Best of LinkedIn: Digital Construction CW 22/ 23

00:00:00: This episode is provided by Thomas Allgaier and Frennus, based on the most relevant LinkedIn posts about digital construction in calendar weeks twenty-two and twenty three.

00:00:09: Frenness is a B to B market research company that supports industrial automation and ICT companies with Market Intelligent across the construction industry to prioritize segments identify high value accounts and validate use cases.

00:00:23: you can find more info.

00:00:26: So welcome in, everyone.

00:00:27: If you're joining us today we have a really exciting deep dive lined up for you.

00:00:32: We are cutting through all the noise online to explore The absolute top digital construction trends that Are just buzzing across the industry right now.

00:00:41: Yeah and there's A lot of noise out There Right Now

00:00:43: Exactly.

00:00:44: And our mission Today Is Really To Just Break Down What'S Actually Working.

00:00:47: But Before We Get Into All The Flashy Tech We Can Have To Talk About The Messy Reality Of What Actually Feeds It.

00:00:53: Yeah...the data I mean.

00:00:55: imagine you've got this massive construction firm, right?

00:00:58: And they're sitting on a decade of golden project data.

00:01:01: Oh yeah the holy grail

00:01:03: Literally The Holy Grail for future bidding.

00:01:06: talking real material costs Precise labor hours highly specific supply chain timelines everything You'd need.

00:01:13: but then you know A junior estimator comes in Asks to see that data to price out at new high-rise and a manager just points To a guy down the hall and says ask Dave Dave remembers the old numbers.

00:01:24: Oh man, ask Dave!

00:01:26: I mean that is just... That's a terrifying business model

00:01:29: It really is.

00:01:30: and The Kicker?

00:01:31: Dave is retiring next month?

00:01:33: Of course he

00:01:33: is.

00:01:33: Yeah This was an actual scenario shared recently by Victor Agustio And it perfectly captures the biggest issue we have to tackle first

00:01:41: Which is this terrifying reliance on Dave?

00:01:44: Right We were treating millions of dollars completed project data as Dusty totally inaccessible archive instead of you know, a strategic operational asset.

00:01:53: exactly

00:01:53: the Dave problem is just A symptom of a much larger structural crisis in The industry.

00:01:59: Mathieu Lefebvre recently spoke about this.

00:02:02: He was referencing insights from the Bentley illuminate conference.

00:02:04: Oh right?

00:02:04: Yeah and You Know the construction Industry Is staring down This massive wave Of infrastructure investment.

00:02:10: but There's his widening Structural gap as all these seasoned engineers retire.

00:02:15: And Matthew pointed out this staggering statistic, roughly ninety-five percent of the information an infrastructure asset produces over its lifetime is considered dark data.

00:02:25: Ninety-five percent?

00:02:28: Ninety five percent.

00:02:29: That is wild, so it just vanishes into the void.

00:02:31: and you know for you listening when we say dark data We aren't talking about like mysterious computer code No.

00:02:37: We're talking about hundreds of thousands of unstructured PDFs daily field notes Just buried in some random server Outdated CAD files and closed RFIs.

00:02:47: right those formal requests for information that clarify plan discrepancies?

00:02:50: Yeah, there's massive context just rotting in a hard drive

00:02:53: exactly.

00:02:54: And okay let's unpack this for a second.

00:02:56: If a firm's data is essentially a disorganized junk drawer, aren't we just setting our AI up to completely fail?

00:03:03: Oh absolutely!

00:03:03: I mean it's like trying to run a high-performance sports car on dirty unrefined fuel right or building a fifty story skyscraper on a foundation of loose gravel.

00:03:14: the steel framing might be state-of-the art but

00:03:18: That structural analogy is spot-on.

00:03:20: I mean, Satish Durakarjhan made this exact argument when he was observing how organizations are currently spending

00:03:26: just

00:03:27: vast amounts of time and money evaluating AI use cases.

