Scoring the Queue
// the model
Three hypotheses is a conversation. Thirty is a backlog, and a backlog needs an order that survives the loudest person in the room.
ICE is the simplest model that does that: score every item on three things, one to ten, and sort by the average.
| Impact | If this works, how much does it move the metric? |
| Confidence | How sure are you it will work, given the evidence you actually have? |
| Ease | How cheap is it to ship? Ten is an afternoon, one is a quarter. |
what it's really for
Not precision. A 7.3 is not better than a 7.0 in any meaningful sense, and anyone defending that gap is doing numerology. What the model buys you is three specific arguments instead of one vague one — when two people disagree on the order, ICE tells you which of the three they disagree about, and that argument is usually resolvable.
where it goes wrong
- confidence is the number everyone inflates. It's the one that should come straight off the evidence — a finding backed by two cards in the snapshot is an 8, a hunch you like is a 3. Scoring your own hunch a 7 is how a backlog becomes a wish list
- ease is the number everyone underestimates, usually by whoever isn't building it. Ask the person who ships it, and have them include the measurement work, not just the change
- averaging hides a one. Impact 10, confidence 10, ease 1 scores 7.0 — same as three straight 7s, and they are not the same item. Scan the components, not just the total
- impact and confidence collapse into each other when you're tired. Impact is how big if it's true; confidence is how likely it's true. Score them in that order and keep them apart
Score all three hypotheses on one criterion before moving to the next — all three impacts, then all three confidences. Scoring item by item makes you anchor on whatever you scored first.
// hypothesis one
Impact — how much does it move the metric?
1 = barely measurable10 = changes the quarter
Confidence — how sure are you, from the evidence?
1 = a hunch10 = two cards agree
Ease — how cheap is it to ship and measure?
1 = a quarter10 = an afternoon
// score once all three are set
// hypothesis two
Impact — how much does it move the metric?
1 = barely measurable10 = changes the quarter
Confidence — how sure are you, from the evidence?
1 = a hunch10 = two cards agree
Ease — how cheap is it to ship and measure?
1 = a quarter10 = an afternoon
// score once all three are set
// hypothesis three
Impact — how much does it move the metric?
1 = barely measurable10 = changes the quarter
Confidence — how sure are you, from the evidence?
1 = a hunch10 = two cards agree
Ease — how cheap is it to ship and measure?
1 = a quarter10 = an afternoon
// score once all three are set
Hold onto the order these produce. On the next page you'll find out that the top-scoring one may not be the one you run — and that's not a flaw in the scoring.