Rapid experiment cycles can be applied in many settings to speed up the pace of learning and reduce the gaps between expectations and reality. One of the common terms in literature dealing with that concept is the idea of minimizing the "blast radius".
In short, minimizing the blast radius means making sure you don't screw up something major in your worst-case scenario. That way you can keep executing experiment cycles and improve the pace of your learning. So how can one think about mitigating their blast radius in order to continue uninterrupted learning cycles? I recommend a three part test:
- Are you acting unilaterally?
- Are your actions irreversible?
- Is the potential damage unacceptable?
If your planned experiment only violates one or two parts of the test, proceed! If, however, you find yourself tripping all three, you might want to reconsider your next experiment. What does it mean to trip one of those tests? Let's take them one by one.
Acting unilaterally. In a classic unilateral setting, one person is making a decision to do something without consulting anyone else. That definition can apply quite readily to testing blast radii, but it should also be noted that groups can act unilaterally. In the case of groups, "unilateral" can be equated with "one perspective". Multiple people in a group might all agree to an action, but if their perspectives are mostly overlapping, then that action is still unilateral. In a setting of us vs. them, you need to get input from them if you want to avoid acting unilaterally.
Irreversible actions. Technical domains often have an ability to "undo", "restore", "revert", "roll back", or otherwise reverse an action just taken, but by no means is that capability restricted to technical domains. Putting a physical item somewhere can be reversed if you can just move it back. Announcing a change to a tight-knit team can be "reversed" with a similar level of communication. Limiting the population exposed to a change means the bulk of the population stays with the current reality, essentially letting you revert to old norms easily.
Unacceptable damage. Of the three parts of the test, this is probably the trickiest, and can seem a bit recursive: isn't the whole point of thinking about blast radii to avoid unacceptable damages from an experiment gone wrong? Thinking about the other two parts of the test can help one suss out unacceptable levels in a given situation. As an extreme example, if you decide to make an omelette for yourself, you're acting both unilaterally and irreversibly, but clearly it's an acceptable level of damage. Perhaps more realistically, if you're considering a major product change that you think will be irreversible, you can strengthen your case by seeking different perspectives.
The point of your experiments should not be to do, but rather to learn so that you can do better. Not every experiment will turn out as you expect it to, so you should mind your blast radius, but odds are if you haven't critically assessed your blast radius, you're either being too cautious and taking too long to learn or being reckless and inviting unnecessary risks.
Acting unilaterally. In a classic unilateral setting, one person is making a decision to do something without consulting anyone else. That definition can apply quite readily to testing blast radii, but it should also be noted that groups can act unilaterally. In the case of groups, "unilateral" can be equated with "one perspective". Multiple people in a group might all agree to an action, but if their perspectives are mostly overlapping, then that action is still unilateral. In a setting of us vs. them, you need to get input from them if you want to avoid acting unilaterally.
Irreversible actions. Technical domains often have an ability to "undo", "restore", "revert", "roll back", or otherwise reverse an action just taken, but by no means is that capability restricted to technical domains. Putting a physical item somewhere can be reversed if you can just move it back. Announcing a change to a tight-knit team can be "reversed" with a similar level of communication. Limiting the population exposed to a change means the bulk of the population stays with the current reality, essentially letting you revert to old norms easily.
Unacceptable damage. Of the three parts of the test, this is probably the trickiest, and can seem a bit recursive: isn't the whole point of thinking about blast radii to avoid unacceptable damages from an experiment gone wrong? Thinking about the other two parts of the test can help one suss out unacceptable levels in a given situation. As an extreme example, if you decide to make an omelette for yourself, you're acting both unilaterally and irreversibly, but clearly it's an acceptable level of damage. Perhaps more realistically, if you're considering a major product change that you think will be irreversible, you can strengthen your case by seeking different perspectives.
The point of your experiments should not be to do, but rather to learn so that you can do better. Not every experiment will turn out as you expect it to, so you should mind your blast radius, but odds are if you haven't critically assessed your blast radius, you're either being too cautious and taking too long to learn or being reckless and inviting unnecessary risks.
No comments:
Post a Comment