Hacker News Reader

Comment by chewyshine | Hacker News Reader

Notice how there are no clear criteria for evaluation in this space? No math. No models. Just loose concepts strung together with words and sprinkled with calls to authority (e.g., Andy Grove) to add credibility. No evidence. No science.

As an organizational "scientist" it's amazing to me that organizations are ubiquitous and yet we know so little about how to construct good ones. Software design is in a better state IMO but not by much.

Here's a simple question that should be answerable in any approach to org design. What's the optimal *span of control* for management at each level in the organizational hierarchy? If you can't answer this question, you can't "design" an organization.

View on HN

Actually, there is a entire field dedicated to this (disclaimer: I studied it at uni, plus CS and info systems) - organizational studies [1]. It has subdomains such as organizational structure/models, behavior, communication, etc. Much of it backed by theories, studies and more, ranging in disciplines.

Time and motion studies (for example) were part of the scientific management revolution for industrial management [2]. There have been both qualitative and quantitative studies of all kinds of things - organizational forms, people networks (things like, Dunbar's number[3]), power distribution (ex. work of Pfeffer), etc. I could go on.

I do think we are reliving an era of interest in management by data and metrics, much like that of the industrial revolution and scientific management. Nothing wrong with using science and quantitative measures to optimize, but any human who has been subject to purely management by quantitative objective will likely tell you it often becomes.. rather, inhumane. This is often what led to automation, I feel - to remove the human element that was crushed by industrial efficiency.

I suspect this is why the qualitative balance is important (and no less scientific - science can be logic not just metrics right?).

My two cents...

[1] https://en.wikipedia.org/wiki/Organization_studies

[2] https://en.wikipedia.org/wiki/Time_and_motion_study

[3] https://en.wikipedia.org/wiki/Dunbar%27s_number

OP said they were an organizational scientist, so I would assume they know this. However the article rather hypothesizes and references recent books like Working Backwards, rather than long-established theories about org-design (like Mintzberg, who dates to the 60s). I’ve studied many of these and they are very thorough (though none are perfect)

I had a similar thought on another commoncog.com post. The author didn’t seem to research deeply before postulating opinions. Looks like this site allows members to post so I don’t know if it’s the same person

https://commoncog.com/blog/the-hierarchy-of-practical-eviden...

I would ask the OP a simple question: what organisations have they built, and where are they now?

Longer, but establishes the epistemology of the blog: https://commoncog.com/blog/practice-as-the-bar-for-truth/ and https://commoncog.com/blog/four-theories-of-truth/

Sorry but providing three more links to your blog does not contribute to the threaded discussion. If you'd like to ask OP that question, you can ask OP on OP's comment.

Otherwise, I'm still not sure why you don't include both (well established) theories, in addition to your own practice in your post.

Ok, since I have to spell it out for you:

I have not included any such theories because I have not found them useful. The links are to say something simple: my entire epistemology is pragmatic. That is, true knowledge should lead to effective action. If I cannot apply it and get results, then it is useless to me, and irrelevant when writing up notes for other practitioners.

I have noticed that your claim is that 'here is some theory that is well established and rigorous and old'. I have also noticed that your claim is not 'here is some theory that I have found useful, and <insert notes from application>.' I pay attention to arguments of the latter form, because it usually indicates something that I can integrate into my practice. Because you have not included notes from actual application, I am not particularly interested in your argument.

(But if you can provide an applied account, I’m all ears!)

That is not to say that org theorists are useless, or that research is useless. I have found Herbert Simon's work on organisational decision making useful as a lens on practical rationality, for instance. I’ve also spent a lot of time digging into expertise research for applied ideas.

I think the bar I use is simple: when reading a theory, I ask myself if there are actionable handles. If so, I tend to pay attention. If not, or if I’ve applied it and it doesn’t work particularly well, I discount the research.

That said, I have noticed that good organisational builders — with a track record of actually building orgs — say different things from organisational theorists. And I think the reason for why is interesting: I suspect that org builders are interested in actual org outcomes, while org theorists are too far removed from actual application; they are interested in tenure.

(Edited to soften tone.)

How do you account for survivorship bias in your chosen group of "good organisational builders"? Organisations which have been successful in their domain could have internal organisational problems but succeed in spite of them because of other exogenous factors (luck, overwhelming product/market fit etc). Are the people who built those organisations good builders? Conversely you could concieve a brilliant org design but the org itself fails. That would not make you a bad org builder.

