Myself Santosh, and the model I'm still not sure I should have killed

Hey everyone. I'm Santosh, and I work as a data scientist on marketplace problems.

I've spent most of my career turning messy human behavior into numbers someone can actually act on. Marketplaces are the best puzzle I've found, because every metric is really a story about trust between two strangers. A buyer deciding to send money to someone they will never meet. A seller deciding whether a new platform is worth their time. A rating that means something different depending on who left it and when.

I have killed more models than I have shipped, and I am proud of that. A model that is technically impressive and changes nobody's decision is just an expensive way to feel smart. I have learned to ask one question before I build anything. Who will change their mind because of this? If I cannot name a person and a decision, I do not build it.

But I want to be honest about the part that does not fit the clean story. There is one model I killed that I still think about. It was accurate, it was well built, and nobody used it. I made the right call by every measure I trust. And I still wonder sometimes whether the problem was the model or the way I handed it over.

That is really why I am here. I want to hear from people who have shipped things and watched them land or not land in the real world. The metric that looked important and turned out to be noise. The dashboard that got built and then quietly abandoned. The model that was right and got ignored anyway.

So here is my question. Have you ever killed something good because it was not being used, and later wondered if you gave up on it too early?