Feature prioritisation frameworks, honestly compared

That does not make them useless. It makes them useful for a narrower thing than people claim, which is worth knowing before you spend an afternoon in a spreadsheet feeling productive.

RICE

Reach times Impact times Confidence, divided by Effort. Built at Intercom and popular because it looks rigorous.

Good at: comparing features that are genuinely similar in kind, when you have real usage data to put in the Reach slot. If you know that four hundred people hit a screen every week and eleven hit another one, RICE will sort those two honestly.

Where it breaks: before you have users, Reach is invented and Impact is a mood. Confidence is supposed to correct for this, but in practice people set Confidence high on the features they already wanted, so it amplifies the bias it was meant to catch. Dividing by Effort also quietly favours small things, so you end up with a tidy list of minor improvements and nothing that changes the product.

MoSCoW

Must have, Should have, Could have, Won't have. Old, simple, still everywhere.

Good at: forcing a conversation with other people. The Won't category is the whole value, because writing something in it is a decision you have to defend out loud.

Where it breaks: alone, with nobody to argue with, everything becomes a Must. There is no mechanism stopping you. I have made MoSCoW lists where twenty of twenty four items were Must have, which is not a prioritisation, it is a to do list with extra steps.

Kano

Sorts features into basics people expect, performance features where more is better, and delighters they did not know to ask for.

Good at: explaining why a beautifully polished feature got no reaction. If it was a basic, doing it well earns nothing, and doing it badly costs everything. This model is the reason your login screen will never impress anyone.

Where it breaks: proper Kano needs a survey with paired questions, which almost nobody runs. Done from memory it becomes a way of labelling your favourite feature a delighter. Categories also move over time, and faster than you expect. Dark mode was a delighter and is now a basic.

Value versus effort

Two axes, four quadrants, do the top left first.

Good at: being drawn in four minutes on paper and being roughly right. If you only use one thing, this is defensible.

Where it breaks: effort estimates are wrong in a specific direction. The features you understand well look expensive because you can see all the work. The vague ones look cheap because you cannot see it yet. So the grid systematically rewards the things you have thought about least.

The problem all of them share

Each of these takes your opinion, multiplies it by another opinion, and returns a decimal. Nothing new entered the system. What changed is that the output now looks like the result of a calculation, and it becomes much harder to argue with, including for you.

A score is not evidence. It is your assumption wearing a suit.

The tell is what happens when the number disagrees with you. If you quietly adjust a Confidence value until the ranking matches what you already wanted, the framework did not make the decision. It ratified it, and you have spent an afternoon buying confidence you did not earn.

How to use them without lying to yourself

Write the inputs down separately from the score, in words. Not "Impact 3" but "I think this matters because two people mentioned it in interviews and one of them was already paying for a workaround". Then the thing you check later is the sentence, not the digit.

Score in a way that admits uncertainty instead of hiding it. If you genuinely do not know whether a feature matters, that is not a low score, it is a different category entirely, and the honest next step is a test rather than a build.

And be suspicious when everything lands close together. Real priorities are lopsided. If your list comes out with eleven features scoring between 6.2 and 7.1, the model has told you nothing and the flatness is the finding.