Just writing down my train of thoughts here. This article made me think of a machine learning-like approach of learning the customer needs. You want to find out

Just writing down my train of thoughts here. This article made me think of a machine learning-like approach of learning the customer needs. You want to find out the true distribution of them. Apparently, the more clearly you can identify the needs, the more win for your products. But in some areas, the customers are "noisy" and the boundary (of what's okay and what's not) is not clear. So you want to try various approaches to figure out the shape of what the "needs" look like. In some sense, brainstorming (without criticizing the ideas) is a bit like generative methods. You only want to find the center of distribution. But sometimes you want to know the exact boundary. In that case, you need to employ a discriminative way; to try both sides of the boundary to define the better shape of it.