For infrequent features, memorability matters more than learnability
Aliases: infrequent features · return design · memory-first
What it is
For features used two or three times a year (taxes, expense reports, annual settings), "quick to learn" is nearly worthless—users almost always arrive in a returning state, facing not "learning from zero" but "can't recall." The design target for infrequent features is fast recovery for returners, not zero-cost first learning.
Why it happens
Infrequency means learning decays almost fully between uses, so each arrival lands in the deep trough of the forgetting curve: learnability investments (tours, tutorials) serve only the first time, and every later use restarts at the bottom; memorability investments (stable structure, familiar anchors, recovery cues) serve every return. The investments point differently—tutorials optimize use one, structure optimizes use n—and in low-frequency scenarios use n is the norm.
Studying it
Compare the two intervention classes on return performance in spaced-use designs: first-use guidance (learnability investment) versus structure stabilization plus recovery cues (memorability investment), measuring return completion and time. Always analyze first-time users and returning users separately; conflating them is the most common error in infrequent-feature research, because in real usage returners dominate the population.
Where it stops holding
The claim does not fit features whose frequency will rise: a cold-start phase still needs learnability investment before the weight shifts. For low-frequency features with very high first-use failure (unfamiliar domain concepts), recovery cues cannot rescue "never learned," so a learnability floor remains. And "low-frequency" must come from real usage data—team imagination runs systematically high.
Applying it
- Cut one-shot tutorials from low-frequency features; spend the budget on stable entries, clear naming, and in-context hints.
- Keep low-frequency features isomorphic to high-frequency ones in information architecture, so spatial memory built in one transfers to the other.
- Restore context at return time: last selections, frequent items, example values—shortening relocation.
- Verify the "low-frequency" assumption with access-interval data, not the plan's imagination.