metrics are holding people accountable to.<p>An example that comes to mind in personal life is fitness. Weight is such an easy metric to get your hands on... doubly so since you can quickly judge weight by looking at yourself in the mirror. It’s really easy to tunnel on trying to lose weight! The activities (both physical and mental!) encouraged by fixation on weight can be incredibly destructive. If instead I first start with a strong vision statement that aligns with my values: My body is my most important possession; I want to keep it healthy so I can live a fulfilling life. In the context of that purpose weight is an important metric to keep in mind as a risk factor, but it is also clear that there are a multitude of other metrics that will allow me to measure, stretch, and grow towards a productive end.
There’s a sweet spot that’s hard to hit. On one end you have people who ignore metrics. They get themselves into trouble breaking things without making the easy measurements to know they’re breaking them. They make obvious mistakes.<p>The next group makes subtle mistakes. They’re the ones who have learned how to use metrics and now they have a hammer, everything looks like a nail. I see this a lot in less senior data scientists. They know how to drive with data, and so everything must be done that way. But they make the mistakes this article gets at.<p>The group I like takes a middle ground approach. We recognize that certain network effects are so coupled that you can power maybe one proper experiment on it at a time, and even those experiments might take too long. Other experiments take too long simply by virtue that the thing you want to measure might be long term behavior. So you experiment on what you can, get everything you can by proxy, and make sure to at least measure health via the metrics that matter, but you don’t try to base every decision in something measurable simply because it’s measurable.
<p><pre><code> > A resume will never tell you if a person is genuinely interested in uncovering unknown truths about nature. But a 30-minute conversation will.
</code></pre>
A 30-minute conversation with a friend might, but this is implicitly talking about an interview setting. How confident are you in your own ability to tell whether "a person is genuinely interested in uncovering unknown truths about nature" knowing that nobody has been able to come up with a similar formula for interviewing programmers, where there are at least in theory easily measurable knowledge levels?
'OKRs before KPIs'as it says in our culture deck
<a href="https://citizenshareholders.com/culture" rel="nofollow">https://citizenshareholders.com/culture</a>
If achieving your KPIs compromises your Objectives and Key Requirements you have the wrong KPIs.
In the start-up/scale-up world this there can be enormous pressure to meet investor expectations on short term revenue/growth forecasts.
Part of the job of founders is to know their market, their company's unfair advantage.
'Don't be Bullied' should probably be the title of a blog post calling on founders to resist investor pressure to follow the norms in achieving early revenue, not because early revenue is a bad thing, far from it, it's amazing. But if the cost of generating that revenue puts OKRs at greater risk, its better to raise more money to extend your runway.
If you meet your OKRs, assuming you've chosen the right OKRs, your KPIs will follow.
In many ways this is a culture issue:
Culture trumps Strategy, OKRs trump KPIs
Provided you have the cash.
Citizen Shareholders is now looking for that cash. When we're asked why we have no revenue projected for 2 years, it's because being able to survive for 2 years without revenue will derisk us to the early B2B adopters.
Having that early revenue in the bank alongside those investor funds is a problem we can live with.
Just in time supply chains are so very 2019, we plan to make sure we have enough to succeed by, at times, having more than enough.
If you want to find out more: CitizenShareholders.com
Thank you!<p>Your post hit really hard, because a) Sole focus of my team IS MRR. b) I wrote something very similar.
Sorry if my comment is a little maendering, I am a dad as of 7 days and what they tell you about sleep is no joke.<p>I want to plug in one of the less famous Drucker quotes:<p>> Working on the right things is what makes knowledge work effective. This is not capable of being measured by any of the yardsticks for manual work.<p>It's true that "What gets measured gets managed" by the converse is also true - if you focus on one metric only, the rest will fall into disarray.<p>It's good to keep your eyes at the gauges when you are flying, but just as the map is not the territory, the measurement is not the real thing.<p>People love metrics, because they give them an illusion of control over a complicated, real world, messy reality. The core of the problem is unfortunately human bias to simplify. We as a wider tech community are contributing a bit introducing yet-another-dashboard.<p>But as you rightly point, the metrics people pick are the ones that are easiest to measure. NOT the ones worth measuring. In fact I published a post quite similar to yours: <a href="https://piszek.com/2019/11/24/metrics/" rel="nofollow">https://piszek.com/2019/11/24/metrics/</a><p>To broaden the theme, I would even argue that Taleb is complaining about roughly a similar set of problems: Gaussian curve is an oversimplification as well and has caused a lot of suffering when misapplied.<p>Despite all of these caveats, there is a benefit to having a focus (like MRR you are mentioning). Having a singular focus can be transformative and can make one developer more effective than 30-people team with no real direction.<p>In that sense, there is benefit to metrics. As with all real-world things, I think the only solution is a dynamic equilibrium, where you change from one approach to the other to cover your bases.<p>As I wrote in the beginning of this comment, the Team I Lead (we build simple ecommerce tools) was tasked with maximizing MRR and the side-effect of having that focus was that we ignored the one-off sales features that our customers wanted desperately.
