I’ve been trying hard over the last several weeks to wrestle a very tough idea to the ground: economies of variety. Yes, there is such a thing, and I don’t mean either the Starbucks menu of mass-customized combinatorial choices or some charming favela economy that has variety, but not economies of variety. Economies of variety are related to, but not the same thing as, the idea of superlinearity.
I’ll leave that subject for another post, when I beat the thing into some sort of submission, but the process of wrangling the idea has led me to a much deeper appreciation of the two existing economies — of scale and scope respectively — that characterized the industrial age. So this is a sort of prequel post. If a well-posed notion of “economies of variety” can be constructed, it will need to be really solidly built in order to punch in the same weight class as these two mature ideas. A business that achieves all three will be close to unbeatable by competing businesses that only manage one or two out of three.
Amazon is the first company that is getting dangerously close to 3/3. That should give you a hint about where I am going with the economies of variety idea. But let’s figure out scale and scope first.
If you’re like me, you probably encountered the idea of economies of scale in a freshman economics course. You probably never encountered the economies of scope idea at all (I’ll explain why you didn’t in a bit). You were probably introduced to the former in book-keeping terms: the idea that large production runs allow you to amortize fixed costs over more instances, driving down unit costs.
This is going to sound really obvious and dumb once I explain why, but this is not the right definition of economies of scale. This is the effect of successfully achieving economies of scale.
Economies of scale and scope (and variety, though we won’t go there today) are both types of learning.
- Economies of scale are the advantages that can result when repeatable processes are used to deliver large volumes of identical products or service instances. Scaling relies on interchangeable parts either in the product itself, or in the delivery mechanisms, in the case of intangible services.
- Economies of scope are the advantages that can result when similar processes are used to deliver a set of distinct products or services.
As a first approximation, you could say that economies of scale result from learning the engineering, while economies of scope result from learning the marketing. The first is primarily a one-front war between a business and nature. The second is primarily a two-front war where a business fights nature on one front, and market incumbents on another. As an aside, both kinds of learning are war-time learning: they proceed in an environment where failure equals death for the firm.
More on this after we look at the details of the two learning processes.
Learning in Scaling
The key to economies of scale is process learning of the sort that the consulting firm BCG codified with its experience curves in the 1970s. Amortization of fixed costs across many instances is merely what makes the learning worthwhile, but the work of scaling lies in the learning. Getting to repeatability in an engineered process takes conscious and deliberate effort.
You can also think of scaling as the process of proving a steady-state financial hypothesis in a specific case. In other words, the amortization argument, which does not include the learning costs in getting to the design scale, is a hypothesis that you must set out to prove by construction. The equation is only true once the learning is over (and as we’ll see, it is therefore a “peacetime” model of business that applies during periods of detente between periods of business-war). The unknown learning costs are what might kill you. And usually they do, which is why pioneers rarely own markets that they create.
The ingenuity involved, I am now convinced, actually exceeds the ingenuity involved in coming up with the unscaled idea in the first place. Why do I say this? Because people who come up with great product ideas are a dime a dozen. People who figure out how to successfully scale an idea are far rarer. We tend to lionize “inventors” but the real heroes are probably the “scalers.”
Why exactly is there learning involved in scaling at all?
- The law of large numbers: the more you scale, the more you expose your operations to rare phenomena that are expensive to deal with. Scaling is about dealing efficiently with events that occur with a predictable frequency. Hard disk failures are rare catastrophes for individuals. They are an operating condition for data centers.
- Staircase effects: Capacity increases follow a staircase curve, but demand changes smoothly. You can only buy one airplane at a time. You cannot buy half an airplane for an airline. So you’re constantly undershooting or overshooting your capacity requirements while scaling. A particularly severe (but non-commercial) example is scaling an ordinary navy into a blue water navy with aircraft carriers, a challenge China is currently taking on. You generally need 3 carrier groups to have one in deployment at all times, and it takes a couple of decades (or a very active war) to climb the three-step staircase.
- Loss windups: When you are running a small bakery, if your oven is malfunctioning, you might lose one batch of cookies before shutting down to fix the problem. In a scaled operation, due to the larger distances between loci of problem creation and discovery, and the sheer speed of operations, huge losses can pile up before you intervene. Soft failure cases are predictable inventory problems. Hard failures? Think about events like the Firestone tire recall and various instances of contaminated food products being recalled.
- Accounting illegibility: Chances are, while scaling, you are slashing prices as fast as you can to grab the largest possible share of a new market. Such phases are called “land grabs” for a reason. Margins may seem strong but that’s only because the accounting simply cannot model and track a growing and learning operation accurately. Effective margins, after factoring in risks and crisis response costs, may be much lower than you think. Contributing to this is poor financial governance during scaling phases leading to a lot of waste, both justified (getting a major new order by any means necessary) and unjustified (people taking advantage of the chaos to indulge in profiteering)
- Process Design Evolution: There is an enormous amount of iterative process redesign involved in successful scaling. As quickly as you discover rare conditions, unexpected operational risks and other blindside phenomena, you need to bake the knowledge into the process. This process must not only proceed very fast, but it has to be very elegant. A bad process adaptation to handle a contingency (think TSA security procedures following 9/11) can end up being both costly and ineffective, and add entropy to the process without increasing its capability.
- Human Factor Variances: If people are involved, such as in scaling a sales operation, you have to very suddenly turn tacit, creative knowledge in the heads of the pioneers into explicit knowledge that can very cheaply be imparted to the cheapest available brains capable of handling it. In the process you may discover that your tacit knowledge is simply too expensive to codify and scale. This training failure can kill your business.
- Gravitational Effects: When you scale, you start to influence and shape your environment rather than merely reacting to it. When you launch a small satellite into space, you can ignore its effect on the earth in orbit calculations. When you are talking about the Moon, you get a proper 2-body problem. One manifestation of gravitational effects is litigation. Get to a sufficient size, especially in America, and you are suddenly worth suing. Another interesting gravitational effect is late-stage growth investment flooding in: dumb money with growth expectations that might be unreasonable/greedy enough to kill the company.
- Lucy Effects: Think about the classic scene in I Love Lucy where Lucy is working on a chocolate assembly line that moves faster and faster. When she fails to keep up, she has to start stuffing her mouth with chocolate. As with fluid flows going from laminar to turbulent, process flows too, experience phase transitions. To keep them efficient (“laminar”) with increasing velocity, you may need to reinvent (or refactor) the process entirely. These hidden reinventions can sometimes be harder than the original inventions.
