A rate such as trial conversion gets an interval too. The textbook Wald interval is p ± 1.96 × sqrt(p(1 - p) / n). It works when n is large and p isn't close to 0 or 1, and it fails badly otherwise: with 0 conversions it says the rate is exactly 0.
The Wilson interval stays sensible for small samples and extreme rates, and is the better default.
statsmodels does both: proportion_confint(k, n, alpha=0.05, method='wilson') with k successes out of n, or method='normal' for Wald. Note that alpha is the error rate: 0.05 for a 95% interval.
To plan a sample, solve the margin for n: n = 1.96² × p(1 - p) / margin².