I used to have a motto: "If something is worth doing, it's worth over-doing!"
While I have got older and have somewhat resiled from that degree of extremism I must concede that my younger self had a point. Here's why: Consensus is boring. If you want to be creative in your examination of issues and ideas, do not be content with looking at both sides of the coin. You will also need to turn it on its side, cut it in half, dye it blue, and compare it with similar coins, learn its history, and, ..., you get the idea.
In the end you may come back to a fairly uncontroversial consensus, but you will have done so with a thorough understanding of the available options. You will have learned interesting stuff, and have acquired an understanding rather than a pre-digested second-hand overview. This will enable you to make decisions or recommendations from a solid base.
Application to Software Design
A feature request comes in or -- more likely -- becomes top priority. You decide to implement it in its simplest form. Is it actually useful? Or is it simply the case that it could be useful? In the latter case this is a token feature. Maybe it could become useful with further work, maybe it should be left in for a while (to find out for sure).
If a feature turns out to be fairly unused and is not the subject of ongoing then surely it is a candidate for removal. Of course this will annoy the 2% of the user-base who do use it, so the usual practice is to baulk at this degree of ruthlessness, and leave it in, or merely deprecate it (i.e. "we plan to remove it, say in 20 years time"). If removal would break backward compatibility, we tend to be very reluctant to remove a feature.
So maybe we did introduce the feature to serve a genuine need, but have not yet gone far enough. Perhaps there is a Tipping Point where with sufficient integration and polishing a feature or feature-set suddenly becomes compelling.
So again it becomes a case of tinkering, polishing, experimenting and, reflecting to get there: maybe incrementally; maybe with a sudden jump.
This approach risks that you overshoot and pass the point of diminishing returns, or end up throwing good money after bad. But, your guts should tell you when to quit. I contend that, having committed to adding a feature, it is a greater danger to do a bit too little than to do too much.
In golfing terms: It is better to push a putt long rather than leave it short.
Practical exercise: Review your software and look for partially realised features. Mark them for removal or improvement.
Question: Given limited resources, is it better to have a smaller, thoroughly-realised feature-set, or a larger, partially-realised set?