Sea-level rise and associated flood hazards pose severe risks to the millions of people globally living in coastal zones. Models representing coastal adaptation and impacts are important tools to inform the design of strategies to manage these risks. Representing the often deep uncertainties influencing these risks poses nontrivial challenges. A common uncertainty characterization approach is to use a few benchmark cases to represent the range and relative probabilities of the set of possible outcomes. This has been done in coastal adaptation studies, for example, by using low, moderate, and high percentiles of an input of interest, like sea-level changes. A key consideration is how this simplified characterization of uncertainty influences the distributions of estimated coastal impacts. Here, we show that using only a few benchmark percentiles to represent uncertainty in future sea-level change can lead to overconfident projections and underestimate high-end risks as compared to using full ensembles for sea-level change and socioeconomic parametric uncertainties. When uncertainty in future sea level is characterized by low, moderate, and high percentiles of global mean sea-level rise, estimates of high-end (95th percentile) damages are underestimated by between 18% (SSP1-2.6) and 46% (SSP5-8.5). Additionally, using the 5th and 95th percentiles of sea-level scenarios underestimates the 5%-95% width of the distribution of adaptation costs by a factor ranging from about two to four, depending on SSP-RCP pathway. The resulting underestimation of the uncertainty range in adaptation costs can bias adaptation and mitigation decision-making.