When you step back and think about all this, you realize that scaling is basically the equivalent of deliberate practice (the 10,000 hours idea) for companies. The COO is typically the unsung hero leading this scaling process (and often is promoted to CEO during the transition to a scaling phase).
By leaving the unpredictable learning costs out of the equation, Economics 101 professors tend to make scaling sound like a matter of so it shall be written, so it shall be done pronouncement. In practice, the outcome of scaling efforts is anything but certain, even for a wildly successful product. If you can find the right sort of talented people to drive the process the first time you attempt it, you will find that you can improve your process capabilities just slightly faster than you are increasing production volumes. Enough to deliver something approximating the cost lowering promised by the micro-economic calculations. The equation is only true if your learning costs come in under the hidden, assumed threshold. Otherwise you win a Pyrrhic victory, or get killed along the way.
If you succeed with one product, you’ve achieved something far more precious than that one product: an organization that has learned-how-to-learn the scaling challenge for a class of processes. The next time around, you can use your past (i.e., “experience curves” — now you know why they are called that) to learn faster, better.
Learning in Scoping
The key to economies of scope is transaction-cost learning. Walmart for example, can take advantage of the fact that whether you are buying a can of beans or a new toaster, the process of finding it in a large store, putting it into a shopping cart, and checking out, is the same. On the other end, buying a large catalog of things in bulk from China involves similar procurement process skills.
Economies of scope therefore revolve around questions like:
- What is the right level of diversification for our product line?
- What new market needs can we address with products/services we already make?
- Is the industry naturally horizontally or vertically integrated? Can we force it to flip orientation?
- What is the right way to bundle and price related products?
- Should we cover the entire range of price/performance possibilities for this product category? Are there customers with full-scope needs who would actually want one-stop shopping?
- Which customers are we under-serving? Which ones are we over-serving?
- What should we build? What should we buy?
- What knowledge should we learn and store? What knowledge should we rent?
Each of these questions is driven by a market structure concern that can be phrased as “should this be traded via the marketplace or managed through internal cost accounting?” The answers have some engineering implications, but they are marketing questions first. They help determine the operational shape, size and boundaries of the firm relative to its financial size.
The last question in particular reveals a relationship with scaling. Scoping decisions lead to scaling commitments.
Scope learning proceeds much more slowly than scale learning. While every instance of a production or service process is a learning data point, scope changes are much rarer, so the data accumulates more slowly. Events like mergers and acquisitions, changing suppliers, trying out a new outsourcing pattern, building out a new channel, all drive scope learning. The most important kind of learning event for scope learning is fighting off a new competitor and surviving.
Unlike scale learning, which can be reasonably be expected to plateau into an efficient state that will then deliver high-margin revenues for a period, scope learning may never plateau at all.
You may never get to a point where you can claim you have right-sized and right-shaped the business, but you have to keep trying. In fact, managing the ongoing scope-learning process is the essential activity in business strategy. If you ever think you’ve right-sized/right-shaped for the steady state, that’s when you are most vulnerable to attacks.
The Historical Scale/Scope Asymmetry
Economies of scope were only recognized and studied in the 1970s. By contrast, economies of scale have been recognized since at least Adam Smith in the 1770s. Why the two-century gap? And why the continued obscurity of scoping dynamics relative to scaling dynamics?
There are three basic reasons.
Pristine vs. Competitive Market Formation
First, the idea of economies of scale applies to pristine markets (the first successful company or companies in a new market, that is/are able to form on virgin territory). Economies of scope, by contrast, primarily exist in competitive markets: markets where incumbents exist and evolution is driven by the entrance and exit of a changing cast of players.
I am borrowing the terms pristine and competitive from political science, where they refer to two types of state formation. The result of this difference is that economies of scope emerged as a factor in the business landscape only after a few industrial sectors had matured enough to enter the competitive phase in their evolution.
Priorities in Mainstream Economics
Second economies of scope are fundamentally about transaction costs and the theory of firms — Coase stuff — and as such, is not part of mainstream economics education. If you’re like 80% of people who endured an Economics 101 course, chances are you only encountered Keynes. If you were lucky, you also encountered Friedman. If you were really lucky, you also encountered the Austrian school. The study of larger actors than individuals or small firms in economics isn’t exactly heterodox, but it is not exactly wildly popular either. Studying large state actors leads to developmental economics, comparative advantage and so forth (the original subject matter when the field was still called “Political Economy”). Political science education tends to pay some attention to this.
But larger firms with significant gravitational fields? Forget it. The subject falls right through the crack between management science and economics. You may have formed some loose ideas about how dynamics at the firm level work, through exposure to Joseph Schumpeter’s “creative destruction” phrase, but chances are, you have no real idea about how the process actually works, how firm size distributions shift and morph, and how an economic landscape gradually settles into a relatively mature market structure.
So those are the two abstract reasons economies of scope aren’t as well appreciated: the pristine/competitive distinction and the relative obscurity of firm-based economic thinking. The third reason is a person rather than an abstraction.
Michael Porter
It is rare for a single individual to have as much historical influence as Michael Porter has, but I believe — and this is going to be a controversial claim — that by providing a rudimentary zeroth-order model (a largely ahistorical and structuralist model) for analyzing economies of scope, Porter helped prematurely close a line of investigation that required far more attention than it has received. So the study of economies of scope stalled by the mid-eighties.
It is sort of unfair to blame him though. Forces beyond him conspired to basically shut down a fertile direction of inquiry, and turned his promising early models into a set of static truths, around which a variety of business cargo cults formed. More on Porter later.
Scale First or Scope First?
Scale and scope have a chicken-and-egg relationship. So it is worth investigating their interplay to some extent.
In a pristine market environment, you typically scale first, in whatever scaling direction you discover, and start discretionary scoping only when serious competition emerges. Initial scope in a pristine market is determined mostly by what you are forced to create in order to exist.
- Sears had to develop and scale the catalog model to establish modern retailing.
- Amazon, Google and other Web majors had to invent large-scale data-center technologies. They didn’t have a choice about whether or not to put those activities in scope because there wasn’t anyone else to do it.
In a competitive market environment by contrast, you always scope first, and the task is very non-trivial. Your initial scoping is in fact your entry strategy, and it is structurally similar to an exploit in hacking computer systems.
