How-to

To prioritise growth experiments, score every idea in your backlog with a simple framework, ICE, PIE, or RICE, then run the highest scorers first; the framework forces you to compare impact against effort instead of chasing whatever feels exciting. Pick the model that fits your data and stay consistent.

TL;DR in 60 seconds

• ICE scores Impact, Confidence, and Ease. Fast, rough, good for a busy backlog.
• PIE scores Potential, Importance, and Ease. Built for conversion-rate optimisation and page testing.
• RICE scores Reach, Impact, Confidence, and Effort. Best when reach and effort vary a lot across ideas.
• Consistency beats precision. The scores are relative rankings, not truth; use the same model across the backlog.
• Agents can score continuously. An agent re-ranks the whole backlog against live data as conditions change.

The three frameworks

Each model multiplies or divides a few factors into a single comparable number. None is objective. Their value is that they make you argue about impact and effort out loud, before you spend time building.

ICE = Impact x Confidence x Ease. Three quick 1-to-10 scores. It is the fastest to run and the easiest to game, so it works best when a single owner scores everything and just needs to order a long list.

PIE = Potential + Importance + Ease, averaged. Potential is how much room a page has to improve, Importance is how much traffic or value it carries, Ease is how hard the test is to run. It was designed for CRO, so reach for it when you are prioritising page and funnel tests.

RICE = (Reach x Impact x Confidence) / Effort. Reach is how many people the change touches in a set period; Effort is the cost in person-time. Dividing by effort makes RICE the right call when your ideas range from a one-hour copy tweak to a multi-week build.

A quick comparison

When to use which

• Choose ICE when you have many ideas, one scorer, and you want an order by the end of the day.
• Choose PIE when the work is mostly website and funnel optimisation and you can judge each page's headroom.
• Choose RICE when ideas differ wildly in audience size and build cost, and you have reach numbers you trust.

The wrong move is switching models mid-backlog. Because every score is relative, mixing frameworks makes the rankings meaningless. Pick one, score the whole list, revisit quarterly.

Common mistakes

• Treating scores as facts. They are a shared guess. The number orders your list; it does not predict the result.
• Inflating confidence. Everyone believes their idea. Anchor confidence to evidence: prior tests, data, or a clear mechanism.
• Ignoring effort. ICE and PIE both fold effort into "Ease", which is easy to under-weight. If your builds vary a lot, RICE keeps you honest.
• Scoring in a vacuum. A prioritisation done once and never updated goes stale as traffic and costs move.

How agents score a backlog continuously

A human scores a backlog on a Monday and it is out of date by Friday. Agents remove that lag. Wired into your joined data layer, an agent can re-score every idea as conditions change: a page that just started leaking conversions rises, an audience that got more expensive falls.

That turns prioritisation from a quarterly ritual into a live signal, which is what keeps experiment velocity high without running low-value tests. Humans still pick what actually ships; the agent just keeps the ranking current and evidence-based. The scored backlog then feeds straight into the experiment loop.

FAQ

Which framework should a small team start with?

ICE. It needs no reach data, takes minutes, and gives you a usable order immediately. Once you are running steadily and your ideas start ranging from tiny copy tweaks to large builds, graduate to RICE so effort is weighted properly. PIE is the specialist choice if your work is almost entirely CRO.

How do I stop confidence scores being wishful?

Tie confidence to evidence, not enthusiasm. Reserve high scores for ideas backed by a prior test, a clear data signal, or a well-understood mechanism, and cap anything based on a hunch. Writing a one-line reason next to each confidence score exposes the guesses quickly and makes the whole backlog more honest.

Can I combine frameworks?

Not within one backlog. Because each model produces relative scores, mixing them makes rankings incomparable. Choose the framework that fits your situation, apply it to every idea in the list, and only switch when you deliberately re-score the entire backlog under the new model.

How often should I re-prioritise?

Manually, at least once a quarter, because traffic, costs, and priorities drift. If an agent is scoring against live data, prioritisation becomes continuous and you review the ranking rather than rebuild it. Either way, treat the list as a living document, not a decision you make once.

Cadence keeps your experiment backlog scored against live data, so the next test you run is always the one most likely to pay.