Keynote: Highly Predictable Schedules
June 13, 2022
In this keynote, Dr. Gui unveils his vision for highly predictable schedule forecasts by using reference class forecasting to target project duration while preserving probable P-value as the project is executed. This is attained, in part, with Level 3/4 forecasting schedules that algorithmically extrapolate future schedule performance (right of the data date) from patterns discernible from as-built schedule data.
I believe that the time has come for schedulers to mine the wealth of actual durations—specific to this project—accumulating left of the data date and to convert the heretofore hyper-detailed deterministic schedule into a true forecasting model that relies on actual performance, not just on estimates, to predict completion likelihood. The value of a progressed detailed schedule incorporating actual, statistical duration data as a reliable project completion forecasting tool far exceeds using the schedule for predicting when, say a 10-day activity two years in the future, may be started.
The mission is nothing short of reforming project-wide scheduling into project completion forecasting and, to once and for all, solve the conundrum of most projects completing late despite the billions spent year after year on scheduling projects. The outcome of a schedule should be algorithmic likelihood of completion, and conversely, to compel replanning to restore the targeted likelihood when periled by progress—or by the lack thereof.
Dr. Ponce de Leon is recognized as one of our nation’s foremost planning and scheduling subject matter experts. His broad professional experience includes executive and senior roles as investor’s developer, program manager, construction manager, and EPC contractor planner/scheduler. He has continually pioneered innovations in project management and written widely on the use of CPM in construction contracts as well as on schedule, delay, and acceleration analysis.
First, academics, knowledge purveyors (e.g., PMI, AACE, et al.), and scheduling professionals’ notion that scheduling that undermines project completion forecasting is settled knowledge.
Second, schedulers willing to stick to software deploying a 65-year-old algorithm that offers little of the real-time, visual interfaces standard in other endeavors (e.g., Google maps).
Third, case law that reinforces continued reliance on a deterministic critical path schedule as valid for ascertaining responsibility for the delay.
Once schedulers embrace stochastic vs. deterministic forecasting and software tools become available, there will be an opportunity to measure the impact on completion based on reducing critical path total float and lowered completion risk. As projects complete and the analysis based on P-value is validated by actual outcome and eventually case law, specifications will institutionalize completion risk.
Because schedulers have disregarded the schedule left of the data date as worthy of attention, case law has validated expert opinion on delay based on ex post facto as-built schedules, with as-built schedules and analysis from opposing experts rarely leading to the same outcome. If you have a GPM built schedule with accurate actual (as-built) data left of the data date, you will contemporaneously know actual delays and whether or not they are on the then-current as-built critical path.
By contrast, if you have a CPM schedule, you can export it to Project Summit and reveal then-current as-built total floats and critical paths, making it unnecessary to spend millions on experts’ ex post facto as-built schedule analyses.
It takes a scheduler who understands the difference between schedules and project duration/completion forecasting and how they complement one another. We need schedule specifications, textbooks, academia, and RFPs to include reference class forecasting to set realistic project duration and increase knowledge and expectations. You can get certified as a reference class forecaster.
It’s not so much the size of the project but how crucial it is that you meet your forecast. The expense of the extra as-built data effort will scale up/down with project size.
Neither textbooks nor academia have kept pace with complexity and innovation in project schedule forecasting. Schedulers can get caught up in the schedule detail, forget that they also need to be forecasters, and then lose sight of beginning with the end in mind. To have an accurate forecast, you need an excellent, current schedule. Risk assessment plays a vital role in creating a schedule with an accurate delivery date. When we do a schedule risk assessment, those schedules are on the project completion forecast because you are getting a completion distribution curve.
You will replace any distribution you started with by sampling from the actual (assumed accurate) duration distributions in reference class forecasting for activities. So, if you had 2,000 activities right of the data date of your simulation, maybe 1,500 you’re sampling from as-built distributions. For the other 500, you will be using whatever function you prefer.
Absolutely not—a project team is still better. Ideally, the project team includes one or more schedulers that can forecast the duration and risk assess the schedule. In reality, you can be a little good at everything but not really good at everything.