For example, if the parameter of interest is a proportion, one could elicit the plausible range (quantiles, formally) of values seen, from multiple samples of 100 independent draws from the population, and reverse-engineer a corresponding prior; the math involved is not challenging.