Some of the elements of the initial scoping are widely recognized. You have to think about whether to develop your own data-center technology or use Amazon services, a choice Amazon itself did not have to make. A century ago, product makers had to decide whether to sell through the new supermarkets and department stores, or run their own little stores.
Other elements of initial scoping are much more obscure, and only recognized in hindsight. The successes of Dropbox and the iPad reflect the workings of scoping decisions that were very obscure before the fact.
If you manage to create an initial scoping exploit and break through barriers of entry created by incumbents, you can then scale in a very selective way. Indeed, you must scale to survive. If you are not scaling along any dimension (alone, or along with an aggregate of small players that adopt the same strategy, such as bloggers versus old media), you aren’t actually playing in the market. The time pressure exists for the same reason as it does in hacking: you have a limited period of time before the intrusion is detected and lockdown processes kick in. You have to steal what you can and hope to get away with enough to earn a bargaining position.
Finding an initial scope that provides access to a vector along which to scale, that incumbents have failed to recognize, or have cooperatively ignored, is the definition of breaking into an existing market during the competitive market formation phase.
Once you do, you’ve caught a tiger by the tail. You have to scale as fast and as effectively as you can, usually rescoping as smoothly and quickly as possible along the way, shaping the clay as fast as dump it onto the table. Speed is of the essence here, because it is your only hope of growing to a sufficient size, and accumulating enough experience-curve learning before the incumbents react and get ready to contest your position.
This idea is often misunderstood. Iterating fast before you find a scoping exploit to break into a market is useful, but not essential. If you have a day job, or other means to stay in bootstrapping mode for a long time, you can go as slowly as you like. But once you find a scoping exploit, the clock starts ticking immediately. That’s when operating at a much faster tempo than the opponent helps. The more you can get done while the defenses are down, the better positioned you will be when hostilities begin in earnest.
How big do you need to grow in your limited window of opportunity before the incumbents launch a credible counter-attack? Our list from earlier provides the answer: you need to grow to a gravitationally significant size. Big enough to shape and warp the environment around you, because ultimately, it is the environment (smaller economic actors who, in aggregate, form the crowd that determines destinies) that decides whether you get to live.
To get the environment to keep you alive, you need to be big enough to influence thinking in a specific way: you must appear to be a part of their landscape rather than a part of the mobile population of small actors navigating the landscape.
In the eyes of smaller actors, you must transform from being a peer to being part of the terrain, too big to be wiped out in terrain shifts.
Right-Sizing, Right-Shaping
Since most industrial era sectors (defined as those that do not possess any economies of variety) are now mature, there are very few pristine markets. Almost all markets are competitive. This means, the lifecycle of a business is as follows:
- Becoming a Contender: An initial scoping exploit that we typically call the entry strategy. Success in initial scoping leads to the discovery of an unknown or dormant scaling vector. This is almost entirely a marketing challenge.
- Earning a Title Shot: A rapid scaling phase to get to gravitational effects scale before the incumbents can react, accompanied by rapid adjustments in scoping, to get to a “fighting shape and size.” This is not the final right-size/right-scale personality of the business, but one that is capable of winning the fight that has been set in motion. Here tempo is everything. You have to move as fast as you can.
- Title Fight: A wartime phase, when the incumbents finally recover from the effects of surprise and launch a counterattack to preserve the existing market structure (usually marked by a technological detente, which is why the newcomer can suddenly scale before the incumbents realize what is going on). This sets up the title fight: the newcomer through the undetected exploit and scaling phase, has set up a positional advantage, despite being smaller (generally), and is set for the tougher melee phase, after the surprise has been milked (see my post positioning moves vs. melee moves for more on this). This is a two-front war, since the newcomer must continue to learn along the scaling vector to preserve the position won.
- A New Champion (Possibly): The challenger either wins the title fight or loses. In either case, there is a new market structure, and a new order. And as I argued in my previous realtechnik post (linked above), some incumbents may exit. Here, as in guerrilla warfare, the newcomer wins if he survives the fight with enough left over to stay in the market. Beating the incumbent is not necessary.
- Right-Sizing, Right-Shaping: The title fight leads to a new detente, and a period during which the newly enthroned (or admitted) challenger has a chance to grow and mature in relative peace to the right size and shape. Incumbents have accepted the presence of the newcomer, resign themselves to making enough room for it, and retreat to lick their wounds in peace as well.
- A New Contender Emerges: The old contender is now one of the incumbents. Like everyone else, it is caught napping when another incumbent discovers a new scaling vector, or reopens hostilities along a detente frontier that was declared closed in the last war.
A key characteristic of this lifecycle is this: businesses almost never die of old age; they die of fatal war wounds in periodic wars.
Drucker, Porter and Grabowski Explained
The distinction between scale and scope maps very clearly to the distinction between engineering and marketing. In the days before computer-aided engineering tools, you could make a distinction between engineering and innovation aspects of technology. Today, they coincide.
Economies of scale and scope help explain two familiar, and one not-so-familiar idea in business.
The first is Drucker’s idea that there are only two basic functions in a business: innovation and marketing. This can be restated by substituting the word engineering for innovation today. The bulk of engineering in this sense actually revolves around scaling, not invention. We can therefore restate Drucker as follows: there are two, and only two economies in business: economies of scale and economies of scope.
The second is Porter’s famous five forces model of competition (and various associated constructs such as value-chain analysis). Scale and scope explain both its strengths (it gets forces and masses and “pieces” in play right, as well as the basic rules of engagement in competition), and its weaknesses (it either gets the evolutionary arguments and explanations for the dynamics of the warfare mostly wrong, or ignores them). Porter’s models are somewhat helpful in explaining relatively pristine markets where the cycle of war and peace has not yet heated up and positioning is an occasional, rather than continuous exercise. Once the cycle accelerates beyond a point, the models become too brittle. You need a more dynamic model that accounts for the fact that economies of scope are a learning process, and therefore one driven by human factors rather than structural ones.
The third idea — this is the obscure one — is the Grabowski Ratio, which has strongly influenced my thinking. If you’ve been reading this blog for a while, you’ve encountered the idea, but if not, it is probably new to you. The ratio captures the idea that the success or failure of a new product or service depends on the ratio of marketing spend to engineering spend. The ideal ratio (empirically discovered) seems to be around 1. The scale/scope model helps explain why: in a pristine market environment, scope is a forced variable that is relatively simple to manage (with little scope learning involved), while scaling must be actively managed through learning. Engineering spend can outpace marketing spend in pristine markets without hurting the business. But once a competitive market forms, initial scoping is non-trivial (since it is basically “exploit design”), and once the game begins, both scaling and scoping must be fluid and agile. So both activities will require comparable amounts of funding.
