Product & Delivery · Chapter 1 of 13
Prioritisation that survives contact
Frameworks produce rankings, not decisions. This is about the three things that actually determine what ships.
Every product manager can describe RICE or MoSCoW. Almost none of the difficulty is in the scoring — anyone can move a reach estimate to change the ranking, and pretending the framework decided hides who actually decided. The difficulty is in three things the framework does not touch.
First: what is the business actually measured on this year
Not the mission statement — the number the executive team reports upward. Nearly every prioritisation argument is a disagreement about that, conducted without anyone naming it. Once it is explicit, a surprising number of contested items resolve themselves, because it becomes obvious which ones cannot move it.
Second: what is the constraint
Is the business constrained by demand (we cannot sell it), delivery (we cannot build it fast enough) or retention (we lose them after we sell it)? A forty-item backlog will contain plausible work for all three, and only one of them is your quarter. Building beautifully against the wrong constraint is the most expensive mistake available to a product manager, and it is invisible at the time.
Third: how much capacity actually exists
What a decision looks like, as opposed to a ranking
Three things, each with what it is meant to move, plus an explicit not doing list. The second half is what makes it a decision; a ranked list of forty items is a ranked list of forty items.
And state which of the three you expect to be wrong about, and how you would find out by mid-quarter rather than at the end. A roadmap with no falsification point is a forecast dressed as a plan.
What the frameworks model, and where the arithmetic fails
| Framework | What it is really estimating | Where it breaks |
|---|---|---|
| RICE — reach × impact × confidence ÷ effort | A crude expected return per unit cost | Multiplies an ordinal impact ladder as though it were a ratio scale, then divides. The output rank is dominated by whichever input has the widest range — usually reach, which spans orders of magnitude |
| ICE — impact × confidence × ease | The same, with cost inverted | Same defect, one fewer input to argue about |
| WSJF — cost of delay ÷ job size | Economic loss per unit of capacity consumed | The best-grounded and most often invalid. It is the weighted-shortest-processing-time rule, provably optimal for one machine, no dependencies, known durations. A product backlog has many teams, pervasive dependencies and guessed durations |
| Kano | The shape of the satisfaction response, not value | Its most-taught claim — that attributes migrate from delighter to expected over time — is asserted everywhere and I could find no empirical support for it |
| Opportunity scoring | Which desired outcomes are underserved | Sound logic. The vendor's published success rate carries no methodology, sample or control and should be read as marketing |
RICE's confidence term double-counts, and the direction of the error matters.
Confidence is already what your reach and impact estimates are uncertain about. Multiplying by a separate confidence factor applies the risk discount twice — and it does so hardest on exactly the high-variance ideas that produce outliers.
A diagnostic you can run on your own backlog in ten minutes: hold reach constant across every item and re-rank. If the order collapses into an effort ranking, your score is measuring effort with extra steps.
And note what one prominent discovery practitioner does: excludes effort from opportunity assessment entirely, on the grounds that any given problem can have both easy and hard solutions, and teams routinely rule out an opportunity as too hard and then discover a simple solution once they actually look. That is a direct repudiation of the RICE structure at the problem level.
Is there any evidence these work?
No — and the honest answer is more useful than a hedge
The peer-reviewed requirements-prioritisation literature evaluates a completely different set of techniques. The most-cited recent systematic review screened over a hundred papers across twelve years and concluded that the analytical hierarchy process is the most accurate and most industrially used method, with scalability as its main limitation. RICE, ICE, WSJF and MoSCoW are essentially absent from that literature.
Targeted searches for empirical evaluation of RICE return nothing relevant. For WSJF, the only directly relevant work is an uncited master's-level case study — whose own abstract notes that the available case studies are largely produced by consultancies implementing the framework and focus on positive effects.
And there is a structural reason none of this evidence exists. To show that RICE improves outcomes you would have to randomise organisations to prioritisation methods and track product outcomes for years, with the counterfactual backlog permanently unobservable. It is close to undoable.
So the defensible position, and it is a strong interview answer: prioritisation frameworks are alignment artefacts, not decision procedures. Their value is that they force assumptions to be explicit, expose disagreement between stakeholders, and leave an auditable record. None of that requires the arithmetic to be valid — which is fortunate, because it is not.
The case that should change how you think about backlogs
At a major search engine in 2012, someone proposed a change to how ad titles were displayed. It was stack-ranked below other features and sat in the backlog for six months.
An engineer eventually built it in a few days on his own initiative. It raised revenue by 12% — over a hundred million dollars a year in one market alone. For context from the same account: hundreds of engineers working on that advertising system raised revenue by about 1.5% in a good month. The change was worth the equivalent of more than a hundred person-years of work.
The lesson is not “prioritise better”. No scoring system would have ranked that idea highly — it would have scored low on impact and low on confidence, which is exactly what an untested outlier looks like from the outside.
Scoring frameworks are structurally unable to find outliers, because the inputs are estimates and nobody's estimate of an unknown large effect is large.
The only mechanism that does find them is making the cost of trying a low-ranked idea small enough that low-ranked ideas get tried anyway. That is an argument about cycle time and experimentation infrastructure, not about prioritisation method.
Which reframes the whole discipline: the return on reducing the cost of being wrong exceeds the return on being right more often. That sentence is worth having ready.
Check yourself