00:03:30: Oh

00:03:30: everyone wants the flashy new toy

00:03:32: Exactly!

00:03:33: They want predictive analytics they want automated scheduling but their completely struggling to get meaningful outcomes Because AI does not create clarity from chaos.

00:03:42: Right,

00:03:43: it just doesn't?

00:03:43: No!

00:03:44: Yeah

00:03:44: If you feed a large language model conflicting spreadsheets and like outdated structural codes It doesn't intuitively know which document is

00:03:51: correct...It just amplifies that existing chaos Exactly

00:03:55: And confidently hands the wrong number Just much faster than human would Which is

00:03:59: terrifying, and it means the actual competitive advantage here has nothing to do with buying The Flashiest New AI tool.

00:04:06: It's entirely about who owns that foundational data layer?

00:04:10: And how clean they can keep it right.

00:04:12: like Peter Mitrov highlighted a National Institute of Standards & Technology Estimate That puts A huge price tag on this!

00:04:18: The cost Of poor interoperability To the US Construction Industry alone Is fifteen point eight billion dollars annually.

00:04:26: Wow Fifteen

00:04:28: billion annually.

00:04:29: interoperability is basically the software equivalent of plumbing.

00:04:33: if you're estimating Software cannot seamlessly flow data into your project management software.

00:04:38: The data just dies in transit.

00:04:40: Exactly, it's just dies

00:04:41: and solving that flow of data requires a pretty fundamental shift In how we interact with our own documents.

00:04:47: Goulnor shared really highly tactical solution regarding this.

00:04:50: Oh right?

00:04:51: With the dense PDFs?

00:04:52: yeah Go observe the nightmare team's experience when they try to use standard AI, just chat with massive construction documents.

00:05:00: Right because if you take a four hundred page steel detailing manual or some super dense local fire code and dump it into a standard AI prompt window... The whole system breaks down.

00:05:11: Exactly.

00:05:12: Because of token waste right?

00:05:13: Yeah!

00:05:13: Token Waste.

00:05:14: Because large language models process text in chunks called tokens And have a strict memory limit.

00:05:20: If you stuff a massive manual into the prompt, The AI essentially forgets middle pages by time it reaches end.

00:05:26: Right

00:05:27: and then it starts hallucinating clauses or inventing safety standards that don't even exist

00:05:31: Which is obviously an unacceptable risk in construction.

00:05:34: Exactly So Gatal stopped treating these dense documents as mere inputs for a chatbot.

00:05:41: Instead they restructured this documentation to what we call on-demand AI skills.

00:05:46: Okay

00:05:47: I love framing.

00:05:48: They vectorized data meaning broke the documents down into mathematical representations so that AI can search for conceptual matches rather than just looking at keywords.

00:05:57: Oh,

00:05:57: that's smart!

00:05:57: Yeah...

00:05:58: So now when an engineer asks you know which fire reading spec applies to this specific drywall assembly The system uses retrieval augmented generation To pull only exact relevant paragraph in the AIs memory.

00:06:13: So no token waste?

00:06:14: Exactly And no generic hallucinated guesses.

00:06:17: That is massive, and honestly it's a perfect pivot to our next theme.

00:06:21: because if organizing that underlying data Is step one Step two is figuring out how To actually embed those tools into daily operations

00:06:29: without causing A revolt on the job site.

00:06:31: exactly.

00:06:32: Elliot Christensen wrote a brilliant piece arguing that for most of his career Construction tech was treated as an entirely separate entity from construction Operations.

00:06:40: yeah It was always the new iPad after The field teams were just forced to use

00:06:43: right.

00:06:44: But Elliott argues we have to stop treating AI as a side experiment.

00:06:48: It has to be woven in, you know designed to reduce repetitive work without simply ballooning the back-office headcount.

00:06:55: And integrating it is an operation strategy requires complete mindset shift and procurement deployment like Chaitanya Buresh shared framework for this.