If all this is too complex, just keep a simple idea in mind: scaling is engineering, scoping is marketing, both are types of learning, you have to do both to survive in competitive markets, which are the only kind around today.
The Inevitable Biological Metaphor
Scale and scope are two dimensions of growth. In the process of biological ontogeny, a single fertilized egg grows to a full organism along both dimensions.
Scale is a matter of replicating the same kind of cell. The ecological niche of an organism determines its overall mode of survival, which in turn determines how big it needs to be, and how many cells it needs.
Scope is the fixed variety that emerges in the collection of cells as it expands to cover the structural needs corresponding to the functional and behavioral range of the grown organisms. On Galapagos, cormorants lost their wings, for example, a case of scope evolution.
Both scale and scope in ontogeny can be regarded as the effects of learning that actually happens at the genetic level, but are encoded into the phenotype and tested through survival of individuals. So skin cells represent learning about how to optimally design the organism’s physical boundary. The size of a whale represents learning about how big you should be if you feed on plankton and live in a certain gravitational environment.
And so organisms grow in size and scope until they become complete: capable of autonomous survival in their environment (the “autonomy” is of course defined relative to the sociability of the species). Older species that are the first to colonize an environment primarily represent engineering learning about that environment. Species that enter environments already colonized by competitors, must learn both the physical and biological environments around them.
Fixed scale-and-scope designs are only enough for survival in a single ecological niche during a given, relatively stable time epoch with a stable detente among species in that environment. Throw domestic cats onto isolated Pacific islands and all hell can break loose. Put Japanese culinary preferences together with industrial whaling technologies and size is no defense for a whale. Put traditional Chinese medicine together with guns and poor people living near national parks, and tigers, rhinos and elephants turn into pills and potions.
More robust sorts of survivability require learning processes that have greater capacity and speed and crucially, are more open-ended, and don’t stop during detente periods. The two extremes in biology are bacteria and primates. Bacteria learn faster by keeping their scale and scope very limited and speeding up the genetic process. The represent variety in the favela sense that I referred to in the beginning. Primates also learn faster by keeping physical scale and scope relatively fixed, and moving learning to more programmable (and more sociable) brains. More agile firmware versus more agile software. They represent the true economies of variety I am trying to understand right now.
In between, you have animals with the bulk of their learning in hardware, through high scale and scope — a lot of fixed and embodied intelligence — but sharply circumscribed survivability. Small brains, slow genetic mutation rates, complicated, specialized and big bodies. The problem here isn’t that hardware intelligence isn’t adaptable, but it is only adaptable within the narrow range of possibilities that are explored before the ecosystem niche stabilized into a long detente. As a first approximation, we can say that hardware learning is bounded in both total capacity and closed in the sense that it can only learn certain fixed kinds of things.
Economies of both scale and scope learning represent hardware learning varieties that must stop once the organization reaches a certain condition we can call “adulthood.” Species that cannot do either firmware or software learning don’t survive very well when there is a lot of environmental variability, either through movement into new environments, or changes in their existing environments, or incursion of new species. Firmware and software learning in biology are the (mixed) metaphors for economies of variety that I am exploring.
If you want to get a head start, try the paper Environmental Hypotheses of Hominin Evolution by Richard Potts, Yearbook of Physical Anthropology 41:93–136 (1998). Google should help you find you a bootleg PDF.
The Boyd Connection
This post has been brewing for a while, but what finally made the ideas come together was discussions at the Boyd and Beyond conference at Quantico (the headquarters of the US Marine Corps), which I just attended (and spoke at — one of the most fun talks I’ve done in recent memory).
For readers familiar with Boydian/OODA thinking, many of the elements of the broad argument in this post can be restated in terms of the basic ideas in Boydian strategic thinking, in particular Fingerspitzengefühl and Schwerpunkt. The entire model of how to operate in competitive (as opposed to pristine) markets can be summarized as “get inside the tempo of the market and then ratchet up the tempo to compound your gains in the window of time before counter-reactions form”
Pristine market creation can be equivalently thought of as “getting inside the tempo of nature.” I talked about these ideas at the conference, but since it was a no-slides conference, I’ll try to reconstruct the talk in the form of slides and post shortly (probably on the Tempo blog, not here).
Within the Boydian world, I want to acknowledge Chet Richards and Ho-Sheng Hsiao. Both provided extremely helpful pieces of the arguments I needed for this post. I also plan to blame them if these ideas don’t work out.
Economies of Variety: Preview
I am still working out the economies of variety idea, but here is a quick preview.
If I am right that there is such a thing, Drucker’s idea that there are “two, and only two” essential functions in business will have to be revised to “three, and only three,” the third one being a business function (like engineering and marketing) that maps directly to the type of learning involved in economies of variety.
Amazon’s recommendation learning models are an early example.
Whatever this third function, it will be heavily dependent on technology: machine learning and data technology in particular. But those will be necessary rather than sufficient elements (just as interchangeable-parts manufacturing is necessary, but not sufficient, for economies of scale to exist). Just as the prize, for winning economies of scale is a highly favorable amortization-of-fixed-costs equation, and the prize for winning economies of scope is a highly favorable pattern of transaction costs, the prize for winning economies of variety will be a highly efficient learning capability that will result in a sort of “dynamic Coasean” firm that will have the equivalent of a primate brain.
Grabowski’s model will need to be turned into a three-way ratio. Instead of M/E, we’ll have M:E:?
The Boydian-dynamic version of Porter’s models will need to be evolved to reflect a potential three-front battlefield in business (and war, for that matter). The as-yet-unlabeled ? third function of business will provide a capacity for what Boyd called “fast transients” (today, some businesses exhibit this capacity occasionally, in specific moves, but no business has a capacity for sustained fast transients).
One of the things holding me back from finishing the model properly is the lack of sufficient examples to think about. So if you know of a company that is navigating a three-way M:E:? challenge, and wants to help me work this stuff out, send them this post. I’ll offer some sort of discount on my consulting services in exchange for the opportunity to learn about their business and analyze it.