>> Who else manages by metrics? Politicians!<p>This is completely false. Politics is the least metrics driven, mostly intention driven.<p><a href="https://www.washingtontimes.com/news/2014/apr/25/sowell-good-intentions-bad-results/" rel="nofollow">https://www.washingtontimes.com/news/2014/apr/25/sowell-good...</a>
<i>> You know who manages by metrics? Big companies like Google, Amazon and LinkedIn.<p>And what do they have in common?<p>Their core product got notably worse over time.</i><p>It’s interesting he doesn’t mention Apple. I have no insight but from the outside of say they have very different things they optimize for than revenue. Money merely pours out of their customer obsession instead.
As usual: If you can't measure what's important you declare important what you can measure. That wouldn't be so destructive if management at least was aware of that problem. But they aren't. They'll have their report of metrics which are green or red. Green is good. Red is bad. End of story.
All employees have radically different ideas about how to organise for success, ranging from free market capitalism to the government literally seizing control of this specific company. The skill range is from "needs constant supervision" to "good at a very limited range of activities". Any specific manager has 40 hours a week, which gets eaten up very quickly (10 reports = 4 hrs each to figure out if you like what they are doing assuming no actual work the manager needs to do or surprises that mean they can't work to schedule).<p>The issue isn't that anyone believes that metric-driven management is a good option, it is that at the scale of 2,000+ people there are not a lot of alternatives. I could easily design a system that works at large scale ... but only by the employees and owners being the same people. Apart from that, the erratic us-them nature of stockholders and employees makes all the good structures untenable.<p>Lay out an org chart for 2k people then ask how long the manager has to work out what their reports are doing. Expand the hierarchy far enough that everyone has time to do a good job and then ask what the odds are that all the people in positions of importance agree on how things should be done by whom. To settle the number of disputes that arise in a big tree-like org structure there isn't have much of a plan B apart from metrics.<p>It would be nice to see more governmental support for limited liability cooperative business models. The organisational opportunities that would open up could allow for so much. Not to mention maybe workers could earning more.
> You know who manages by metrics? Big companies like Google, Amazon and LinkedIn.<p>This isn’t true through. On this very site’s Startup School you will find lots of experienced entrepreneurs advocating for a metric-driven approach to growing your startup. Maybe not “one metric that matters” all the time, but Lean Startup is about measuring your company’s metrics and running experiments week-by-week to see what moves the needle.<p>Sure, slavish and myopic adherence to a bad metric will lead you astray. But management without metrics will lead you astray too.
That's not true and true at the same time. There are myriad instances where management by metrics leads companies totally astray, bit it can easily argued that they are badly managed or have a bad strategy.
The core of any kind of measurement effort is a clearly defined objective that sits in balance with other objectives, and the important thing is these objectives are sometimes in conflict with each other, like finance and UX, measuring both gives the company a chance to find the right balance.
Eric Reis' "The Lean Startup" covers this topic in detail, and makes a useful distinction between "vanity" metrics and "actionable" metrics. Aside from the book, you can read a blog post by him on the subject of metrics here:
<a href="https://tim.blog/2009/05/19/vanity-metrics-vs-actionable-metrics/" rel="nofollow">https://tim.blog/2009/05/19/vanity-metrics-vs-actionable-met...</a>
By using metrics we make it a lower dimensional model, that can be managed at scale. However in real world, it often becomes a overfit one. In large scale execution, people only care what metric measures.<p>A talk vs metrics is just higher dimension vs lower one. The information within a talk is complex. Different people received vary information from the same talk. So it's hard to managed by scale.<p>So a metrics system is just a trade-off for its scale.
