Early gains are fast and later performance approaches a plateau
Aliases: diminishing returns · power law of practice
What it is
A characteristic learning curve has rapid early improvement followed by smaller gains with each repetition as performance approaches a plateau. A plateau is not absolute cessation of learning; it means that low-cost improvements available under the current task and device have largely been captured.
Why it happens
Early users solve large problems: knowing the next step, locating a control, and avoiding obvious error. Later time is dominated by movement, perceptual confirmation, and system response, which are harder to compress. A new strategy or tool can create another local rapid-learning phase.
Studying it
Plot time and errors on every trial and test whether early and late slopes differ; do not compare only first and last trials. Include enough trials to distinguish a plateau from fluctuation, and log strategy changes. Examine curves by starting skill rather than declaring everyone “learned” from a mean.
Where it stops holding
Late slowing does not prove an interface is optimal: people may stop exploring better strategies or be bottlenecked by waiting. High-risk work should preserve checking even if it prevents further time reduction. Inferring an ability ceiling from curve shape requires extra evidence.
Applying it
- Focus onboarding and feedback on the first few uses, where correct action–outcome mapping yields the largest gains.
- Offer progressively discoverable expert shortcuts without sacrificing the novice path.
- Compare learning curves, not one endpoint; when the goal is faster adoption, inspect early-trial time and error first.