Do we kill the feature nobody uses, or keep it because three people love it?

I've got a feature that barely anyone touches. My analytics say under 2% of users have ever opened it, and most of those used it once and never came back. By every rational measure, it's dead weight — extra code to maintain, extra surface area for bugs, extra thing to explain in onboarding, extra line in the pricing table that confuses more than it sells.

But here's the problem. Three customers use it every single day. Not casually — religiously. For one of them, it's the entire reason they signed up. They've emailed me about it, they've told me it's the reason they'd never leave, and one of them said they'd churn immediately if it disappeared.

So now I'm stuck between two kinds of truth. The spreadsheet says cut it. The humans say keep it. And I can't tell if the three people are a signal that I'm one niche away from a real market, or just three people who found a home in a product that isn't really for them.

I've killed features before and felt relief. I've also killed one and lost a customer I never replaced. How do you decide when something is dead weight versus under-discovered? And do you ever try to save it first, or just cut and move on?