Sometimes, doing the right thing is just way too hard. So you have use the best approximate substitute available. When you can’t fly like a bird, you can aspire to be a frog that can jump really high, or a flying squirrel.
Decision-making is like that. There is, in my opinion, a “right way” to do decision-making in complex, dynamic environments (VUCA conditions — Volatility, Uncertainty, Complexity, Ambiguity), but most of the time, the right way is way too hard. So like most people, I use approximations tailored to current conditions (I am partial to the geeky joke that life isn’t just hard, it’s NP-hard).
To explain the right way and the approximate way, it helps to think in terms of high-speed maneuvering as a metaphor. Think of the dog-fighting in-an-asteroid-field scene in Star Wars. There are unpredictable moving obstacles and adversaries in the environment, and potential/kinetic energy considerations arising from the physics and energy levels of your own vehicle.
The “right” way to engage such a domain is with high situation awareness and calm mindfulness. Such a mental state allows you to maneuver smoothly and efficiently, with surgically precise moves that minimize entropy generation while achieving your objectives. This is the peak-flow-state, with your OODA-loop humming away at Enlightenment Level 42.
Unfortunately, if you’re like me, you’re in that state perhaps 1% of the time. What do you do at other times?
OODA for Zombies
The rest of the time, there is a good chance you are an over-caffeinated, under-exercised, manic-depressive, financially precarious, junk-food-eating zombie. And even if you’re not a total train-wreck, you may have over-reached so far beyond your capabilities that you might as well be.
Once such crutch is replacing the satisfying full-blown dogfight-in-an-asteroid-field metaphor with a simpler and less satisfying one: an auto racing metaphor that is locally and approximately correct a lot of the time.
The racing metaphor is simpler because it involves no moving dumb obstacles like asteroids, adversaries who maneuver in simplified ways, a constrained competition model, and a playing field that’s a closed circuit with a simple geometry.
The best-known use of the racing metaphor is the heuristic, overtake on the turns.
The reasoning behind this heuristic is that a turn, where potential and kinetic energy must be traded off in controlled ways to reorient a vehicle, is where skill and higher situation awareness can beat raw power. When power is roughly equal between competitors, skill differences at the turn are the only thing that can change the leader board.
A closed-circuit race decouples agility-first and energy-first epochs. If the course is simple but very long, you can even map planning/execution to turns and straights.
The Turn and the Straight
But I don’t hear much about the obvious companion heuristic (I had to make up a mnemonic phrase): overwhelm on the straight.
If you have a true power advantage, you have to use the relatively straight portions of the course to pull away from the rest of the field.
And you have to overwhelm, which means using your power advantage to really put a lot of distance between you and competitors. It is not enough to get just a little ahead. You have to get as far ahead as you can.
Why? Because under zombie, non-peak-flow operating conditions, a trade-off holds between power on the straight and agility in the turn. You are unlikely to be good at both.
To take a simple example, imagine that you’re racing with just one other adversary on a rounded-corner rectangular circuit. You gain 6 meters per straight due to greater power, and lose 5 meters per turn on every turn due to poorer turning skills. So your net advantage isn’t 24 meters/lap. It is 4 meters per lap, and you have to go all out to get it.
You cannot be lazy or hold back. Depending on your overall power reserves for the race, however, you can and should cleverly time when you choose to go full throttle.
The key to the timing there is the length of the race.
Short versus Long Races
The shorter the race, the greater the advantage for the competitor with better turn performance.
What’s more, when you’re neck-to-neck, the more agile competitor has an even higher advantage, because the more powerful adversary is likely even worse on a neck-to-neck turn than a solo turn. So the agile adversary will likely gain more per turn when the gap is smaller.
On the other hand, long races favor the more powerful players (so long as they stay focused on the race and don’t get distracted).
There are two reason for this. The first is decision fatigue.
Navigating a turn smoothly drains the executive function because it is more demanding. By contrast, powering through a straight requires far less executive function control. It is more a function of raw energy levels.
This means, the advantage of an agile adversary is going to slowly decay through multiple turns, as he/she gets mentally fatigued.
The second reason is local learning.
Situational learning is easier and cheaper than generalized learning. An agile competitor has a starting advantage based on generally superior turning skills that are most potent when the course is equally unknown to all. But on a specific simple course, such as our simple rounded rectangle, a power-competitor can quickly and cheaply learn the local turns and neutralize the advantage of the generally agile competitor.
Fast-Following as Co-Opted Agility
When the turns are a repeating pattern of the same four turns, rather than a sequence of unexpected turns, generalized agility quickly becomes useless. It is equivalent to a starting position advantage on a completely straight and long course: a finite advantage that just delays the inevitable against a more energetic racer.
This means the longer the race, the more you can rely on greater power. In our simple example, assume that the more powerful competitor always gains 6 meters per lap. But if the turn performance advantage of the agile adversary erodes to nothing after (say) 100 turns (25 laps), the race turns into a pure straight race with a 24 meters/lap advantage instead of a 4 meters/lap advantage.
