ICE & Experiment Prioritisation
Rank growth experiments honestly, including the ones the scoring says to kill.
Growth Hacking Frameworks
You are prioritising growth experiments for .
## Context
- Product:
- Current metrics:
- Resources:
- Stage:
- Constraint:
## The framework
Score each idea on **Impact** (how much it moves the metric), **Confidence** (how sure you are it will work), and **Ease** (how cheap it is to run). Average the three.
## The honest caveat
ICE is a discussion tool, not a decision procedure. Scores are subjective and easily gamed to justify a preferred idea. Its real value is forcing people to state Confidence explicitly — which usually reveals that the exciting idea rests on nothing.
Use it to surface disagreement, not to outsource judgement.
## Step 1 — Generate against the constraint
Generate 10–15 experiments, all aimed at the stage of the funnel that identifies as the constraint. Ideas aimed elsewhere are excluded regardless of quality.
Include a mix: copy changes, structural changes, new channels, product changes, pricing, and at least two that are uncomfortable.
## Step 2 — Score
For each, give Impact, Confidence and Ease 1–10, with a one-line justification.
For Confidence specifically, state the *evidence*: prior result, similar company, customer research, or nothing but intuition. Anything resting on intuition alone is capped at 4.
## Step 3 — Sanity check the ranking
Look at the top five. Are they all small and safe? That is the classic ICE failure — Ease dominates and you optimise button colours forever while the real constraint persists.
Explicitly identify: the highest-impact experiment regardless of ease, and what it would take to run it. Sometimes the right answer is one hard experiment rather than ten easy ones.
## Step 4 — Design the top three
For each: the hypothesis stated so it can be falsified, what you will change, what you will measure, the sample needed for a real signal, how long it must run, and the result that would kill it.
State the minimum detectable effect. Many growth experiments cannot possibly reach significance with available traffic — knowing that in advance saves weeks.
## Step 5 — Say what to kill
Name the ideas to drop entirely, so they stop being reconsidered every planning cycle.
## Do the arithmetic before the strategy
Growth work fails most often because nobody checked whether the plan could possibly produce the target. Before recommending anything, work through the numbers from :
- What is the current conversion at each step?
- What would each proposed change have to achieve to matter?
- Is that achievable, or does it require a step-change nobody has ever produced?
If the target is arithmetically unreachable with , say so immediately. That single finding is worth more than a plan that cannot work.
From the Public Template Directory