I came across this word in the context of Machine Learning, but I like the concept for other domains too, like organisational design and startup strategy.
Overfitting means creating a model for a set of data that fits it so closely that it’s not useful in other situations. Imagine if the only cat you had ever seen was a ginger three-legged cat and then thinking that if other cats don’t look the same, they are not really cats.
In organisations, the concept of overfitting can apply to process design. Do you need a separate SOP that is specific to each situation? Or can you create one that allows for some variation? Where do you set the boundaries? It can also apply when creating teams for new ventures or areas of work. If you don’t know exactly what work the team is going to do, don’t make hard decisions about the team roles and structure – look for a way to accommodate a wider range of outcomes.
For startups, the sheer number of areas where overfitting can arise is dizzying. Finding a niche market that is also valuable enough. Hiring staff who have the right balance of generalist vs specialist skills at each stage of growth. Developing the product for your first customers while keeping it aligned with likely market needs. Positioning your company against others that are also emerging in the same emerging space.
The time to think about whether you may be overfitting is when you look to double down on a certain decision, to close off other routes. Think about how you can maximise the reversibility of the decision, adapt to a variety of routes forward or test your areas of highest risk.