RICE (Reach, Impact, Confidence, Effort) is the most adopted and most quietly abused prioritization framework in product management. I used it for years on roadmaps at CaaStle spanning a ~$50M+ ARR portfolio, and I used a leaner version at WisOwl AI. It earns its keep, but only after you fix the ways teams routinely game it.
Where RICE goes wrong in practice
- The confidence column becomes fiction. Teams assign 80% confidence to guesses because 50% feels embarrassing. My rule: confidence above 70% requires citing evidence: a past experiment, cohort data, user interviews. No citation, no score.
- Impact gets estimated in adjectives. "High impact" is not a number. At CaaStle every impact estimate had to be denominated in the metric it claimed to move (subscriber retention points, funnel conversion, ARR), which is how our A/B testing program stayed honest enough to lift cancellation saves to 50%.
- Effort is scored by whoever isn't doing the work. Engineers size effort or the whole exercise is theater.
- The score becomes the decision. RICE ranks your assumptions; it doesn't replace judgment. Strategic bets with long payoffs will always score badly against quick wins. If the spreadsheet alone ran the roadmap, nobody would ever rebuild infrastructure or enter a new market.
How I actually run it
Scoring happens once a quarter in a working session with engineering and design in the room, instead of asynchronously in a doc where nobody can argue with the numbers. Anything scored above 70% confidence gets its evidence read aloud. The output is a ranked list plus a short written note on where and why leadership overrode the ranking. That override note matters more than the scores, because it makes the strategy explicit instead of hiding it inside arithmetic.
Used this way, RICE does its real job: turning prioritization fights into evidence discussions. If your backlog scoring has drifted into ritual, a one-day reset session usually fixes it for a year.