How Much to Float Activities During Monte Carlo Simulation

If you’ve made it this far, you know that float is used in many ways: to level resources, to pace progress, to respond to shifts in priorities, design holds, and competing projects, or for other strategic reasons, and that this use of float presents a major risk to project completion. However, modeling float use during simulation brings up new questions. For example, which activities should be floated or paced? How likely is it that float will be used? What impact could float use have?

When it comes to choosing activities, stakeholder interviews can help us understand which activities the team thinks wouldn’t be floated. Keep in mind, however, that these decisions may be tentative. In the face of a significant delay or risk, the incentive to pace may outweigh any preconceived notions.

As project managers, we know empirically that float gets used often and early. However, unlike other project-specific or systemic risks, we don’t yet have any historical data to draw upon. One approach could be to put a higher likelihood of float use (e.g. 90%) on earlier activities, reducing it to 50% toward the middle and to 10% toward the end of the schedule. Interviews may help us understand how risk-tolerant the contractor may be in response to owner-caused delays and vice versa, and we can increase or decrease the likelihood of float use on their corresponding activities.

When it comes to impact, we are less well-equipped, since the major determinants of how much float to use are other delays, risks, or resource usage on the project. For example, if a risk occurs on the critical path and some parallel work is floated, there’s an equal chance that all of the float could be used or half of it, depending on the risk being paced. If activities are pushed out in response to crew demand, the amount of float could be anywhere from 0% to 100%, depending on what is necessary to level the resources.

Overall, given the number of scenarios, their complexity, and the lack of data, we’ve found that a good approach for an unmitigated scenario is to use a uniform distribution from 0% to 100% on every activity with a likelihood of 100%. This means, in any given iteration, there’s an equal probability of using no float, all float, or anything in between. In risk analysis, the uniform distribution is often used as an approximation when little or no data exists, and until historical data or ever-more sophisticated algorithms are available, it helps us represent what would otherwise be very difficult to model.