Think on how to optimise for outsized gain without outsized risk exposure.
The premise is that there is a non-linear relation between risk of failure and potential gain in any set of choices. It’s not always true that one must take on more risk to create the possibility of greater gain – and it’s not always true that aiming for X increase in potential gain increases risk by some equivalent factor of X
The easy argument is that we should seek to maximise the possible gain per risk – this makes fairly universal sense.
The hard argument is that the logic applies even for extremely long shots, areas where probability of a huge gain are super tiny. The only thing that matters is the ratio between risk taken and potential upside, diversification efforts, and a personal evaluation of risk appetite.
In business we tend to accept some performance variance in low-risk environments, and control variance more in high-risk environments. (assuming risk to mean the potential for randomness to take down the business)
Instead, we should seek to maximise variance in low-risk environments, and work to build high-risk environments that are resilient to high variance.
What a “low-risk” environment is will vary between businesses. For some, it’s a sandbox or incubator, for some it’s an experimental business unit or team, and for some it’s just that one guy, Bob, who has to test all the new features of the platform betas. For others it’s just a specific budget and time set aside for testing and learning.
When we maximise variance in all scenarios we optimise for black swan events. When we do it in low-risk (low spend, low effort, low danger) environments, it feels a lot like hunting cheap thrills.
Cheap thrill example:
When planning advertising spend, don’t just launch 10 things you have specific performance expectations for. Launch 8 things you know what are going to do, and 2 things that you have a hunch about. Worst case, you take 20% of your effort and some insignificant portion of budget and burn it, black hole, gone. Best case, your hunch is right and you go on to win a nobel prize for world-changing advertising campaigns, stop the global warming crisis and win the race to Mars.
Having a high capacity to run tests with low downsides and MASSIVE potential upsides is a powerful situtation to be in.
Having the technical ability to launch those tests provides a platform for creating an exciting array of cheap thrills, some of which will expose you to some portion of their potential upside.
There is no formula for determining your hit rate, but there’s a gradual skill that develops. Hunching. Proper hunching about what might do something, even though you don’t really get why.
The entire cheap-thrills scenario only really works with epic people calling the shots on what tests to run. Otherwise the hunching doesn’t develop.
When we try to design systems that are resilient to variance in higher-stakes environments, we seek to leverage the same effect at a larger scale. Opening ourselves to the potential for huge wins, without increasing risk correspondingly.
Variance-accepting-but-resilient-system example:
If at all possible, have people develop their own ways of getting things done. Introducing someone to a new ad platform? Give them a loose guideline/strategy intro, a budget, and time to develop their own tactics for hitting the targets. Make sure they know they can also trash the strategy in favour of their own, better version.
Control this environment with stop-loss rules and similar, instead of top-down “this is how we do things” instruction.
In multi-brand companies such as P&G, the multiple business units and brands are another way of embracing a high resilience to variance. Each unit has autonomy to expose the business as a whole to massive potential upside, while being risk-controlled by strict performance criteria.
Random riff concluded, maybe this’ll get edited someday.