Once you’ve fatigued the executive function of an adversary, and done your local learning, you can pull away steadily. In fact with a limited overall energy budget, it might be useful to stay slightly behind initially, as a racing strategy, making the turn-advantage competitor work harder for their turn gains, fatiguing their executive function more quickly while learning faster yourself.
So the presence of a more generally agile competitor means you can learn a localized (and therefore cheaper) version of their applied generalized tricks via imitation.
Once you’ve exhausted them, you can start developing your straight advantage and pull away. This strategy works even better if the adversary also has a lower total energy store. You simply wait for physical or mental exhaustion (whichever happens first) before you get started, just keeping up until then.
They waste their limited energy learning, you imitate and save your greater energy reserves for scaling and winning.
Many fast followers (such as Microsoft) seem to operate this way, effectively co-opting more agile competitors as unwitting scouts. This is the reason imitation is so much more successful than innovation, in terms of returns.
The racing metaphor actually understates the advantage of being a power player, because in many real-world situations, gains compound rather than accumulating linearly.
Imagine a weird sort of race where the more you are ahead, the more you can pull ahead. So instead of always gaining 6 meters/straight, you gain 6 meters/straight if you are 1 meter ahead, 12 meters/straight if you are 2 meters ahead, 24 meters/straight when you are 3 meters ahead.
You get the idea. You can really overwhelm on the straight in long races, and the longer the race, the more you become impossible to catch. It can be a crippling demotivator to watch an exponential breakaway from behind.
I’ve struggled to articulate this in previous posts, but I think I finally understand the phenomenon correctly. You can check out my old post on the subject if you want a more complete (if more confused) deep dive.
I am interested though, in whether there can be a similar advantage for the agile under sustained VUCA conditions.
Let’s call the corresponding idea of a runaway agility advantage an exponential turnaway.
I don’t think such an advantage can exist in a simple situation like racing on a known closed course.
But when we return to the full-blown metaphor of dogfighting in an asteroid field, it may be possible.
Here, exponential turnaway may not depend on compounding gains for the more agile competitor, but compounding losses for the less agile one.
Military strategists often talk about the idea of a “target-rich environment.” An asteroid field is sort of the flip-side of that idea: a “projectile rich environment” that disproportionately penalizes bigger players (size is generally correlated with power and lower agility).
Instead of a known sequence of turns coming up with known periodicity, the non-agile player faces an unpredictable environment of unknown maneuvering challenges.
From the point of view of how quickly decision fatigue sets in, I suspect, the less agile competitor will quickly get overwhelmed, with compounding errors, bad decisions leading to worse situations, and eventually, an impossible-problem situation.
Of course, the more agile competitor is also experiencing decision fatigue, but he/she doesn’t have to survive forever in the asteroid field. Just long enough for the competitor to crash and burn. At that point, he/she can exit the asteroid field, and recover in relative peace.
Exponential turnaway can particularly affect big companies that try to “fast follow” an entire swarm of little startups into a new market, instead of just a single agile scout. The leader emerges via a shakeout in the swarm rather than a powerful incumbent fast-following an initial leader. Neither Microsoft nor Google won the social network game. Facebook and LinkedIn did, via a shakeout in an initial field of many players (so it is inaccurate to say that Facebook was a “fast follower” to MySpace. It wasn’t. The sector wasn’t really viable at that point in the game, and Facebook did not really win by imitating MySpace the way Internet Explorer imitated Netscape).
In such a situation, even acquisition may be an impossible strategy. By the time a clear leader emerges via a shakeout, it may be too big for a wannabe “fast follower” to swallow. This is why, despite the generally greater returns for imitation over innovation, the innovation game is still worth playing, at least for the rare shakeout winner.
If you are a pioneering startup afraid of a big company fast-following you into markets you created, it might actually help you to help out other related startups a bit. They will tweak your formula and pursue it slightly differently, creating a shakeout game rather than a fast-follower game.
This is not about socialist cooperation, it is about creating VUCA for a more powerful adversary.
There is a Hunger Games joke in there somewhere (I just watched the first movie yesterday).
The Lean Startup types among you will have noticed some obvious connections to those ideas. In fact, this post partly grew out of my efforts to try and figure out what bothered me about the model.
It is by now widely recognized that the Lean Startup model can, in many ways, be regarded as an approximate version of the OODA loop. But the precise nature of the approximation involved has been eluding me.
The symptom of the simplification is that the Lean Startup replaces the core “get inside the tempo of your adversary” competition model of OODA with a “faster iteration” competition model.
Here “adversary” can stand for either a literal competitor, or a responsive market (i.e., an asteroid field that responds to your actions, such as customers with some control of their own decisions and learnable reactions to your moves).
I spent some time thinking about conditions under which the simplification is valid, and the best I’ve been able to come up with is this: “faster iteration” is a valid substitute for “inside the tempo” when for some reason, the learning per iteration has a fundamental rate limit.
What do I mean by this? Imagine a dumb startup and a smart startup trying the same first A/B test on identical minimum viable products (MVP). Both get the same results, but the latter learns a lot more, by extracting more intelligence, and is able to pick a second experiment (or pick a pivot direction) much better.
