Choosing the framework
Pick RICE, Kano or cost of delay for the question in front of you, report a score as a range when its columns can move the order, and name a bet as a bet.
On this page
Three frameworks, three different questions. RICE orders a backlog; Kano classifies what a feature means to a customer; cost of delay sequences work against a calendar. The failure is rarely choosing the wrong one. It is running one of them past the point where its number means anything — and that point is visible in the arithmetic if you look for it.
Three questions, three instruments
A backlog to order: RICE. Twenty to forty items, one team, one quarter, and the question is which ten. RICE wants a reach you can count, an impact on a fixed scale, a confidence on three settings and an effort estimate; it returns an order that a second PM could reproduce from the same inputs. It is the instrument for the planning meeting whose output is a ranked list. The worked lesson scored three items as 6,000, 2,133 and 2,000.
A conversation about what to build: Kano. A customer advisory board, a sales team asking why their request is not on the roadmap, an executive asking why summaries are sold as an add-on and export is included. Kano does not rank; it classifies, and a category is what those conversations are about. "Export is a must-be — 27 of 38 — so it ships and it is never the headline" is an answer a sales lead can carry back; a score of 6,000 is not. Kano is also the only one of the three that says what a feature is worth to the price, which is a roadmap question before it is a backlog question.
Sequencing under a date: cost of delay. When the question is order, not selection — the ten are chosen and the calendar decides what goes first — and when one of the items has a cliff. CD3 wants value per month and duration, and it is the only one of the three that knows a two-week item ahead of a two-month item is cheaper than the reverse. It is also the instrument for the argument with sales, because its input is their number.
The order of the questions matters. A date, if there is one, overrides everything: an item on a fixed date nine weeks out is sequenced first and scored never. A conversation about categories is not a ranking problem, and bringing a RICE sheet to a customer advisory board turns a discussion of what matters into a discussion of why impact is 2 and not 3. Only when the question is a list, with countable reach and a confidence of at least 50%, is RICE the tool — and when confidence falls below 50%, no framework is: the item is a bet, and the next step is the one-day prototype, not a cell.
When the number is precision theatre
RICE returned A = 2,133 and C = 2,000. The 133 is the gap between a feature the sheet says to build and one it says to defer, and it rests on A's impact being 2 rather than 1 and C's confidence being 0.5 rather than 0.8. Move either by one notch on its own scale and the order flips. Showing four digits implies the sheet knows something it does not.
The honest form of a score is a range: the score across the plausible settings of its least certain columns. For C, reach is counted (8,000) and impact is stuck at 3 until an agent has seen a summary, so the uncertain columns are confidence and effort:
C = 8,000 × 3 × confidence ÷ effort
effort 4 effort 6 effort 8
confidence 0.5 3,000 2,000 1,500
confidence 0.8 4,800 3,200 2,400
confidence 1.0 6,000 4,000 3,000
C runs from 1,500 to 6,000 — a factor of four — and beats A (2,133) in seven of the nine cells. The sheet has not ranked A against C; it has reported that A against C depends on two things nobody has measured. B at 6,000 is a different case: B's columns are all counted or built before, so it has no uncertain column to range over, and no cell of C's grid beats it: the best, where C is certain and cheap, only ties at 6,000. The ranking that survives the ranges is "B first, then A or C", and the sheet's job is done at that sentence.
Two rules follow. Report the range when the uncertain columns can move the order, and skip it when they cannot — B's does not need one. And when ranges overlap, the sheet has not decided; you decide, and the decision doc says what evidence would narrow the range. For C that is a two-day prototype in front of five customers, which moves confidence from 0.5 to 0.8 or removes C from the sheet. Two days to settle A against C (every cell in the 0.8 row beats A) is the cheapest arithmetic on the roadmap.
When the decision is a bet
A framework ranks what is known. Some of a quarter is spent on things that are not, on purpose, because a backlog of certain items is a backlog of CSV exports, and the Kano lesson's attractive feature, summaries, sits at 50% confidence. The honest way to carry a bet is to name it as one, size it, and keep it off the ranked list.
The size is the cost of finding out, not the cost of building. C's bet is two days of an FDE's time and five customers' attention, against a possible six person-months; the ratio is the argument. A bet whose cost of finding out is the cost of building — the only way to learn whether tenants want a feature is to ship it — is a different and more expensive thing, and the launch lesson handles it with a 5% ramp behind guardrails.
A split many teams use, stated as an assumption to argue with: roughly 70% of the quarter to the ranked list, 30% to bets with a named learning goal and a date by which the goal is met or the bet is closed. The split is not a framework; it is the admission that the frameworks rank the 70%.
What a framework never knows
None of the three carries the capacity check: twelve person-months is the quarter, and the sheet will happily rank fifteen months of work. None of them carries strategy: a scoring sheet that returns ten small certain items has ranked correctly and may still describe a product nobody chose. And none of them carries the thing the FDE brings back from the customer on Thursday — the ask that is on no sheet because it was discovered this week. The frameworks make the known arguments checkable. The part of the job that is not yet on the sheet is the PM's, and knowing where the sheet's number stops meaning anything is how to tell which part that is. Saying so out loud, with the range and the column that drives it, is the same skill as communicating a trade-off in a design review: name the alternative you rejected and the number it turned on.
In an interview
"How do you prioritise" is a question about judgement dressed as a question about method, and the interviewer scores whether you choose the instrument for the question rather than reciting one. Name the three and the question each answers in one sentence apiece, then pick for the case in front of you and say why: "this is a sequencing question with a contract date, so cost of delay, and I would ask for the date before the value." Then show the limit: "the RICE gap between these two is inside the range of one column, so the sheet has not decided this — I would."
The mistake that loses points is defending a score to the digit. A candidate who argues that 2,133 beats 2,000 has told the interviewer they will defend a spreadsheet against a customer. Say the range, say which column drives it, say what two days of work would collapse it, and let the sheet rank what it can.
Check yourself
1. Fifteen items, all certain, all small, no dates, one team. Which instrument, and what do you add that it does not provide?
RICE, because it is a list to order with countable inputs. What it does not provide: the capacity line (sum the effort column and draw it at twelve person-months) and the strategy check (if the top ten are all exports and settings, the sheet is right and the product is drifting — reserve the bet share and name a bet).
2. Two items score 4,000 and 3,600 on RICE, and both sit at 50% confidence. Has the sheet decided?
No. A confidence notch moves a score by 60% (0.5 to 0.8) and the gap is 11%, so the ranges overlap (4,000–6,400 against 3,600–5,760): either could be the higher. Decide on something the sheet does not have — which is cheaper to de-risk, which has a date, which the FDE can prototype first — and write down what result would have made the sheet decide.
3. Sales asks why their feature is not on the roadmap, and it scored 900 on RICE against a cut-off of 2,000. Which number do you answer with?
Not 900. Run the Kano pair on their customers if it has not been run, and answer with the category: "two of the five customers who asked classify it attractive, three indifferent, and its reach is 300 agents". A category and a reach is an answer a salesperson can take back to a customer; a score below a cut-off is an answer that starts an argument about the cut-off.