00:07:06: that really clarifies operational hurdle.

00:07:14: It is a new hire.

00:07:16: If you think about hiring a brand-new junior engineer, You'd never hand them complex stack of blueprints on day one and just say run the project!

00:07:25: Right...you would be setting up to fail.

00:07:27: Sit down with your specific company manuals show how their work breakdown structure categorizes tasks And closely review there for several weeks.

00:07:37: and AI requires that exact same rigorous onboarding process.

00:07:41: Yeah A firm needs to invest hundreds of hours formatting historical cost codes so the AI can actually read them, And senior estimators need to review the AI's first fifty outputs To provide a real human feedback loop.

00:07:54: The firms pulling ahead aren't looking for an out-of-the box miracle.

00:07:58: They're putting in the grueling work.

00:08:03: Nate

00:08:08: Feller brought up some pretty sobering research from Princeton University regarding frontier AI models.

00:08:12: Okay, we often see these models succeeding wildly on academic benchmarks.

00:08:16: right completing complex logic tasks autonomously.

00:08:20: but construction is a vastly more difficult environment than pure software.

00:08:24: Oh definitely you're dealing with multi-party delivery methods highly fragmented data silos massive contractual complexity

00:08:31: exactly and they pointed out that there Is currently A fifty X cost cap.

00:08:35: wait fifty times

00:08:38: between the highly accurate AI agents that perform well in academic settings and The affordable AI agents you would actually deploy for everyday construction workflows.

00:08:47: So an agent sophisticated enough to accurately navigate a multi-party Construction contract without making catastrophic errors carries.

00:08:55: A massive

00:08:56: premium.

00:08:56: yeah, I'm massive premium.

00:08:57: That's not

00:08:58: just an academic challenge procurement reality for any builder trying to scale this technology across a whole portfolio of projects.

00:09:07: It's a massive hurdle, however and here is where it gets really interesting when a firm does manage to deploy sophisticated AI effectively the results are just staggering.

00:09:18: yes let's talk about lisha longs post.

00:09:22: He highlighted this massive win from Suffolk Construction using Alice Technologies' targeted optimization AI on a delayed life sciences project.

00:09:30: This is such a perfect example of moving beyond basic AI, because in traditional construction scheduling teams use the critical path method.

00:09:37: it's a linear very rigid sequence.

00:09:40: if a concrete pore is delayed by a week Every single dependent task down the line shifts by a week, creating what is known as negative float.

00:09:49: Right!

00:09:49: The amount of time.

00:09:50: that project is now officially behind schedule.

00:09:52: Exactly and normally human scheduler has to stare at a forty two thousand line Gantt chart And manually guess different scenarios To try recover those lost times Which takes days and super prone error.

00:10:04: But using targeted optimization the AI doesn't just shift lines on a chart.

00:10:08: It runs massive simulations across thousands of resource allocations.

00:10:12: On this Suffolk project, the AI analyzed entire complex web dependencies and surfaced exactly three specific trade tasks where just a thirty percent productivity lift would completely eliminate all remaining negative float.

00:10:26: That is incredible.

00:10:27: By finding those non-obvious needles in the haystack, Suffolk successfully recovered forty two days on the schedule.

00:10:35: Forty Two Days!

00:10:35: In an industry where a single day of delay triggers massive liquidated damages.

00:10:40: recovering forty two.

00:10:48: If your back office is suddenly using optimization algorithms to churn out schedules and estimates twice as fast, it really seems like you're field teams are going be hit with a massive pipeline of physical work.

00:10:58: They just do not have the headcount actually build.

00:11:00: Oh wow yeah that's exactly danger.

00:11:03: Robert Dong provided very sharp reality check on this dynamic.

00:11:06: Yeah he did.

00:11:06: He noted that AI might solve the Back Office bottleneck making estimators and planners hyper efficient but does not eliminate friction in building.