Parts of this post were also strongly influenced by my ongoing research for the Leading Edge Forum (a division of CSC) on the “Future of Data.” Though I didn’t talk about the role of data in this model, the catchphrase of the Big Data movement, “Volume, Velocity, Variety”, should give you an idea about why there is a connection here.
About Venkatesh Rao
Great article!
If I read it correctly (and am not injecting too much bias from my own background), you might find it helpful to look at some PE firms – especially those known for cutting and stitching together their acquisitions (e.g. Cerberus), or with large venture arms (e.g. Apax), rather than those only engaged in pure financial restructuring and resale. Identifying the correct variety of their holdings (for a given period) is, after all, how they earn their profits.
I basically know nothing about the PE industry, so I’ll have to take your comment at face value. Sounds plausible. It’s a complex meta-example though, since PE is basically firms that trade in other firms.
Fascinating and very rich posting. Would love to have a discussion with you around both the Austrian school and potential other forms of economy of scope that are not (at least explicitly) covered in your exploration above.
Definitely, let’s plan on lunch or dinner soon. Should be in and out of the Bay Area over the next few weeks.
Very good read. Very enjoyable. I especially loved the reference to Econ 101 and the Austrian school. You talk of Dropbox and iPad having scoping questions as the came to be. A “for instance…” there would have been helpful.
Also, I know you’re not writing about economies of variety yet, but you hint that you’re going to talk about superlinearity. I associate that with loosely coupled, largely decentralized organizations like cities not hierarchical organizations like Amazon, thanks for Geoffrey West, so I’ll be interested to see how you develop that.
West’s model is descriptive, not analytic. I don’t think organizational structures (“loosely coupled, largely decentralized”) have as much to do with it as certain very specific patterns of informational openness and a porous boundary. The “loosely coupled, largely decentralized” idea by the way, is decidedly NOT true of thriving urban areas. Usually that’s just the top few visible social layers on top of a couple of layers of tightly coupled and quite centralized hierarchical systems. Without this core of tightness (John Hagel’s “pull platform” infrastructure piece), the informational openness that makes for superlinearity would tear the system apart.
Viewed this way, there is nothing particularly special about cities. I think the same architecture can be achieved for corporations. In fact, I think Amazon has already pulled it off. They have other brewing problems, but they are more “city” than “20th century corporation” already.
Instant classic rivaling and out-scoping (yes!) The Gervais Principle.
Considering you once told me I’d peaked with GP, this is reassuring.
The applicabilty of Boyd to an overly beaurocratic military establishment is understandable.
The applicablity of Boyd to German operational concepts somewhat alludes me. While it is generally true that the Germans tried to move quickly and keep their opponents off balance, what they actually did is not particularly well described by Boyd and his loops.
Russell: Google “Patterns of Conflict” for a (rather cryptic and long) analysis in slide-deck form, of Blitzkrieg by Boyd. He was very heavily influenced by German WW II thinking. The influence was second only to Sun Tzu. The OODA loop is just the tip of the iceberg of Boyd’s thinking on German models. Chet Richards’ “Certain to Win” has a lot more (and less cryptic) detail based on what Boyd actually said during his briefings.
I just obtained audio recordings of the Patterns of Conflict briefing, which I plan to listen to. Eight hours, so it is going to be a bit of a slog.
O.K. I have looked at it.
It is non-sensical.
The whole idea of the decision loop is rather linear (in a time sense) in the first place. In a complex situation, where you are being hit by information/opponents from all angles, there is no single “opponent” whose loop you can get into. It is exactly the type of thinking that you would expect to get from a fighter pilot though. It reminds me of how the German fighter pilots in North Africa shot all their opponents fighters to pieces, and lost the air war – in part because they didn’t like attacking bombers. A bomber, even a two engine one, had lots of eyes and guns, and other bombers flying close by. You couldn’t assure success with tricky tactics. You just had to go in and hammer it out and hope for the best. Which is a lot what real life is like.
I would also reccomend you stop using the term blitzkrieg. It is not actually how the Germans viewed their opperational concepts, but a term that was promoted by the media.
The Germans excelled at tactical doctrine, and the heavy use of radios to enable closer tactical/operational cooperation. Their fast dash system tended to fall apart when they had to push beyond 300km, or if defensive weapon densities became too great. Most everything else they had, their opponents had as well. They were the best army at the start of World War 2, got a free practise session in Poland, and then pretty much stayed ahead of the training curve until somewhere around 1943.
It should also be noted that Boyd’s idea about using retreat as a disruptive mechanism, was used by the Germans as well. Moltke the Eldar was fanatical about the battle of Cannae. The Germans kept their AT guns close behind the front wave of attack, so that the counter attacking enemies initial successes would land on a wall of steel. If you look at how they set up their defensive postions, you can also see that they developed a system of non-linear defense through chaos. A system of defense that they developed in WW 1.
I entirely disagree, but this is not the time and place for that conversation, and I am not the right person to make the case. Suffice it to say that I am skeptical you could have grokked PoC that quickly. The fact that you find it nonsensical while a lot of very smart people have spent years understanding, successfully applying and teaching it, suggests you didn’t get it. I didn’t start getting it until the third or fourth reading.
Part of the reason may be that you’ve already made up your mind about German WW II doctrine. I’ve heard your reading of German doctrine before and I don’t buy it basically. Boyd’s reading made a lot more sense to me.
I’ve personally found PoC an endlessly rewarding document to study, but it does require participating in the community of people who were at the live briefings etc. and able to clarify cryptic bits (Boyd was known to be deliberately cryptic — remember, these briefings were developed to influence Pentagon conversations, one general at a time, not for classroom exposition. This intent was quite spectacularly achieved).
For a different and more detailed takedown, you could look at Jim Storr’s The Human Face of War p 12 – 14.
I bought book you mentioned and checked the pages you referenced, Storr seems to have a fundamental misunderstanding of what Boyd’s OODA loop is. Storr’s concept seems to be that the OODA loop is strictly cycling through Observe->Orient->Decide->Act in that order. Boyd’s concept incorporated both feedback and feedforward between each of the Observe/Orient/Decide/Act stages. Based on those pages you referenced, Storr’s ideas actually seem quite similar to Boyd’s. In fact, the Gulf War example he uses was architected by Boyd. Chet Richards and others have leveled these same criticisms at the “OODA-lite” that Storr describes. FWIW, Storr doesn’t cite Boyd/Richards/Osinga/etc, Lind seems to be his primary source on OODA.