I find this to be very very rich.<p>A guy who lambasts metrics, but whose projects are all based on scraping metrics.<p><a href="https://productexplorer.io/" rel="nofollow">https://productexplorer.io/</a>
<a href="https://newsletterspy.io/" rel="nofollow">https://newsletterspy.io/</a>
<a href="https://gumspy.com/" rel="nofollow">https://gumspy.com/</a>
I don’t think there is an alternative to management by measurement if scalable is what you’re trying to be.<p>That doesn’t mean you don’t have long term goals or nuanced objectives even if your measures are coarse, nor does it mean your measures stay the same.<p>I think optimising for measures focuses activity , which again doesn’t mean you have to optimise for the same thing all the time using the same measures.
I'm a but cynical about the recommendation of talking to people. The interviews then turn into its own optimisation game (I think politicians for instance do base their decision making on talking to people, hence lobbying). The same thing happens with job interviews (what is the most useful skill to getting a great SWE salary? Interviewing well).
Metrics in software is a bag of issues in most companies I see. But there ARE useful ways to use metrics that actually help (software). I did a talk about this at GOTO: Lies, Damned lies, and metrics.
<a href="https://www.youtube.com/watch?v=goihWvyqRow" rel="nofollow">https://www.youtube.com/watch?v=goihWvyqRow</a>
The article fails to acknowledge that there is in fact two very distinct families of metrics.<p>1. Taylor metrics - The ones invented by management.
2. von Braun metrics - The ones invented by the guy that put men on the Moon. These are everywhere it really matters.
Article shows companies like Google and Facebook. However doesn't talk about big companies managing in a different way. Why? They are all dead. Entropy and decay is part of product cycle. There is no way around it. You push them far away.
The metrics being discussed here are largely quantitative. OTOH, a <i>qualitative</i> metric is better for measuring if your product is turning to shit over time. Both types are needed to manage big complex things.
I very much subscribe to combining this 'Common-sense filter' to metrics when choosing a course of action, and came up with a term for it: inform your gut.
The conclusion of this article is basically "I don't understand how to use common sense and data at the same time, so I'll hark back to old-timey "gut instinct" and good-ol-fashioned selection bias hidden by my own lack of self-understanding to validate my ignorance, oh and here's some BS about lockdown policies."<p>What a joke, all of these posts have the same conclusions: don't deal in absolutes.
> Their core product got notably worse over time.<p>> Google’s search results are dominated by ads<p>Googles core product are ads, not search engine!
While is is a good (and often discussed) point that the use of metrics causes distortions as the metrics themselves affect behavior, I am not sure that the author's proposed solution of relying instead on intuition is an improvement. While the data may introduce distortions, ignoring it in favor of "going with my gut" is how we get vaccine denialism, conspiracy theories, racism, and so many other negative outcomes.<p>I am also not convinced by the inclusion of the R number in the list of examples - it is unlike the others in an important way. The problem with metrics usually arises when the metric is a proxy for the goal. For example, test scores are not the same as knowledge, and so using test scores as a metric can result in teaching to the test, such that the metric is no longer a good proxy for knowledge. But the R number is not a proxy - it is the actual thing we want to change. (Of course, there are many knock-on issues, such as how we measure that number, which do introduce proxies. Discussion of that would be a much more interesting article.)<p>Which actually points towards the real solution, which is not to simply rely on gut instinct, but to be aware of what you are measuring, and how it relates to what you want to know.
Much boils down to the question about the quality of the metrics. In my work they are just numbers that are manipulated to look good so that everyone gets their bonuses and the bosses are happy.
If anyone is interested in reading about this topic in terms of late stage capitalism I would highly recommend a book by Mark Fisher titled Capitalist Realism.<p>In it he talks about “Stalinist Capitalism” I.e. a capitalist society that is obsessed with metrics which optimize for arbitrary goals at the expense goals which are intangible or hard to measure. It focuses on the metric obsessed cultures that have developed within many private organizations.<p>Two examples come to mind, universities prioritizing publishing count vs publishing impact and lines of code written vs impact of code.