This is an interesting question! The answer is to hold conclusions loosely and test; always test.

One way to pick who to pay attention to is ‘believability’ — meaning that you listen to those with at least 3 successes, and a coherent explanation when probed.

Pointing out survivorship bias is a common rejoinder to this view. But when you’re trying to put things to practice (not get at some universal truth like a researcher would) you often cannot wait for perfect samples. So you pick certain practices from believable people and test them against reality, and then hold the lessons loosely, making sure to update based on further experiments (which is necessary because life is messy and full of confounded variables).

Over time, it becomes clearer what is useful and works for you, and what isn’t and is perhaps a quirk of the other organisation’s context. But I’m not saying anything new; this is how we learn from life.

Related: Brian Lui’s loose feedback loops https://brianlui.dog/2020/05/10/beware-of-tight-feedback-loo...

And the problems of learning from experience: https://commoncog.com/blog/the-hard-thing-about-learning-fro...

For what it's worth, I actually do understand your sentiment. I am also a pragmatist and practitioner and find that academic theory sometimes has to "bend" for lack of a better word.

That said, I was brought up to find what is useful and good regardless of whether it comes from academia or practice. It's a beautiful thing to be able to find a balance of building an operational world view that borrows from both theory and practice.

It is my personal observation that much theory is based on observation of practice + theoretical projection, followed by more observation - i.e. the scientific method.

I loved Adam Savage's comment that sometimes science boils down to doing stuff and writing it down. I think as long as we find a good balance in learning from the past and building on it, while not letting the past limit our future endeavors for no reason, we can have the best of both worlds.

Before demanding that, first come up with metrics for the optimal attributes of a marriage. What is the optimal number of loads of laundry to do each month? The proper number of silly dances to invent?

https://en.m.wikipedia.org/wiki/McNamara_fallacy

This is a ridiculous critique of the argument. Just because everything can’t be quantified doesn’t mean we can’t quantify some things.

I worked in quant finance for many years so I’m very familiar with low signal to noise in complex systems. You can’t throw your hands up simply because you’ll never capture everything in your models.

This field is so far from my areas of expertise but I imagine there are lots of smart people investigating and putting structure around these questions.

> You can’t throw your hands up simply because you’ll never capture everything in your models.

You can however, decide that the key drivers in your domain are essentially impossible to capture quantitatively and decide not to model the domain scientifically. This applies especially well to cases where 'tacit knowledge' is important. Because that knowledge is hard to formulate, let alone formalize, it is really hard to quantize.

Taleb makes an argument in Black Swan that a bad model can do more harm than no model at all, and that "we can't throw our hands up" is not a valid excuse either: sometimes that's exactly the right call.

Right. It is also false to say there are no useful numbers in this space. For example: Dunbar’s number is 150.

But it is misleading to expect an employee to maintain relationships with 150 people—they also have a family and friends.

https://en.m.wikipedia.org/wiki/Dunbar%27s_number

From the linked wikipedia:

> However, enormous 95% confidence intervals (4–520 and 2–336, respectively) implied that specifying any one number is futile.

what’s more interesting about dunbar’s number is that it suggests a potential maximum for the size of an effective organization (say, 520) rather than pinpointing an optimum. the idea of a maximum like this appeals to intuition, so it’s worth studying more (and more quantitatively), but opposes ambition, which is probably why we don’t have plentiful research in this area already.

Do regular blood tests on a randomized sample of married couples and measure the stress levels.

Once you've found the couple with the lowest overall levels, book them on a touring circuit so audiences can ask them how many silly dances they invented. Since the couple doesn't know whether that's a source of their happiness or not, it won't really get us any closer to an answer. But at least the couple's resulting stress from the tour and impending marital problems will teach the audience about the limits of their method of inquiry.

It'll be the couple with the highest income to work hours ratio.

Below a certain [unknown] ratio, absolutely. Above it, I'm not so sure.

When you can't measure what's important, what you can measure becomes important.

The more drawn out / harmful form is: when you mandate that only the measureable is valid, and something is hard (or impossible) to measure, everyone comes up with reasons it's not important.

Thanks for the link.

I think there can be a useful middle ground, where "fuzzy" descriptions are used with models to explain strategies that are developed organically.

I empathise with the OP on the lack of modelling in this space. I think it shows a lack of maturity of the field since good, simple models are usually used to produce fundamental understanding in a field.

