Projects Need Predictive Analytics. And to convince you, I'm going to have to convince my heckler, Bob.


Before you even start, let me jump in. All you are talking about is adding more overhead and more resources, with no value. I can already tell.

Hold your horses Bob. It costs money to have headlights, windshield wipers and brakes in your car but they add serious value despite the cost. Value is the appropriate way to perform this analysis.

Okay…so analytics are like car parts?

Absolutely. What do headlights and windshield wipers do? Let you see farther and more clearly when visibility is poor. What do brakes do? Increase your ability to control your speed and direction. Good analytics do exactly the same thing for a project.

That sounds slightly less idiotic than it did at first. I won't leave quite yet – say more.

Thank you. I hope what you hear next helps convert you from "hold" to "buy". What is a common way that Project Managers usually communicate with their Program Manager (and Program Managers with their Project Board)?

Email? Just kidding. Dashboards.

Right. With Key Performance Indicators and color status. Seen these before?

Sure.

Okay, now think of a few projects that you have seen get into trouble.

Okay.

In your experience, did the dashboards and health reporting for those projects do a good job of conveying project health throughout the project? Tell me about your experiences.

Honestly, they did not. There is often a large gap between the dashboard and reality, particularly between when the problem/issue/risk starts to materialize up until when it becomes a blocker or increases schedule or cost. Sometimes managers can be afraid to put up the flag until they are sure there is no way of getting back on track. I don't see how analytics would necessarily help with this though.

Let's take the easiest example first. Do you agree that there is often a bias in the project budgeting process?

Yes, absolutely. When projects are under-budgeted or over-budgeted, managers will report accurately on history but may bias projections to match budget. This happens over or under budget. I have seen this on many projects. How can predictive analytics help?

There is generally good historical information by project type on original budget projections and final costs. It is a straightforward exercise to assemble this information across a variety of projects, labeling each project by type, scope, delivery slippage and complexity. If you use this historical information, you can build analytical models to explain each project's ultimate performance as a function of these variables. This will give you a powerful way to do a "sanity check" when new projects are in the budgeting and planning stage. Run new projects through the model before they are mobilized, and compare the resulting projection to actual results. Look for a clear explanation when there is a material difference, and use this information to improve your model. You will improve the quality of your projections and find surprising insights. A common learning is that system specific TCO (Total Cost of Ownership) is substantially different than expected when maintenance and project costs are taken into account.

I totally agree with this. But are you talking about real modelling or just good reporting?

Take a look at this spreadsheet.

It includes example data for 50 projects. It was created using very conservative assumptions. The sheet includes the following tabs: Field Guide, Data, Created Metrics, Correlation and Regression. Note in the ‘Created Metrics' tab that we've created a composite independent variable called "System Count * Interface Count" and a composite dependent variable called "% Over Budget". You can see in the ‘Correlation' tab that there is a 28% correlation between this and budget overage. You can see in the ‘Regression' tab that both the coefficient for the intercept and "System Count * Interface Count" are significant at the 95% level…a common threshold used by scientific journals to suggest significance. A 10 point increase in our independent variable increases expected budget overage by (.0042*10) = .042 = 4.2%. This gives us real actionable information. Our budget process is systematically and significantly underestimating budget when projects deal with large numbers of systems and many interfaces. Note that this was a very simple regression that assumed a linear relationship, which might not be the case here. We could probably increase our predictive power and explain more of the budgetary overages by using a more nuanced model.

Okay I can see the value in this from a control perspective. But you are just talking about a budgetary/planning exercise here. What else can analytics do for a project?

You can also use analytics to compare the performance of two vendors, or project resources within and across vendors. A common "top down" sourcing approach looks at $/hour for a role or skill-set, but this is not a best practice from a value based management perspective. Senior leadership often relies on this simple measure because it is objective (if inaccurate). A well run project analytics program can provide strong evidence that one vendor or type of resources is delivering more value per dollar than an alternative that has a lower dollar per hour cost. You may already have the data that you need to run a simple regression that contains both explanatory and predictive power. Look at all your projects for the past 3 years and gather some basic information about their complexity, resourcing and performance to calendar and budgetary projections. Work up a viable model and see what it tells you. Analyzing projects this way will help you to reduce the cost of delivering a unit of value while simultaneously improving projections and reducing risk. It will also allow your organization to develop analytical expertise and enforce quality and discipline on service procurement, reporting and budgeting processes.

Well heck. Thanks! Can I hire you to help me do this?

Yes you can Bob.