Fitting a logistic curve from early data is extremely difficult because small measurement errors cause huge errors in estimating the asymptotic value. When data points are all on one side of the inflection point, the curve approximates an exponential, making the limiting value estimate highly sensitive to tiny changes in the observed values.