00:11:15: It shifts the bottleneck

00:11:17: Right, AI tools cannot hang drywall.

00:11:20: They can not weld pipe

00:11:21: Exactly and the industry is already suffering from a historic labor shortage.

00:11:26: Alternatively The bottleneck shifts to the business development team who now suddenly have to win twice as much work just To keep this hyper efficient operational machine

00:11:35: fed?

00:11:37: right?

00:11:37: And that collision between a hyper-efficient digital back office and the messy physical reality of the job site Leads us directly into the next theme The architectural challenges of BIM and digital twins.

00:11:50: Oh, this is the

00:11:50: big one!

00:11:51: Yeah because bridging the gap between a pristine three-D model And that dirt on the ground Is basically the defining technical challenge for decades For sure.

00:11:59: And Florian Humor sounded massive alarm regarding This warning that ninety percent of enterprise scale Digital twin architectures are going to choke and fail the second they hit real world production.

00:12:08: Ninety

00:12:09: percent failing in production?

00:12:10: And the reason it fails so reliably is rooted In how their actually built.

00:12:14: right.

00:12:15: Florian argues that teams are designing these digital twins like standard web applications.

00:12:20: Exactly, and in a normal web app the user clicks a button or request goes to a database And the data is returned.

00:12:26: very simple.

00:12:26: But a digital twin of a commercial building Is getting bombarded with thousands Of continuous real-time signals from IoT sensors.

00:12:34: you know temperature HVAC pressure occupancy levels.

00:12:38: Right, and if you try to pipe that raw unfiltered fire hose of data directly into a heavy three D virtual simulation the server simply collapses under its own weight.

00:12:48: it just

00:12:48: crashes.

00:12:49: To prevent that crash Florian points out dual pipeline engine.

00:12:54: You have to physically separate the workloads?

00:12:56: Yeah,

00:12:56: you have two.

00:12:57: one pipeline handles The physical data ingestion using edge computing To clean and filter out the noise of those iot sensors before it even reaches the main system.

00:13:06: right.

00:13:06: And then the second pipeline handles the heavy three d virtual simulation.

00:13:10: and Those two pipelines only meet in the middle at a high-performance Computing layer

00:13:15: which is so crucial.

00:13:17: if you don't build that dual architecture from day One your digital twin is dead on arrival

00:13:22: Dead On Arrival.

00:13:23: And Antonio Cianzuli built directly on this technical hurdle, looking specifically at geospatial digital twins.

00:13:30: Oh right when you scale up?

00:13:32: Yeah!

00:13:32: When you scale-up from a single building to entire city blocks by combining OpenBIM and GIS geographic information systems.

00:13:40: Yeah...and OpenBim relies on universal file format so different software ecosystems can actually read the same.

00:13:48: But when you aggregate an entire district into a common data environment or CDE, You are no longer managing a few models.

00:13:55: No!

00:13:56: You're asking a single central repository to render millions of federated models and billions of geometric objects simultaneously

00:14:03: Billions.

00:14:04: And if that common data environment is not engineered to natively handle that sheer polygon count and data weight, the latency becomes unbearable.

00:14:12: Oh totally!

00:14:12: An a digital twin that takes five minutes to load a room?

00:14:15: Is it a Digital Twin that the facility manager will simply refuse use

00:14:19: exactly which raises a critical question for Mark Queenen ahead of Digital Construction Week.

00:14:24: Mark asked if we are overproducing information at the handover phase.

00:14:28: That's a really good question

00:14:29: because We've gotten incredibly proficient at generating massive information dense BIM models.

00:14:36: But when we hand over the keys to the operations and maintenance teams Are we just handing them?

00:14:41: A massively complex digital artifact they literally cannot consume?

00:14:46: And the answer is often yes.

00:14:47: That disconnect between the theoretical capability of a digital tool and practical usability for human beings starts long before building is finished.

