Real Life PM Predictive Analytics

 

I believe that one of the most, if not the most, valuable information bits that come from even a basic Earned Value Management System (EVMS) is its ability to accurately predict the at-completion costs and durations of Project work. Decades before “Predictive Analytics” was a thing, EVMSs all across the PM universe were doing an outstanding job of generating this highly coveted bit of management information, and doing so rather easily. So when it comes to discussing Project Management in Real Life (ProjectManagement.com’s theme for September), this management information stream simply has to be the first to consider.

David Christensen is widely acknowledged to have performed the seminal work[i] in evaluating the stability of the Cost Performance Index, or CPI. As a reminder,

CPI = BCWP / ACWP

…where BCWP, or the Budgeted Cost of Work Performed, is the Earned Value figure, and ACWP is the Actual Cost of Work Performed, or just “actuals.” CPI stability isn’t just some wonkish parameter relevant to EV enthusiasts – it’s the heart of the most common (and, as we shall see, the most valuable) way to calculate the Estimate at Completion, or EAC. That formula looks like this:

EAC = BAC / CPI

…where BAC is the Budget at Completion, or total budget, and CPI we’ve already defined. Now, eagle-eyed members of GTIM Nation will immediately recognize an opportunity for algebraic simplification, since:

BCWP = % Complete * BAC

…and they would be correct. Without going into the steps, the EAC formula above can be simplified to:

EAC = ACWP / % Complete.

In other words, all one has to know to calculate an Estimate at Completion is the cumulative actual costs of a project/task/activity, usually readily available from the organization’s general ledger (except in those instances where the GL isn’t set up to collect costs by Work Breakdown Structure), and a reasonable estimate of the project/task/activity percent complete is available from the managers in charge of that piece of scope.

This formula is so simple, and with such readily-available data points needed to feed it, that it’s become fashionable in some PM circles to claim that it couldn’t possibly be reliably accurate. But this is where Dave Christensen’s work comes in, for if the CPI is relatively stable for most of the Project’s life, it follows that the simple EAC formula is reliably accurate. So, what does Christensen’s work indicate? That the CPI, for most of the Project (starting from around 20% complete), is indeed stable, to within ten points. And that means that the BAC / CPI formula is usually reliable to within ten points of the true final costs, and that means that the simplified version of the formula is, as well.

But, wait (as they say in numerous television commercials), there’s more! I believe that the same formula works for duration. Simply take the cumulative duration of your project/task/activity (I recommend in calendar days), and divide it by the cumulative percent complete for total Duration at Completion.

So, what are the real-world implications? For starters, the Estimated Costs at Completion can be reliably calculated, serving as a sanity-check for when your PMs are trying to convince the PMO Director in the Project Reviews that their Project is going to finish on-budget, even with a CPI of .89, and the Project itself over 85% complete. That’s virtually impossible, but if the only source of the EAC is that dopey re-estimate-the-remaining-work-and-add-that-to-the-actuals method, the PMO Director may remain completely unaware of the huge Variance at Completion that awaits until it’s too late to do anything about it.

Then we have those very low-budget, primitive capability PMOs that don’t implement widespread Critical Path Scheduling, instead relying on some form of milestone list. Once documented, these milestones are typically “tracked” by contacting its owner, and asking if they think the milestone is on-time, late, or already accomplished, with the answer invariably being “on-time.” Here’s the fix: rather than polling the PM’s opinion of the make-ability of the milestone, instead ask for an estimate of the percent complete. Take that parameter and divide it into the difference between that milestone’s start date and the date of the closing of the reporting period, and you have its likely duration. Compare that duration to the milestone’s originally scheduled duration, and you have the at-completion variance.

So, yeah, kind of like the PM version of an amazing magic trick, we can “do” predictive analytics, in the real world, with just two easily-obtainable data points. This fact must make the non-PM-types (read: our friends, the accountants) envious, but they need not worry. Outside of GTIM Nation, I’m not convinced that a lot of managers know of this little trick.

[i] Christensen, David, Cost Performance Index Stability – Fact or Fiction?, Journal of Parametrics, January 1991.