Rank by dollars, not scores
Every prioritisation framework is trying to approximate one question: how much money might this make, how soon, at what cost. RICE is the famous approximation, and I use it as a tiebreaker, but my primary ranking is more direct: a floor-and-ceiling revenue estimate per experiment, and priority tiers from P0 to P3 based on revenue impact against effort. Writing a dollar floor and ceiling forces the honesty a 1-to-10 impact score lets you dodge, and it produces conversations a board can actually engage with. Even a strong channel idea, Google Shopping ads was one, waits its turn when higher floor-ceiling experiments are unshipped.
The discipline that makes ranking possible at all is a single tracker. At Superpower I consolidated everything into one Growth-Product Experiments Tracker, 22 live experiments ranked by revenue potential in one place, because scattered lists hide the comparison that ranking exists to force. Experimental plays are science, and half-tracked science is astrology with a backlog.
The cadence: graduate or die
The queue only matters if experiments exit it. My cadence: weekly numbers shared with the whole team, and every experiment either graduates to business-as-usual or dies. Public numbers keep the queue honest, and honest nulls are wins, the statistical discipline covered in A/B testing. Funnel experiments get sized by the leak they attack, channel experiments by the revenue path they open, and everything traces back to the number the company is actually priced on, per growth strategy.
Where do queue-worthy ideas come from? Mostly from reverse-engineering competitors, per the 70/30 rule, sometimes from paying for them outright, and increasingly from watching where AI search is reallocating attention faster than competitors notice. The failure mode to guard personally: the rabbit hole. New ideas are more fun than unfinished experiments, I feel the pull weekly, and the ranked queue with public numbers is the only thing I have found that reliably beats it. If nobody owns that queue, that is the moment to read whether you need a Head of Growth.
Last practical note: log the dead experiments as carefully as the graduated ones. The floor-ceiling estimate you wrote before running is the calibration data for every estimate you write after, and a team that reviews its own prediction error quarterly gets measurably better at ranking, which is the entire game.
FAQ
What is the best framework for prioritising growth experiments?
A floor-and-ceiling revenue estimate per experiment with P0-P3 tiers by impact versus effort, in a single tracker. RICE works as a tiebreaker, but dollar estimates force honesty that abstract scores allow you to dodge.
How many growth experiments should run at once?
As many as you can honestly measure, which for most teams is a handful. The constraint is measurement and follow-through: a single ranked tracker of around 20 with a weekly graduate-or-die review beats forty half-tracked tests.
When should a growth experiment be killed?
At its pre-committed checkpoint, when the number says no, including not-statistically-significant. Every experiment graduates to business-as-usual or dies. The ones that linger unmeasured are the real cost.