Prioritisation · Frameworks, worked

Three features, one quarter

Rank three backlog candidates with RICE, decide what ships first and what to do with the AI feature at 50% confidence, and say what result flips the call.

casemediumExercise · 3 min · 15 of 59

On this page

Three candidates, twelve person-months, and a sales team that has already put one of them on a slide. The RICE lesson scored these three; this is the decision the scores were for.

The situation

A support tool with 12,000 agents across all customers, one team of four engineers, one quarter — 12 person-months. The backlog has three candidates. A, bulk-close for duplicate tickets: reach 4,000 agents a quarter, impact 2, confidence 80%, effort 3 person-months. B, CSV export: reach 12,000, impact 0.5, confidence 100%, effort 1. C, AI ticket summaries: reach 8,000, impact 3, confidence 50%, effort 6. Confidence uses Intercom's scale (100%, 80%, 50%; anything lower is a moonshot and is not scored). The FDE says a working prototype of C takes two days, and five customers have agreed to try it in week one. Sales wants C on the roadmap slide. The support lead who runs 1,200 of the agents has asked for B in each of the last three quarterly reviews, and the team has said "next quarter" each time.

The question

What ships first, and what do you do with C?

A strong answer · Rubric · Where answers go wrong

Write your answer first, then open

A strong answer

RICE (reach in agents a quarter, impact on 3 / 2 / 1 / 0.5 / 0.25, confidence 100 / 80 / 50%, effort in person-months): A = 4,000 × 2 × 0.8 ÷ 3 = 2,133; B = 12,000 × 0.5 × 1.0 ÷ 1 = 6,000; C = 8,000 × 3 × 0.5 ÷ 6 = 2,000. B first: one person-month, certain, out in three weeks, and the customer who asked three times sees it. C's score is held down by 0.5 confidence and 6 months of effort, and confidence is the cheap one to change: a two-day prototype in front of five customers either lifts it to 0.8 or shows the moonshot for what it is. At 0.8, C = 8,000 × 3 × 0.8 ÷ 6 = 3,200 and displaces A. If three of the five customers do not use the summaries in the trial, C leaves the sheet rather than being scored lower, because Intercom's scale has no number below 50%. So: ship B, spend two days on C's prototype, decide A against C in week two. Either sequence fits in 12 person-months: B (1) plus A (3) leaves 8 spare; B (1) plus C (6) leaves 5, enough for A if C's estimate holds. What would change the decision: an effort estimate for C above 8 person-months. At 9, B plus A plus C is 13 and no longer fits, so C competes with A instead of sitting beside it, and at 0.8 confidence C = 8,000 × 3 × 0.8 ÷ 9 = 2,133, level with A. A is a third of the effort, so A goes in and C waits a quarter.

Rubric

Shows the arithmetic for all three scores, with the scale each column uses.
Notices that confidence, not impact, is what holds C down, and that it is the cheapest column to change.
Buys information with the two-day prototype before committing six person-months.
Ships B first because it is small and certain and a customer has waited three quarters, not only because it scores highest.
Checks that the chosen sequence fits in 12 person-months.
Names the result that flips the decision.

Where answers go wrong

"C, because AI is the strategy." Chosen by narrative; it ignores 0.5 confidence and 6 months, and the quarter can end with nothing shipped if C slips. The reviewer's note: no arithmetic, and no plan for the column that is wrong.
"B, then A, and C next quarter — the scores say so." The scores do say so, but this answer treats 2,000 as a verdict on C when it is a verdict on the evidence for C. Two days would change the score by 60%, and the answer never spent them.
"Score C at 25% confidence to be safe." The scale has no 25%: an item below 50% is not scored lower, it is tested or removed. Scoring it at 1,000 keeps a moonshot on the ranked list, where it is argued about every week instead of settled in two days.