I like a lot of Storr’s ideas, but I would be the first to admit, he is not always as focused in his concepts as you would like. He places a lot of emphasis on shock and surprise (which Boyds ideas would not preclude) but the writing on them is a bit diffused. I suspect part of the problem is working in an area that has very little quantitative research, and where there are studies, their scopes don’t mesh well.
A lot of his criticism (of the British) seems to be the fuzziness to which their (apparently) Boyd inspired doctrine, and he comments in less detail on American doctrine. Of course Boyd cannot be faulted if his doctrine is adopted in a fuzzy manner – unless their is some fuzziness to parts of it. Of course Clauswitz has some rather fuzzy concepts at times, and their are continual arguments of how to exactly translate some of the German Army concepts. John A. English in his On Infantry does a good job of showing the German tactical principals.
Not trying to get you to buy more books! Robert M. Citino has written a number of interesting books on the development of German war making (a review of one is here: http://www.army.forces.gc.ca/caj/documents/vol_05/iss_1/CAJ_vol5.1_18_e.pdf ) although he doesn’t make a big issue of it, it is Citino (somewhere) who I recall noting the German incorporation of wireless into their mechanized tactics, Colonel T.N. Dupuy’s A Genius For War is the earlies public English language writing that I recall that specifically singled out the General German Staff as the reason for the German successes. A book I have but I have not read yet, is Mary R. Habeck’s Storm of Steel which cover both German and Soviet armor doctrine in the lead up to WW2. You may have sited these works previously, but I didn’t see them in your book Tempo – which of course makes sense as it is not a book about German doctrine.
I believe it is Martin Creveld who I first saw noting that the German system could go about 300km before it ran out of steam. Very effective in small theater warfare. I think it was Citino who noted that the Germans had a tendency to push so hard, that they litteraly ran themselves to exhaustion, leaving little in the tank for second efforts. Given that Germany was a rather midling great power, I would concede that to expect much better results when combined against such a self-inflicted array of forces. Citino compares Israel operational concepts to German ones. He notes that both countries had many hostile large neighbors, and that neither could afford lengthy wars, as there was always the likelyhood of intervention coming from the opposite direction. Thus both countries tended to have very aggressive doctrine that pushed the decision making (starting at a time when the fastest message service was on horseback) to the lowest level.
It is very difficult to really “know” what the Germans were all about. We don’t live in their culture, and don’t make all the same assumptions that they would. A lot of what I have seen actually comes from reading the few good accounts that detail their battlefield activities.
As a for instance. The Germans developed a very effective method of defense using heavy machine guns in 1915. The retained these methods even when they went to a less linear defense. I have never seen anything in American military writing that coherently addresses what it is that the Germans were doing. But the Germans like to scale and reuse everything. As best I can tell they used some of the same methods with their anti-tank guns when they could. You would think that even if the British and Americans didn’t feel a need for these methods themselves, they would do more to make their details understood.
From what I am seeing of Boyd. He picked up some interesting pieces, but it doesn’t seem like he is doing what he is doing for the same reasons that the Germans were doing what they were doing.
Of course, if it works, it won’t really matter. And as we are not Germans setting out to fight a new great war, possibly Boyd is the way to go. But Storr’s comments about the fuzziness of the British/American thinking argues that at least the application (as you note, he does not source Boyd) is lacking.
When you suggest resources are best allocated roughly evenly between engineering and marketing, I assume that sales is included in the marketing bucket? Does startup fundraising fall in marketing, or is it its own separate activity? Would you speculate on how you would account for sales/marketing efforts done by channel partners (when you have them) — is that aggregate marketing spending counting towards your total marketing expenditure, or is it off those books?
Venkat,
another really amazing article. The argument that we are making in B = mc2, is that in the past, companies were limited by the cost of acquisition of new customers and the ability to create (guess?) quickly enough and cheaply enough the products and services that they would want to buy.
“The Platform”, as we call it, has already changed everything. When computing became mobile, it opened a market of 1+ billion customers and growing who are just a thumb away from your business, customers that tell you what they are doing and even what they want or who they are related to. These customers are also providing ample feedback about what they bought. At the same time, the cost of developing solutions is going down rapidly (due to many factors that I can detail, Cloud for instance).
When you can develop products and services more rapidly and acquire customers in a much cheaper way, the successful companies will be the ones that can manage to work across multiple dimensions (products, geographies, customer segments, technologies, form factors …), I think this is what you call the variety of scope.
In a multidimensional world, nothing is context any longer, everything is core (think of why Apple built Cards for example? every MBA, every business strategy framework would have told them not to), yet when you sell to a market that is 1+ Billion users, you don’t get a competitive advantage by standardizing or commoditizing your processes, quite the contrary, delivering solutions across multiple dimensions, in the variety of scope becomes a key strategic differentiator.
To answer your question: “Whatever this third function, it will be heavily dependent on technology: machine learning and data technology in particular. “, I would actually beg to disagree. You are on the right track when you talk about machine learning or big data, because both of these technologies give you some kind of visibility, but IMHO, you need human visibility, you need to do like Apple did and pick across hundreds of features and processes, across dozens of dimensions the one solution that will give you a decisive advantage. This is why we work on improving human visibility, there is no amount of big data or machine learning that could have given Apple a hint about building Cards.
Could an academic chime in on this?
A speed-read says this is pseudo-intellectual at best, and not new, academically.
For example, there are already papers on the complexity/diversity/variety or whatever you want to call it in economies, and even analysis of their mathematical properties.
I’m fairly sure all three “scales” already have names and people studying their properties.
Enjoyed the article. Wanted to emphasize that you’ve conflated two distinct concepts, economies of scale and learning. Neither is necessary for the other. In empirical studies care must be taken that you are not attributed to one of them the effects of the other. See,
http://en.wikipedia.org/wiki/Experience_curve_effects#Criticisms
If I read the criticism in the link correctly, sounds like they *should* be conflated! Agreed that experience curves aren’t always easy to compute, for precisely the complexity of learning.
I agree they can be separate and exist independently, but only in relatively simple cases. In complex cases, they are two sides of the same coin, which is the point of the article. To scale, you have to repeat a large number of times, which causes problems that you have to learn to solve. Conversely, most of the important learning in a growing organization concerns scaling.
