It is found by generating many scenarios for market factors, computing the positive part of the market value at each future date, and taking a percentile of that distribution. Because there is a value for each date, it is shown as a function of time (a profile). Consider a hypothetical example in which the 95% potential future exposure from 1,000 paths is 500 million yen. About 50 paths have an exposure above 500 million yen at that date.
Because it represents how large an exposure to one counterparty can be tolerated, credit limits use this measure rather than expected exposure. In practice around 2017, when it was computed for credit limits, real-world probabilities estimated from historical data were often used.
The regulatory SA-CCR formula also has a term called PFE, but it is calculated by formula from an add-on determined by the notional amount, the remaining maturity, and other factors. It is different from a percentile taken from a simulation.