In practice
Multiply reach by impact and confidence, then divide by effort. Confidence is especially useful because it forces you to say how much of the estimate is a guess.
RICE isn't an answer. 2 people often score impact differently, and the disagreement matters more than the final number.
Why it matters for product builders
The framework makes you ask the same questions of every item in the same order. That consistency is worth more than the precision the formula suggests.
What changes in AI-first products
Effort is the part that changed. Some tasks collapsed from a sprint to an afternoon while others didn't. Honest new estimates can reorder a backlog dramatically.