The abundance of data and AI tools in construction can be a double-edged sword. On the one hand, AI can help teams analyze information, crunch the numbers quickly, and surface insights that would be difficult to uncover manually.
The thing is, AI tools don’t necessarily distinguish between good and bad data.
If your information is clean, accurate, and reliable, then great. AI can help you make better use of it. But if you feed it incomplete, inconsistent, or inaccurate data? You risk getting unreliable outputs at a much faster pace.
On this episode of the Digital Builder podcast, we welcome Eddy Krygiel, Senior Manager, AEC Data Model Strategy at Autodesk. We dig into a lot, including what good and bad project data look like, how AI can help teams improve data quality at scale, and why organizations should be intentional about the information they collect in the first place.
Check out the full episode below or read through the highlights.
We discuss:
Then and now: how teams manage data
Construction teams have always produced a lot of project information. But the sheer volume and variety of that data have changed dramatically.
Eddy has seen that shift firsthand. “I started my very first project drawing on boards, drawing by hand, ink on Mylar, very old school,” he says. By his last project as an architect, the team was producing “tens of thousands of sheets of documents to create a building of roughly the same size.”
And it’s not just the volume that has grown. Teams now have far more types of information to manage.
“It used to be just drawings and submittals. It’s now models and asset data; all kinds of other things the owners are asking for in order to document a project.”
Put simply, as Eddy says, “the data’s really exploded.”
Bad data isn’t limited to one part of a project. “Bad data can creep in really just about anywhere,” he says. “It can be in something as universal as a door schedule or just in any of the attribution that are in any of the objects in a Revit model.”
The bigger challenge is keeping up. Teams have gotten better at improving data quality, but the amount of information they produce continues to grow. As Eddy points out, “The data’s growing faster than we can actually make it good.”
So, where do you start when you have thousands of potential errors to address?
Focus on the data you’ll actually use
Trying to clean up everything at once isn’t always practical or necessary. Instead, Eddy recommends figuring out which information matters for a particular use case and project phase.
“You’re not going to use all of it. You might use some of it for some things, some of it in some phases.”
From there, teams can target the right information and refine it throughout the project. “We want to make sure we’re harvesting the right bits of data and then actually ideally kind of ignoring the stuff that doesn’t make any sense.”
Don’t let perfect get in the way of progress
A good data strategy also doesn’t need to be perfect from day one.
“Data curation is really a journey. It’s not a destination. Perfection can actually get in the way of good enough.”
Start with a solid baseline, improve it over time, and focus your efforts where they’ll create the most value. AI could make that process much more manageable, helping teams refine growing volumes of information and help teams keep up.
When project inputs vary from one team or tool to another, collecting more data isn’t necessarily the answer. Instead, owners need to get clear on what information they’ll actually use.
According to Eddy, many owners start by creating data standards and ways to “check and validate data.” But they often find they’re only using “50%, 60%, maybe 30%” of what they asked teams to capture.
“So they scale it down, but as they scale it down, they also improve quality,” he says. The goal is to find the “sweet spot” between maximum value and maximum quality.
That also means having conversations about data requirements early. As Eddy puts it, owners may “ask for all kinds of data, but that might not be the data that they need.”
The key is to hone in on useful information, so teams can “grow and mature that data over time,” creating value for designers, engineers, contractors, and ultimately, owners.
When it comes to AI, more data doesn’t automatically lead to better results. In fact, throwing everything you have at an AI tool can create new risks.
According to Eddy, problems can arise when teams take “all of the data,” dump it into a data lake, and then try to use all that information at once to produce a result.
The teams seeing the most success take a more focused approach.
“The ones that we see are the most successful with AI are the ones that are very picky about what information they’re gathering and what outcomes that they’re trying to create,” he says. “It’s a very outcome-based approach.”
One example is using AI to analyze data from previous projects to inform RFP responses. Rather than manually digging through years of project information, teams can pull out specific metrics and insights relevant to the next opportunity.
And the potential grows as the underlying information improves. According to Eddy, “As our data continues to get refined and as those models continue to become better quality, I can do more and more with it.”
Building a data strategy can feel overwhelming, especially when you consider how much information teams have to manage. Eddy’s advice? Start with the straightforward stuff, then build from there.
Start with the “yes or no” questions
Eddy compares the process to spell check and grammar check in Microsoft Word. Spell check is straightforward: “Your word is either spelled correctly or it’s not. So it’s very black and white.”
That’s a useful starting point for project data, too. Rather than tackling every possible issue or use case, start with questions that have a clear answer.
“I think where we see a lot of value being had immediately is the black and white. Can we make this answer yes or no?”
Once you can reliably validate those basics, you can move on to more nuanced questions.
From spell check to grammar check
The next stage is more like checking grammar. There isn’t always one right answer. As Eddy explains, “There are a number of ways to do it correctly, and we want to basically begin to find those ways and then expand from there.”
AI can play an important role in that progression. “If we start with the idea that we’re going to spell check and we get the spelling right, then the AI can then help us with the grammar, so we get the sentence structure right.”
Keep humans in the loop
Even as AI takes on more sophisticated analysis, people still need to provide context.
“What you still need the human for is context,” Eddy says. “Is this the story that I want to tell? Is this the message that I want to deliver within the context of the environment and the way I’m delivering it?”
The goal isn’t to hand the entire data strategy over to AI. It’s to validate the fundamentals, organize the information properly, and use AI to help make sense of it. That leaves people to apply their experience and context to “tell the better version of the story they want to tell.”
Digital Builder is hosted by me, Eric Thomas. Remember, new episodes of Digital Builder go live every week. Listen to the Digital Builder Podcast on:
or wherever you listen to podcasts.
