Embracing AI in Project Controls: A Journey Begins with GPT-4, 4,998 Projects and a $6 Billion Dollar Budget
TL;DR Started my journey in AI by experimenting with GPT-4 on a fake scheduling program. The whole process was surprisingly easy. After several hours of testing there are countless workflows anyone in Project Controls could utilize. If you can afford it, I recommend adding it. Even if you save just one hour per month, it pays for itself. In addition to being impressed by the experience, I also have a lot of questions.
Looking for more? Sign up to my weekly newsletter here.
I’ve decided to embark on a bit of a journey 🛣
Through my career, I have become interested in topics and have learned about them.
As of yet, I haven't shared my experiences with anyone.
To make this journey more fun, I decided to bring you along.
Over the last couple months my feed has been full of Artificial Intelligence.
Those who follow my stuff know how much I love technology.
The last wave, however, felt a bit overwhelming 🤯.
There was little substance and a lot of clickbait.
Is there any tangible knowledge, tactics, software, or strategies available?
Well, as the wave settles, it seems like a good time to dive in.
AI could have a massive impact, but I'm not sure how or when.
Through intentionally messing around, I hope to gain a better understanding.
I'm not trying to become an expert or teach you how to become one.
Simply want some clarity and some vibes.
As for my strategy, it will go as follows;
Research and learn
Connect with experts
Hands-on experience
Let’s Get Started
And start with Hands-on experience.
Rather than look at pictures or talk to someone about driving a Porsche, why not experience it for yourself?
I created a fake Construction Program with 4,998 projects, operating in 28 cities, managed by 15 Project Managers, with a 5 year budget of $62 billion.
The schedules for each project are simplistic with 5 activities each (Funding, Design, Procurement, Construction, Commissioning).
There are also 6 systems (Conveying, CSA, Electrical, Fire Suppression, Mechanical, Security) tagged to each project.
The data is here if you want to use it as well.
ChatGPT Code Interpreter Setup
Next using a ChaptGPT Pro account I turned on the code interpreter setting in Beta Features.
Start a new Code Interpreter chat by however over GPT-4 and selecting from the drop down.
Export the Google Sheet data to excel.
And upload the file.
Time to party 🎉.
Starting with something simple.
That is correct ✅.
Let’s spice it up a bit.
Correct as well ✅.
Let’s take it up a notch.
It found my spelling error (corrected) and converted columns into date formats ✅.
Dabble in some money 💲.
That is also correct ✅. Who's managing the budgets on these projects? 😂
Asking a more complex question that covers two different tabs.
This could obviously go on forever.
It seems like any basic query can knock out the park.
Machine Learning (ML) used to be the talk of the town, let’s ask it to do a Linear Regression.
I’m trimming the next screenshots because it’s doing a lot of work preparing the data but the screen shots would be massive. Let’s cut to the end.
No shock here, my random fake data doesn’t appear to be able to predict anything.
I'll have to go back and create some obvious trends to see if it makes a difference.
But for now let's wrap up the first session.
Key Takeaways
For the price of $20 a month, I'd highly recommend adding this tool and capability. If it saves you 1 hour a month, it's paid for itself. It fits within your current budget if you cancel Netflix and don't buy a coffee at Starbucks.
One of the easiest setups ever.
Once set up, the workflow is also easy. Seemingly simple things like correcting spelling mistakes and fixing date columns that are not formatted as dates significantly speeds up the process.
All the time I hear, but the data needs to be “structured”, “aligned”, “standard”, etc, before technology would work. Given this experience, there's massive cracks in that reasoning. It appears that 1.) Might not even be necessary and 2) scrubbing schedule data that would take weeks or not be possible appears doable.
I find exploring data and analysis quite useful and entertaining. This might be because it’s a new toy. There are often so many clicks before you get to anything interesting in spreadsheets and PowerBI / Tableau. You can go 0 to 100mph incredibly fast.
Setting up the linear regression was insanely simple. Very impressed that my fake random data failed. There has to be a catch. This seems too good to be true. I’m left wondering 1.) Is this really legit? and 2.) How many other models and programming can it do?
There are loads of tactics and use cases that I would use today if this was available on my program. The sweet spot is something that can't be done quickly in P6 using filters / layouts, excel or PowerBi. Items on that list are frequent and long.
Keep in mind there's an export function. You can do some crazy merges, vlookups, create columns, etc. and then export it back to excel.
How does this and variations fit into organization or gain mainstream adoption? How are the commercial products better? If this is language, what's visual up to?
