His idea is this: the risk register should not be stove piped as a on-the-shelf database only referred to occasionally. It should be integrated into the project plan by means of an integrated Monte Carlo simulation that simulates the effects of the risk register on the schedule/resource plan.
Here are the steps:
1. Assign risk distributions to the events on the risk register. This is more than just a simple PxI calculation. It means do a three point estimate on each event; the real world distribution, likely unknown, can be modeled as a Triangular distribution for simulation purposes.
2. Also evaluate whether or not the risk event, if it occurred, would actually effect every task on the schedule, or just some tasks, and--here's a twist--evaluate whether or not a true event would actually effect the schedule/cost plan.
That is, if for example labor rate escalation is a risk event, and it actually occurs, does the escalation really get applied or not? Perhaps there are reasons that the escalation, even though authorized, would not be applied in every case; only be applied in some cases. Hulett recommends that the latter assessment be on a continuous scale of 0 to 1.
With this evaluation, you now have two probabilities associated with a risk event: will it occur, and if it occurs, does it have an effect?
3. Now run a Monte Carlo on the schedule; then, run a Monte Carlo on the risk register; then create an intersection between them.
What do you get? You get a simulation result that integrates the probabilistic characteristics of the risk events on the risk register with the probabilistic characteristics of the schedule/resource plan.
A good heuristic is this: when you correlate otherwise independent events, there is a stretching of the timeline. So, even if you don't go to all the trouble to run such simulations--now supported by automated tools for just this purpose--know this: the schedule will stretch out as the risk register is correlated with the schedule/resource plan.
System engineers know it this way: apply a filter to an event, and the correlating effects of the filter will stretch the event. Widen the bandwidth, thereby lessening the filter effects, and the event sharpens.
And, if you are concerned about making risk assumptions, and the accumulation of errors around estimates and evaluation of risky events, read Hulett's paper on that as well: "Assessing Risk Probabilities".
