How to turn your estimates into reliable predictions

Log-normal distribution and the multiplicative random effects in software development is interesting. I wonder if this isn't the case for non-software knowledge work.