Under these conditions, iteration speed is not very relevant. By simply extracting more intelligence per iteration and pivoting smarter, the smart startup can gain on the dumb startup even if they are iterating slower. In optimization terms, the smart startup will be extracting the signal and climbing the right hill. The dumb startup will be thrashing around in response to the noise and failing to see the hill.
But if for some reason, the amount of net learning per iteration is rate-limited, an inefficient learner iterating faster can pull ahead.
This corresponds to conditions where there is almost no scope for generalization either for future decisions in the same local domain, or other similar domains. Or to use a generalized form of our race metaphor, everybody is equally bad at the turns, so the race goes to the better straight performer by default, no matter how short the straights or how long the race.
This is a world where agility is meaningless. There is no such “skill” in the picture. You have a pure straight race where everybody gets knocked down in a completely unpredictable way by nature every few dozen steps. So it all depends on getting back up faster and running as hard as you can till you are knocked down again.
So a lean startup mindset is good if your learning resembles memorizing the digits of a random number, with one digit per iteration. Being smarter doesn’t really help. Being faster does. Every bit learned is expensively bought via a knockdown.
So it is not surprising that the lean startup is most popular in a domain that is effectively close to random user behavior around Web technologies.
Any truly generalizable insight spreads quickly via imitation, since UX design IP is hard to protect. What remains is the random-bits bleeding edge of highly localized user behavior learning (localized down to the single click in a single fixed context, with an additional advantage for enterprise software, which is more localized than the consumer world). The faster you can accumulate these results, the faster you can move (a good sign is that people aren’t particularly good at predicting the results of non-trivial A/B tests).
A clear illustration of this principle is this well-known factoid about memory: chess grandmasters are no better than random people off the street when it comes to memorizing random board positions. But when the board positions are realistic and legal, they do far better.
For a startup, near-random domains make the lean model useful, especially for enterprise software where there is a lot of corporate arbitrariness to be learned for every individual customer. For big consumer web companies, this translates to massive amounts of continuous, almost automated testing.
But when there is more general domain structure, it pays to give up some iteration speed for smarter learning: going for the “inside the tempo” approach.
So paradoxically, if you think you have an agility advantage, it pays to not be at the bleeding edge where all learning is almost random. It pays to withdraw to a less random domain where your superior generalization skills are an advantage.
I am still not completely satisfied with this model, but it’s getting there. I’ve been working on these ideas for almost two years now, and they are finally getting precise enough that I could mathematically model them if I wanted to (starting with this metaphor of a race-to-asteroid-belt spectrum, and generalizing to exponential breakaway/turnaway type domains).
For now, what I’ve arrived at is a set of basic rules for playing the meta-game.
Recognizing, Picking and Creating Games
In business and life, the game you pick or define is far more important than how you compete within the game. If you are a power player, you need to pick a simpler game that is mostly straights. If you’re competing with an agile player, try to draw him/her “out into the open.” If you’re an agile player, try and draw the power competitors into an asteroid field. Guerrillas retreat into the hills. Dictators try to flush them out into the open. It’s an age-old game.
There are at least four levels of the meta-game here.
- Recognizing: As a beginner, you must first learn to figure out what sort of game you are in: racing circuit, asteroid field, or true-random. Pursuing a racing strategy in an asteroid field, or vice-versa, is dumb. Trying to be meaningfully “agile” in learning a random number is beyond dumb. You must also figure out which of the games suits you best.
- Picking: As an intermediate player, you must learn to pick your game. If you have an agility advantage, find a more complex game — an asteroid field. If you have a power advantage, find a simpler game (a race circuit), or a truly random game, which is simple in its own way (note to computational complexity geeks: there is a potential phase-transition diagram here with agility in the middle, I’ll leave you to sketch it out yourself).
- Creating for yourself: As an advanced player, you have to learn how to create a game that suits your strengths. If you’re a power player, you must learn how to simplify the game through meta moves. If your huge starship is being drawn into an asteroid field by a maneuvering rebel fighter, use big-bang type weapons to pulverize the field. If you are an agile player, complicate and confuse the game for your powerful adversary.
- Creating for others: As an enlightened player, you have to create a game not just for yourself, but your adversaries. One of the most extreme techniques is to create pseudorandomness in a domain that actually supports learning. This means taking a relatively clean signal, using it to learn yourself, and mixing in enough noise for your adversaries that they slip into “faster and faster” mode, mistaking an agility-friendly domain for a power-friendly one.
At the heart of this process is increasing self-awareness. You have to understand your own relative energy and agility capacities, and how they change as you gain experience through one game after another.
What makes competition fun is the loop between Step 4 and Step 1, which creates an arms race. It is often impossible to tell whether you are truly in a near-random domain, or whether a smarter player is creating pseudorandomness for you. When you recognize the latter case, you level up.
I am interested in armchair-thinking through these ideas in the context of real cases and examples, so if you’re interested and have a suitable case, give me a call via Clarity.fm.