The optimal number of loads of laundry is whatever rate keeps you in clean clothes within bounds with which you are comfortable.

There are people who are comfortable with the zero margin of wearing their last clean clothes while doing laundry. There are people who are unhappy when they get close to that. And there are people who want to be at the top border, where no more than one or two days' wear can be allowed to be unclean at a time.

The important thing for a marriage is that you are both happy with similar rates, or that you are both happy with your partner's rate even though it is not your own.

Right. Notice the language you are using: "happy", "comfortable". Those are emotion words like "trust".

That's because life is not solely a series of numeric optimization problems; objectivity is a useful tool, but people's desires are by definition subjective.

> Here's a simple question that should be answerable in any approach to org design. What's the optimal span of control for management at each level in the organizational hierarchy? If you can't answer this question, you can't "design" an organization.

This seems so wrong to me. It is the equivalent of stating that unless you can specify the values of all the hyperparameters up front, you can't claim to 'design' a neural network architecture.

All you really need to iterate on (this aspect of) the design of an organization is a way to tell when the span of control is too large and when it is too small.

> Here's a simple question that should be answerable in any approach to org design. What's the optimal span of control for management at each level in the organizational hierarchy? If you can't answer this question, you can't "design" an organization.

Of course you can design an organization without knowing the optimal span of control at each level, just as you can design a logo without math and scientific models. The answer anyway will just be 'it depends, and span of control varies not only between different businesses but also different roles and even different individuals'.

Lots of design is done via intuition and experience rather than concrete engineering anyway, and OD is clearly an area where it is more about understanding the goals of an organisation and building a people strategy around it rather than perfect mathematical optimality.

Or, in other words -- you show me a calculated optimal span of control, and I can show you an individual at my company that would wreck it (either as too broad or too narrow).

Org design suffers from the same problem as economics and psychology: you're designing based on a fundamental discrete unit (a person) that's incredibly variable.

Except unlike the other two, you're typically not dealing in large enough numbers that you can handwave away differences and substitute averages.

Furthmore, any hierarchical org (which is to say, all, either formally or informally) exacerbates the problem in that you have some (variable!) individuals with even greater ability to influence the sum.

Which isn't to say it's hopeless, but is to say (to your point) that any approach needs flexibility and intuition.

Or as the author puts it: "As a result, you cannot predict how the humans in your organisation will react to your changes — not with perfect accuracy, at least. So the nature of org design demands that you iterate — that you introduce some set of changes, watch how those changes ripple out in organisational behaviour, and then either roll-back the change, or tweak in response to those observations."

If we talk about organizational structure, the one real question is whether you can just look at the structure and figure out a organisations success just based on that.

My feeling is that this could be a good way to filter out highly dysfunctional organizations. However I don't think you can find successful ones that easily, let alone come up with an magical organizational structure that automatically leads to success.

That might be because there are a lot of tiny details that might shape a organization much more than the pure structure of departments and roles. Let's say organization A and organization B have the same structure and do the same thing in the same field, but organization A has a good HR department which manages to attract good people and have them work for decades at the company, while organization B has a bad HR department, hires incompetent, fraudulent and downright nasty people, who don't stay on the job for long – wouldn't this make such a huge difference that differences stemming from the pure structure of the organization would be drowned out?

Of course you could now think about how a organizational structure could prevent this from happening, and maybe with the right structure and people checking each others decisions the likelyhood of such a bad outcome could be mitigated – but never fully.

You can look at a chain, and see: This link is broken, the chain is broken. But if it looks intact, you cannot just from looking at it, know how strong it is.

But we do have a science of organisation design; management cybernetics are 70 years old, and as this post hints at, the field of complex systems theory (also 70) applies to organisations. But these fields don't find context-independent answers like the one you asked for; an extremely important trait of systems engineering (as, again, was mentioned in the post under "form-context fit") is that systems have fitness for a specific environment or context. Answers are adaptive and must be approached in-context, not declared to be canon in some "objective" scientific whitepaper.

If it's math you're after, look at Control Theory or Nonlinear Dynamics. They work well for engineering purposes, but good luck modelling individual human behaviour accurately, let alone mathematically.

I would argue that organizations are complex adaptive systems and thus there is no optimal span that holds for all or even most organizations at all or even most levels. As a solution, smart organizations should build in feedback loops and allow rapid experimentation and iteration to evolve the spans so they respond well to the current conditions at each level and each point in time.