Venkat,
Interesting article with lot of food for thought. It seems your three options map to the three Porter generic strategies. More precisely the Treacy/Wiersama remapping into their “value disciplines” of operational excellence (“economies of scale”), production leadership (“economies of scope”) and customer intimacy (“economies of variety”). With big data capturing customer needs at a very minute level, economies of variety seem to be possible at a level not seen before. Not sure if you agree with that.
However, the bigger issue I have with this classification is it seems very inert and doesn’t account for agency or the initial discovery phase of mapping out the approach (i.e, which of the three makes sense). It is hard to imagine someone setting out to build a business by focusing on any of these three focus areas as a starting point with exploration.
Also it is not clear if the strategies are so cleanly independent. To take a well know example, is the introduction of iPad of 2010 a economies of scale play (supply chain prowess)? or a economies of scope play (marketing prowess)? or a economies of variety play (extensive customer insight due to iPod/iPhone usage)? As it stands, it seems these “strategies” are a end result of a activities undertaken earlier and not a guide for doing things per se.
You nailed the issues that arise from a Porter framework, and also exactly where I am going with this. I am actually using the value disciplines model as the starting point for developing the right kind of notion of integrated agility required for my data project I mentioned. It’s a good starting point, but dynamism has to be added.
I think scale, scope and variety are the rows of a matrix where the columns are the value disciplines. You need to go from a diagonal matrix to one with off-diagonal terms, and use the thing in a dynamic model of a company and it’s ecosystem.
Can’t say more here since the work is proprietary for my clients at CSC, but I’ll be exploring this in more general terms in a future post.
possible example: Paypal?
Just realized that this theory explains YC Combinator perfectly as a high-end recruiting firm.
They buy early- or midway-through-process-scaling-learning teams cheaply, accelerate and shape the learning, and then sell them at a premium to the likes of Facebook.
Venkat,
Great piece, this is my first exposure to your blog(s), but definitely not my last.
I’m particularly struck by this:
“I don’t think organizational structures (“loosely coupled, largely decentralized”) have as much to do with it as certain very specific patterns of informational openness and a porous boundary. ”
The reason that I’m struck by this, is that it seems to speak directly to some of my recent work with a colleague… I think that you’d be interested to have a look, as it may help in the conception of your “Third pillar”. Perhaps not surprisingly, we also reference Coase and Hayek.
See here if your’e interested:
http://arxiv.org/abs/1210.2013
If you have any thoughts in this regard, just shoot me an email.
Best,
JP
A really great article. I very much enjoyed reading. This is probably worth at least two years of business school ;-)
Anyway, in terms of “sufficient examples to think about” the mysterious “third function”, why not look at car manufacturing ecosystems and the ways in which this industry learns, collaborates and organizes itself, at least along some shared issues of scaling? I am thinking for example of how integrated companies like Bosch are and how greatly many different car firms co-depend on their inputs and innovations and vice verca. Interestingly, in the automobile industry, just as with Amazon, there seems to be loads of ” informational openness and a porous boundary” (aka collaboration) in terms of the scaling side of things, while the same firms are fiercely competing when it comes to scoping… This might be a great limitation actually.
When thinking about scoping collaborations, examples become much more obscure . I don’t know, but it might be interesting to compare organisations that scale well (e.g. McDonalds) to organization that seem to “spread” well instead (e.g. Chinese restaurants). As Charles Leadbeater noted, there is something strangely peculiar going on, in that you can enter a Chinese restaurant anywhere on the planet and have a consistent experience of “being in a Chinese restaurant” without the organizational and management efforts that McDonalds has to put in play aiming the same sort of outcome. What these spreading models seem to be doing is organizing without an organization, which is hinting towards forms of organizational learning and knowledge sharing that take place in more tacit ways… (Maybe there is something Jungian, as in collective unconscious, at play here). What ever these ways are, they also seem to provide some form of (cities-like?) resilience, given the fact that (at least to me personnally) it seems much more likely that the archetypical Chinese restaurant will still be around long after McDonalds closed shop…
Anyway, a great place to look, in order to explain spreading things like chineese restaurants might be in scociology. Rüdiger Mautz, for example, wrote some interesting stuff explaining decentral diffusion systems that drove/drive the use of on-shore wind-turbines and rooftop solar panels here in Germany, once the feed-in tarifs were in place. Mautz describes it a process of diffusing “Handlungswissen” (that is knowledge of how to do (and maybe scope?) things in order to reduce risk and uncertainty. Also, at the moment there is a new movement emerging here, where people attempt (an sometimes even succeed) in buying back municipal electricity grids from traditional utilities in order to run the show themselves. These grid takeovers are spreading in terms of the underling diffusion mechanism in ways that seem quite similar to what was/is going on with solar panels and wind turbines since the 1990s: It all started with a successful and hugely inspiring takover of a grid in a small town in the Black Forrest that led to the formation of a company called “Elektrizitätswerke Schönau” – or EWS – and which is now replicated or (re-)invented by groups throughout other locales across the county. (If you are interested, I could hook you up with the people who are trying to pull this off in Berlin right now.)
The one social theory, however, that might help make sense of what’s happening here could be Practice Theory, which argues that any Social Practice is a process of integration that results in a structured arrangement (i.e. resulting in a practice that exists as a recognisable entity) and that the elements that need be integrated consist of three things only: (1) material (as in stuff or infrastructure), (2)image (as in the socio-cultural meaning, context and stories explaining why people do something) and (3)skill (as in procedure, knowlege and associated forms of learning). I belive that it might be possible to work up the constituntional elements of practice to blend in nicely with your three functions, whereby (1)material and stuff would exist more on the scaling side of things, (2) Images and stories would sit along scoping issues and (3) skill might help with dealing with phenomena of “institutional openness”. I don’t know…
Practice theory however, can be complex and counterintuitve at times. A great starting point, for me at least, was the article ‘Consumers, producers and practices: understanding the invention and reinvention of Nordic Walking’ by Elizabeth Shove and Mika Pantzar. Here, Shove and Pantzar suggest that new practices arise through the active and ongoing integration of images, artifacts and forms of competence: a process in which consumers and
producers are both involved(!). They also try to to explain the growing popularity of nordic walking – as a practice – in countries as varied as Japan, Norway and the USA –>that’s the diffusion part. Eventually they conclude that practices and associated cultures of consumption are always ‘home grown’: “Necessary and sometimes novel ingredients including images and artifacts may circulate widely but they are always pieced together in a manner that is informed by previous and related practice. What looks like the diffusion of Nordic Walking is therefore better understood as its successive but necessarily localised
(re)invention.” (there is also a video online that tries to explain practice theory in easy terms: http://www.lancs.ac.uk/staff/shove/lecture/filmedlecture.htm )
I am not sure if this is any helpful to you, but these are the lines of thought that were triggerd when reading your hugely inspiring piece. Thought that I might as well feed them back to you. Its not often that I write extended blog comments, though. Take this as compliment. I hope this can be of some help…
Lots of interesting things to explore here. The Chinese restaurant example is one of many such examples I think, where imitation/social proof drive a weak (rather than perfect) standardization.
