AI hits the commercial mainstream

I’ve been getting a lot of questions about AI –what is it, how do I use it, what data do I need to gather to feed the machine, what does it even MEAN?? (very angsty, usually in an “am I already behind” panic)– and thought I’d try to answer some of the questions about where we are at this moment. I’m using an AEC example because Autodesk and one of their premier customers graciously gave me some of their time earlier this year, but this technology will eventually be applicable many more use cases. Even if you’re not building schools or roads, read on!

First, a recap: A subsegment of AI, artificial intelligence, are technologies called machine learning, ML. ML is further broken down into two broad categories: Supervised ML, where humans help teach an algorithm and Non-supervised ML, where the algorithm attempts to learn on its own by classifying things via pattern examination. In supervised ML, for example, a human trains an algorithm to recognize a specific type of car (sedan) in an airport parking lot by clicking on all of the sedans in an image. The algorithm then tries it on another parking lot; the human corrects, rinse and repeat. Unsupervised ML is still pretty rare in commercial applications, but one could see the algorithm identifying all vehicles with truck beds, for example, as belonging to one class.

Think of ML this way: It’s like having an assistant who analyzes whatever data you throw at them, never gets tired or takes a sick day, and isn’t irritated when you ignore their suggestions.

The first real example of an off-the-shelf AI solution that I know of in our world is Autodesk’s Project IQ, now commercialized as Construction IQ.

It arose out of the recognition that Autodesk customers were entering reams of data from a construction site into Autodesk BIM 360 as part of their routine tasks. Hmmm … perhaps we can use that data for other purposes that add value to the project. Wandering around the site and see a safety rules violation? Snap a photo with your BIM 360 field app to bring it to the attention of the site manager, who may call out the individual after repeated violations. See it happening again? or with other individuals from the same supplier? Now we’re into analytics territory: Is this a training issue? Are the rules unclear? How does this affect the site’s insurance or legal exposure? Keeping track of all of the incidents and drawing conclusions is exactly what ML is meant to do. There are dozens of potential use cases on a typical construction job, from subcontractor management to construction site environment monitoring, safety, and planning, to name just a few. The construction industry also uses standardized reporting and stylized terminology, making it easy for pattern-recognition algorithms to wade through data.

Autodesk, along with a number of key construction company customers, chose to focus its first AI efforts on water infiltration aka leaks. Lots of potential causes (improper cladding installation, poor workmanship, shoddy materials) and huge potential impact when owners experience leaks, therefore an important and rewarding challenge to tackle. Workers on the job site were already entering data about potential problems into Autodesk’s BIM 360 cloud apps, so it became a matter of analyzing that data to predict whether an issue would/could cause damage if not resolved.

Remember that the AI can’t actually DO anything to fix the problem; it identifies a trend that a human needs to know about, decide if actionable and then the human does whatever is appropriate.

How does it work? By using the descriptions that construction quality managers already observe and note as they monitor their projects. For example, someone notices that the flashing outside a window is improperly installed and records it in BIM 360, as is usually the practice to spawn a work order to get it fixed. That happens as per normal, but overnight (weekly, etc.) the AI algorithms run through all of the reports and flag any potential water intrusion issues.

The point: AI, in this case Autodesk Construction IQ, repurposes data already entered and used to draw new conclusions and alert humans to potential problems.

Autodesk Construction IQ is very specific right now, focused on quality and safety risks on a single job site as well as giving program managers an overview of risk potential across projects. It’s all based on data analytics that show up on heat maps (as per usual, red is bad) that identify specific items of risk. Is the same subcontractor always messing up the window cladding? What projects are they on? And so on. Project leaders are likely to use Construction IQ daily to check which subcontractors are flagged as high risk and to drill down to understand why the algorithm flagged these risk factors. Many may be flags that arose out of perfectly ordinary circumstances (late materials arrival meant a contractor was late in doing a task — flagged on the facts but not actionable because not their fault), while others need the human to take action. People with cross-project responsibility can track bigger trends, such as a contractor who has 1 safety violation on each project — not flagged in the specific project, but worrisome across them all.

So how does this work in real life? I talked with Michael Murphy, Digital Construction Operations Manager, BAM Ireland about why BAM decided to try Construction IQ, how the use it, and what he sees as reasonable next steps. Mr. Murphy works in the Irish subsidiary of Dutch Royal BAM Group, which prides itself on the quality of its builds, safety, sustainability and, of course, completing projects on time and on budget.

“We aim for right first time,” he told me, which is crucial in meeting the schedule and financial targets of each project, but also contributes to the company’s reputation — helping to secure future work. Part of achieving that goal is a transition from traditional workflows to a “build it (digital) before we build it (physical)’ mindset, which makes it possible to anticipate problems and brainstorm solutions before anything goes critical. BAM has figured out that, at each stage of a project, an error costs 10x what it did in the prior step — so a $20 mistake identified in design would cost $200 to fix in construction, or $2,000 at the end of construction — all the way to $200,000 once the building is in use.

Once on the job site, Construction IQ enables project managers to be proactive in identifying problems, and heading them off before they become crises. Construction IQ, says Mr. Murphy, enables his teams to focus on the issues that matter: “it highlights what’s truly important according to its algorithm, rather than what most recently walked into the supervisor’s trailer”.

BAM may be a bit ahead of the curve in terms of technology adoption, but this is true of all enterprises, everywhere: “We generate so much data in our projects”, Mr. Murphy says, “Construction IQ gives BAM the ability to interrogate this data, gain insights and leverage our field work and design data”.

Honestly, what I liked most about my time speaking with Mr. Murphy and his team was their description of how it works: “It’s just on. We’d gather the data anyway through our normal work procedures. Construction IQ presents the issues dashboard when we log in. It’s seamless — yes, there is AI in the background, but we don’t see it. We do see complex information, presented in a way that’s easy to understand — graphically, with colors and heat maps.”

I asked about how BAM’s field workers are taking to this, thinking that AI is far from most people’s comfort zone. It may be, but that doesn’t seem to matter. Mr. Murphy said that they’ve rolled out tablet apps to many foremen, who see benefits after just a bit of training. In exchange for logging issues digitally, they get access to change orders and other information, instantly. Employees and contractors, who are now being evaluated by Construction IQ and find themselves in red on a heat map, he says, are being motivated to do better.

Bottom line? Construction projects create vast amounts of data anyway –they have to– so why not use it to greater benefit? Finding and resolving issues is a supervisor’s job; if they can get a head start on identifying which issues matter, that’s only a positive.

We see the beautiful building or cutting-edge health care facility. Contractors like BAM see risk: cost, schedule, and safety can be affected by even the most minor incident. ML enables BAM to identify potential problems, be proactive rather than reactive and rely on data rather than intuition.

Outside of AEC? It’s coming — we just need to standardize on terminology and use cases. But learn from the BAM example: what data do you already have, that you could repurpose with ML, to gain new insights?

---

Many thanks to Mr. Murphy and the team at BAM Ireland for speaking with me, and to Autodesk’s AEC PR team for connecting us. The title image is from Autodesk, of a Construction IQ heat map showing safety issues by contractor name. (I believe it’s all made up, so don’t ascribe these values to a real entity with the same name.)