Note that setting up an organization to be so responsive and adaptive is itself a difficult organizational design problem.

A lot of data structures from compsci like trees, graphs, lists, seem suited to form the basis of mathematical modeling of organizational structures, with the goal of optimizing for particular cases.

For example, in modeling an industrial system like a chemical refinery / synthesis unit for optimal throughput, one could also model the human organizational structure needed to safely and efficiently operate that system. Say there were 10 major steps/processes being overseen; failure of any one could be catastrophic. So, perhaps each unit gets its own manager with veto power over the whole process if their unit is down (a flat structure at this level), and each manager oversees a hierarchically-structured team (a tree at this level, perhaps experience-based).

Other organizations would need a completely different structure, but it should be structured around the fundamental goal. Thus, the concept of 'universal organization designer' might be so broad as to be not very useful, i.e. specialization in design domains is probably important.

I recall this coming up in a discussion of why the optimal organizational structure for Tesla is very different from that for SpaceX for example, so just moving 'the best managers' from one to the other wouldn't work out.

IMHO this is wishful thinking. There are thousands and thousands of questions you could pose in this way, each with a range of answers depending on many details specific to the job to be done. Even just designing a single experiment that is not subject to biases of artificial metrics or incumbent market momentum is incredibly difficult.

For instance your span of control question depends on how much individual bandwidth is needed between the levels, which in turn depends on the nature of the work and how it interacts with partner functions and whether it can be routinized or whether there is an aspect of creative problem solving.

It’s a pleasant fantasy to imagine we could get definitive answers using science but it presumes there is a universal maximum when in fact there are many local maxima depending on goals and the individual strengths and weaknesses you’re actually dealing with. And even then org structure is a pretty blunt instrument which is always a huge tradeoff. All orgs rely on extra-organizational effort to address critical problems, whether it be through formal working groups or just individual hustle and resourcefulness.

Models need input and there isn't much knowable input in modern organisations. A unit of organisation is a person with complex internal state. Two units of type M could've competed for another unit of type F and one of them has won, while the other is secretly sabotaging work of the first - an example of chaotic to an outsider behavior because of hidden state. When all sorts of interactions are allowed, particles of organisation interact in all sorts of bizarre ways. It's hard to model a gas where particles have memory, long distance interactions with ten types of forces, mutate into other particles, teleport back and forth according to God knows what reasons and so on. We either resort to making only very general predictions or we cool down the particles, restrict their freedom to bare minimum and make them predictable. Rogue regimes do exactly this: the only interaction they allow is a strictly top-down "who fears whom", so everything is local and predictable.

> Notice how there are no clear criteria for evaluation in this space? No math. No models. Just loose concepts strung together with words and sprinkled with calls to authority (e.g., Andy Grove) to add credibility. No evidence. No science.

Indeed, this is a very good observation from which many ideas occur to me:

Which one do I prefer?

Is one obviously better? (I don't think this is a good question: It's like asking which is better, an API reference (a math textbook with a long list of theorems and definitions) or an API tutorial (a math textbook which holds your hand and explains "intuitively"). It depends on what you need).

Isn't it the case that initial explanations (explorations into a new topic) are like this at first, and over time (usually through work spanning multiple generations) the theories become more mathematical?

All in all I wonder about the difference between these two contrasting approaches towards understanding. And I wonder about it in such abstract (philosophical?) terms that the specific "organizational design" is just an instance of what I'm curious about; which is the different ways to explain the same things and other ways to approach "understanding" in general.

> No math.

"Math" in the social sciences is often a smell for "physics envy". Economics if full of mathy looking things to give the appearance of rigor without actually having any.

Also, I would add one important thing that's being omitted, namely, that science is concerned with principles (at least instrumentally useful ones if not truly accurate ones). So there may be general principles that underpin good organizations, but these principles are general and on their own are insufficient for practical action. You need to be able to combine those principles with the contingent facts to produce action and that requires the capacity for prudential judgement.

I think this might be roughly what you're looking for: https://codahale.com/work-is-work/

But this author is concerned more with 'productivity' rather than longevity or interpersonal relationships.

> Software design is in a better state IMO but not by much.

Looking at all the crap software being built today, I am not sure I agree even with the caveat of "not by much".

Reply on news.ycombinator.com