The Basel Accords set international banking standards to control how much capital banks are required to hold to guard against the operational risks they face.
The regulation sets the amount of equity a bank must reserve (and therefore cannot lend) to cover various operational risks (fire, lawsuit, pandemic, rogue trading, theft…) as a quantile of the probability distribution on the global loss it would suffer as a consequence of such risks. The problem is to evaluate the impact of internal procedures to minimize certain risks to compare the cost of implementing measures against the amount of equity capital to be reserved.
To do this, each individual risk is calculated from two probability distributions: the frequency distribution (how often it happens) and the severity distribution (how much it costs each time) and the global loss distribution is obtained by aggregating the impact of thousands of risks incurred by the bank and its branches.
The Bayesian probability theory provides a single tool which allows to combine estimations of the risk from data (when available) with risk models constructed from expert judgement (when limited or no data is available) for the frequency and severity distributions, and to aggregate the data for all risks. Performing these computations on specialized hardware will allow to explore more possibilities during Monte Carlo simulations to improve the quality of the estimates and to compare more risk reduction strategies before picking the optimal one.