Deal win probability is calculated by analysing the responses to the Qualification Questions for that deal. If you change any of these responses, the win probability is recalculated.
After each deal win or loss the entire forecasting algorithm is recalibrated to improve forecast accuracy. This uses a Bayesian model to calculate the weightings to give to each qualification question answer option.
When you first use ProspectSafari or when you first use a new set of qualification questions, each questions is given equal weighting. After each win or loss these weightings are recalculated. Some questions will be given a lower weighting or even ignored in the win probability calculation as the system learns.
Consequently even if the deal qualification answers have not changed, the question weightings may have if there has been a deal win or loss in the interim. In this case the win probability might change.
Even if two separate users have an equal number of wins and losses, their qualification question weightings may be slightly different. This is because there is often a degree of subjectivity in the questions such that an optimist might give a consistently higher score in their answers compared to a pessimist. The system learns these biases and adjusts the weightings accordingly.