Is AI going to disrupt the world of project controls?

Back in 2012, when I started at Primaned Belgium - now Proove (my first job), I did not know precisely what I would be doing. In fact, it remained hard to explain what we are doing to someone random. I knew it was about Project Management, mostly in the construction industry (I am a construction engineer), very much oriented on time planning/scheduling and it required some IT knowledge/affinities. With this and my dreamful mind, I had the (naïve) dream job idea! I was going to improve the construction industry by implementing tools that would create and monitor the project planning just by dictating to the tool what the project was about, the tool would do the rest for me (I just slightly exaggerate ?)!

And then I started my first week… 4 wonderful days of Oracle Primavera P6 Basic Training in Rotterdam… Don’t get me wrong, I am still working at Proove and fully enjoy what I am doing. However, it was not really what I was expecting a few months before. However, being young, highly motivated and dedicated, I parked that dream somewhere and kickstarted my career as project controls professional based on the current tools and techniques of that market.

Nearly 8 years later, in 2020, time has passed, and lot of things have changed. And something revolutionary came across, I can eventually unpark this dream! Why now? Because Artificial Intelligence (AI), and more specifically, Machine Learning, is coming to support project controls! Therefore, I wanted to write about it and, more specifically, review the value it could bring to the Construction Project Controls area.

Now, let’s try to find a definition which is simple enough for me to understand and, I hope, correct. You will find tons of those on the internet so let’s select one. From Wikipedia, and more specifically Tom Mitchell, McGraw Hill for the first part and Christopher Bishop professor and researcher in Machine Learning for the second part:

“Machine learning is the scientific study of algorithms and statistical models that computer systems use to perform a specific task without using explicit instructions, relying on patterns and inference instead. It is seen as a subset of artificial intelligence. Machine learning algorithms build a mathematical model based on sample data, known as "training data", in order to make predictions or decisions without being explicitly programmed to perform the task.”

In the classical programming world, a computer must be told what to do. From the definition above, you’ll see that it’s not the human who builds the algorithm anymore, neither the mathematical model. In basic English, that would mean, the human explains the machine which problem he must solve, by feeding it with data and explaining the model, and the computer will learn, himself, in order to find the most reliable algorithm.

If you work in the field of project controls, or just know what this means [insert link] , then you also know it is essentially about making data-driven decisions. Given this base, Machine Learning could really be something to add to this project controls world.

How can it support project controls?

If you observe the last decades of knowledge development in the world of Project Controls, you will certainly notice there haven’t been any major disruptions in years. We are constantly improving by managing a larger amount of data and going further into detail thanks to software tools and improved technique support. However, the base remains unchanged. This doesn’t mean nothing new has been invented, far from that. However, from what we can see in the market, we don’t consider them disruptive. The question now is, is machine learning going to disrupt the project controls field of knowledge? Let’s first look at what value it could add.

Machine Learning is essentially based on the data from the past so, of course, everything that follows is based on the fact you would have fed a certain system with your past project data. I then see two main applications; quantitative and qualitative. Both can be split and applied to a lot of project controls techniques and areas. I will also call it “advice” instead of prediction. Project world remains unique and uncertain, there is no tool that will really predict anything. However, it can support the project team by advising on those subjects.

Quantitative advices

This one seems the most obvious to me. It is very much linked with the parametric estimates’ knowledge where you base yourself on the past data and existing knowledge in order to calculate a certain productivity rate, amount, duration, …. However, no human would need to define the related calculation, this part is then done by the machine learning engine. It offers the possibility to take much more parameters and data into account than what a human could handle.

This can be used in two directions. You can receive the number (productivity, duration, resource needed, uncertainty, cost impact of a certain risk, etc.) and then adapt it to match specificities of the current project you’re working on. As you could also benchmark what you have put with what the ML engine would have predicted and the related past data. For example, I estimated this would cost me € 500 000, what is then my confidence level according to “him” (or her)?

Qualitative advices

That one might not come to mind directly when we speak about usage of past data, but I think this has a very high potential in the project controls field. The principle behind this is that, based on what you are doing, the engine will start to give suggestions. For example, when I add some balconies in my estimate, the system will propose to add the related railing and all the related assembly items (quantity of iron, bolts, thermal cut pieces, …).

You could also imagine you’re planning a permitting activity and the tool proposes you to add the “permit refused” related risk. If you include quantitative advices discussed previously, it would also propose you a certain probability and impact. Also, when you add the foundations the tool might suggest to use certain activities, with certain duration and related relationships between activities such that you don’t need to re-develop the schedule every time (btw, if you remember the introduction, that’s what I was expecting in 2012…).

I am not a professional in Artificial Intelligence, but this is what they call “expert systems” where you bring the expertise from other (persons and projects) to the person. Stijn recently posted a blog post about the site disconnect and the difficulty to hand over the site experience to the new generation. I believe this would really help sharing the knowledge and the company specific best practices among the organisation(s).

Going further with my imagination…

If I go further with my imagination, I even think that CPM (and other techniques) could be replaced by a kind of probabilistic schedule based on past data. You would not have links anymore but probability of the occurrence of an activity after other activities. We can also imagine that you describe the project and it would prebuild the schedule, the cost estimate, the risk register, … Ok, maybe I am going too far but… why not? (there are tools that are already going into this direction).

So, is this a disruption in project controls?

I guess, as for any kind of prediction, no one can tell in advance either it’s going to disrupt our project controls field of knowledge. However, I see several takeaways that would really improve the current situation:

• AI can act as a facilitator, enabling more project team member to do what currently requires a specific knowledge
• Hand over experience to the new generations
• Improve accuracy of estimates (time, cost, risks, …) by looking at a larger amount and variety of data that would take years to analyse by a human
• Those more accurate estimates also have a stronger justification

If you ask me, I would say yes, we are at the start of a revolution in project controls where, by becoming more accurate and more accessible, more and more companies will be convinced to go for data-driven project management.

What do you think, do you share this opinion? Do you see other application? Have you experienced the application of machine learning in project controls? Feel free to comment!