Prioritisation · Choosing

The RICE sheet that lied

Find the two scores on a five-row RICE sheet that change when every 100% in the confidence column is replaced by the evidence, and say what the sheet should propose instead.

critiquehardExercise · 4 min · 17 of 59

On this page

A RICE sheet arrives for the quarterly planning review with the ranking already drawn. Every cell in the confidence column reads 100%, which means the column has not been scored at all — and the RICE lesson says that is the column a sheet lies in first.

The situation

The support tool's backlog sheet for next quarter: five rows, one team of four, 12 person-months. Reach is agents per quarter, impact is on Intercom's 3 / 2 / 1 / 0.5 / 0.25 scale, confidence is 100% / 80% / 50%, effort is person-months.

                              reach   impact   conf   effort   score
D  bulk-close duplicates      4,000     2      100%     3       2,667
E  CSV export                12,000     0.5    100%     1       6,000
F  AI ticket summaries        8,000     3      100%     6       4,000
G  sentiment flag on tickets  8,000     1      100%     2       4,000
H  SLA-breach alerts          6,000     2      100%     4       3,000

What is known about each row. E has been built before by this team, and reporting teams export by hand today. G shipped to two tenants as a beta in the spring; its eval scores 90% and both tenants' agents use it daily. H was asked for by name in 14 of last quarter's 40 interviews, and its estimate comes from alerting work the team did last year. G's and H's reach are seat counts from billing, and each of the 14 described a breach they had missed. D's reach comes from the duplicate-rate query in the warehouse; its impact of 2 is the PM's guess, since no agent has used a bulk-close. F has no eval and no prototype, and its impact of 3 is taken from a sales deck; the FDE estimates two days to a prototype. The sheet proposes E, then F and G, then H.

The question

Which two scores change when the confidence column is scored from the evidence, and what does the sheet's proposal become?

A strong answer · Rubric · Where answers go wrong

Write your answer first, then open

A strong answer

Score confidence row by row on the three-setting scale. E: reach counted, impact observed — 100%. G: shipped, evaluated, used — 100%. H: reach counted, impact heard in 14 interviews, effort from comparable work — 100%. D: reach from data, impact a guess — 80%. F: reach is opted-in tenants rather than agents who want summaries, impact from a slide — both judgements, 50%.

Two scores change:

D = 4,000 × 2 × 0.8 ÷ 3 = 2,133   (was 2,667)
F = 8,000 × 3 × 0.5 ÷ 6 = 2,000   (was 4,000)

Honest ranking: E 6,000, G 4,000, H 3,000, D 2,133, F 2,000. F falls from joint second to last; D falls too, but rises past F. The proposal E–F–G–H was 1 + 6 + 2 + 4 = 13 person-months — it never fit — and six of them were going to the one row with no evidence. The honest sheet proposes E (1), G (2), H (4), D (3): ten person-months, every row with at least one counted column, two spare.

F is not scored lower and forgotten; it is the bet. Two days of the FDE's time and the two spare months are the plan: prototype in front of opted-in tenants, and if it lifts confidence to 80%, F = 3,200 beats H and the sequence is re-cut. What would change this answer: a beta like G's, an eval at 90% and agents using it daily, which makes the 100% honest and puts F level with G at 4,000; E, G and F are then 9 person-months, so H's 4 no longer fits and D's 3 takes the last slot.

The sheet's real failure is structural. Confidence identical on every row is a column doing no work, so the ranking was reach × impact ÷ effort — a sheet that cannot tell a shipped beta from a slide.

Rubric

Scores confidence row by row from the evidence given, on the three-setting scale, and says why each row sits where it does.
Finds that F and D are the two rows that move, with the arithmetic for both.
Notices that the original proposal exceeded the quarter's 12 person-months.
Keeps F as a bet with a two-day prototype rather than scoring it lower or dropping it.
Names the evidence that would make F's 100% honest.
Says why a column with the same value in every row has stopped ranking anything.

Where answers go wrong

"F is the only problem — it is the AI feature." It finds the loudest lie and stops. D's impact is a guess too, and at 80% it drops 20%; an answer that checks only the row it already distrusted has not scored the column, it has confirmed a prior.
"Set F to 25% to be safe — then it scores 1,000 and the ranking is clear." The scale has no 25%. Below 50% an item is a moonshot and leaves the sheet for a prototype; parking it at 1,000 keeps it in every weekly argument and spends nothing on finding out.
"Reorder to E, G, H, F — F still scored 4,000 and the quarter has room for it." The quarter does not: E + G + H + F is 13 person-months, the original overrun with the rows shuffled. And it accepts a 100% that rests on a sales deck, which was the thing to critique.