The Nordic walking example and ‘Practice Theory’ are new to me, so thanks. Will think about those things. Useful ideas here.
You mention Coase, but don’t seem to mention Yochai Benkler. I wonder if his paper on Linux and the Nature of the Firm might illuminate some of the issues you’re exploring?
Hi Joel,
Definitely agree in regard to Benkler.
Not sure if you took a look at the link that I posted above, but we address his work, along with Coase and others too, in “The Theory of Crowd Capital”.
See here: http://arxiv.org/abs/1210.2013
I’d be interested to hear your thoughts in this regard!
Best,
JP
Lots of interesting ideas in here.
Well, that explains the Amazon P/E. But will they ever earn enough to justify it, time will say.
The gravitational effects generated if it becomes as big in revenue as would be justified by Amazon’s P/E … hooboy!
This might lead not merely to a third dimension, but a fourth as well, or perhaps this new dimension could be multidimensional. Geoffrey West found that many phenomena he studied ended up being five dimensional with three dimensions of space, one of time, and then a fractal dimension. The business of making use of accellerated tempo to establish operating scale and scope is somewhat like this in that the path taken must outspeed and outflank potential competitors rather like a space filling curve occupying space to deflect travel along a linear or smooth nonlinear path. At any point such a strategy appears to consist of operations that define competitive scope and scale, but it is actually the tempo and path taken that lead to operations being successful.
I finally got to reading this article in full, though I realized that this was (subconsciously) why I asked elsewhere the question about getting into nature’s OODA loop. This one thing came to mind though:
As a first approximation, we can say that hardware learning is bounded in both total capacity and closed in the sense that it can only learn certain fixed kinds of things.
I’m not so sure at this point about that. Jeremy Campbell, in Grammatical Man makes a strong case for evolution being a much more rapidly moving force than people normally think. I found the same sort of argument also existing in Stuart Kauffman’s At Home in the Universe. Both involve the idea of phylogeny having some sort of “grammar” that creates rapid leaps in evolution. I’m also wondering about the connection between this idea and black swans.
This also seems to hit at your question you left in Tempo about tactics and the Efficient-Market-Hypothesis. I also now realize that between markets and evolution, the OODA loop of “nature” may just as well be characterized as the OODA loop of entropy; such a thing does seem to exist, considering how history leaps and yet shows patterns.
Apple’s resurgence had more to do with content exclusivity (single songs) and distribution channel (iTunes) than manufacturing scale and scope. They executed flawlessly on the supply chain side (scale) and design aspects (scope) to solidify that initial lead early last decade.
Moving forward a decade, they have continued to execute on scale and scope at the expense of creating the next sustainable competitive advantage (content). Others have now caught up on scale and scope economies as it is was just a matter of time.
If I may offer a comment… In a conversation about this post at work today, we were wrestling with your remark that “similar processes are used to deliver a set of distinct products or services”, and the distinction you’re drawing to the “identical”-ness of the interchangeable components needed to scale. I think this distinction is quite important, and the subtlety of it is part of the overall subtlety of the distinction between economies of scope and scale (and hence, many people’s apparent trouble to distinguish properly between them — when I talk about “using the same DevOps and other approaches to deliver functionally different SaaS offerings into the insurance market”, someone will often describe that as “economies of scale”, causing me to do a facepalm…).
Anyway, in the course of our conversation, it became very apparent that this conversation about “identical” vs. “similar” and whatnot was a bit clumsy, and awkward.
And that’s when a lightbulb went on in my head, and I realized “Hey, you’ve been here before.”* Because I think that a much more precise set of words for this part of the conversation are “isomorphic” (for the interchangeable components), and “isofunctional” (for the similar processes). Interestingly, there appears to be a coincidental parallel here — the word “isomorphic” is well known, and defined: it means “you can transform one into the other without any information loss”, and it feels to me like a much more precise description of the test that components must pass in order to be interchangeable enough to be useful in solving scaling problems — “identical” is too literal, and makes semi-autistic engineers (like my colleagues) go off on irrelevant tangents.
But, in the same way that scope is far less well known than scale (here’s the strange parallel), “isofunctional” is a much more obscure word. Biologists use it to describe things that may or may not be isomorphic, but do exhibit similar behaviors. If you’ll allow me to continue these word games, and assert that “behaviors”, in the sense that biologists mean, and “processes” in the sense that the pointy-haired (and you) mean here, are isomorphic, then we can say that the line I cited from your post here could be changed to read “isofunctional processes are used to deliver a set of distinct products or services” — and it would be much more precise way of talking about this aspect of economies of scope.
I think.
So I thought I’d throw that all out for your consideration, knowing as I do how deeply you like to think about words.
* I first ran into the notion of “isofunctional” in reading “Gödel, Escher and Bach”, and I blogged about one manifestation of the importance of the distinction in IT here: http://www.jroller.com/MasterMark/entry/raic_what_s_that
Very interesting thoughts !
Now on the question of variety / scale / scope and the analogy with war, I have the following intuition :
The scale effect is analog to attrition. It has been achieved historically by armies through mass levies / conscription (S1 function – see * below) and logistics – mass production, standardization, mass transportation (S4 functions).
The scope effect is analog to operational coordination (essentially an S3 function). Think of the Blitzkrieg innovation : using radio communications as a platform to coordinate actions by front troops & tanks, artillery and air force (tactical).
Now could it be that variety is analog to the remaining function (S2) which is intelligence ? The association with learning seems to point in this direction.
What do you think of this intuition ? Valid or not ? Useful or just a waste of time ?
Cheers,
Gilles
* I refer to the classification of Staff functions used in many armies, based on the US model – I think. S1 = personnel, S2 = military intelligence, S3= operations (and training), S4=logistics (supply and transportation).