00:14:56: Yes,

00:14:56: it starts in training.

00:14:57: Exactly!

00:14:59: Milgin and Djelkovic shared an interesting observation about this – he actually left his position at one Europe's largest engineering firm, Suiko Architects specifically to teach Revit full-time.

00:15:10: Wow.

00:15:10: Yeah, he made that drastic career shift because he recognized a massive gap between knowing a software tool and possessing actual office-ready skills.

00:15:20: Right is the difference between knowing where the draw wall button Is versus knowing how to structure complex project template so it doesn't crash when fifty different architects are working on at same time.

00:15:29: Exactly Milgen noticed half of people in The Office theoretically knew Revit But when asked to produce clean, coordinated documentation that wouldn't require ten rounds of fixes they were completely lost.

00:15:43: Knowing the software features is entirely different from understanding workflow

00:15:47: Which brings up ultimate limiting factor for everything we have discussed today.

00:15:51: Yes our final theme The people.

00:15:54: Exactly You can have most elegant dual pipeline architecture and optimized AI schedule in world but it's entirely useless if workforce not supported capable and willing to use it.

00:16:06: Five percent!

00:16:07: John N. shared research from Columbia University's Dr.

00:16:10: Ibrahim O'Day, who spent a year studying how major architecture engineering and construction firms adopt generative AI.

00:16:17: Okay... Dr.

00:16:18: O'day found that seventy-percent of successful digital transformation comes down entirely to people culture and capability.

00:16:25: the technology itself accounts for Maybe ten to fifteen percent of the equation.

00:16:29: Ten to fifteen per cent?

00:16:30: That figure is wild when you consider that technology consumes almost all the oxygen in industry.

00:16:35: discussions and focusing on people brings us face-to-face with this severe skilled labor shortage.

00:16:42: Nicholas Johnson shared a powerful perspective on this, The industry spends massive amounts energy trying recruit more young men into trades.

00:16:51: but women currently represent only about eleven percent total construction workforce And when you look specifically at the physical trades, that number plummets to just over four percent.

00:17:02: It

00:17:02: is a massive systematically untapped talent pool.

00:17:06: Yeah construction is rapidly transitioning away from being purely a brute-strength endeavor.

00:17:12: it's a highly complex knowledge industry now requiring immense logistical coordination technology management dynamic problem solving Exactly.

00:17:20: If the industry wants to solve the capacity issue and run these advanced digital systems, significantly increasing the representation of women is one of the most obvious and underutilized solutions available.

00:17:30: Absolutely!

00:17:30: And regardless who we bring through this door... We have to fundamentally rethink how we develop that young talent.

00:17:37: Rachel Manna observed that career progression and construction is notoriously slow, especially when compared to the tech sector like the FA in companies.

00:17:45: Oh

00:17:45: for sure

00:17:46: Young engineers in construction often get siloed into incredibly narrow scopes For years spending two years doing nothing but detailing doors or tracking elevator submittals with zero exposure To the broader business mechanics Or the overarching digital strategy

00:18:02: right?

00:18:02: And the firms that will capture and retain The next generation of builders are the ones offering faster learning velocity, genuine mentorship and actual ownership over outcomes.

00:18:13: And that holistic support system is vital for career progression but it also extends to literal survival on the job site.

00:18:20: Zulkarnane Malek brought forward a deeply humanizing point regarding mental health.

00:18:24: we absolutely cannot ignore here.

00:18:27: The statistics in this space are just brutal.

00:18:29: Seventy-five percent of construction workers will not report mental issues to their supervisors?

00:18:35: Tough guys break, too.

00:18:37: They just break silently

00:18:38: and the cost of that silence is devastating.

00:18:40: It materializes in higher on-site injury rates missed days Substance abuse and workers walking off the site permanently.

00:18:48: Yeah

00:18:48: Construction is the second highest industry for suicide in Canada.

00:18:53: No amount of AI scheduling optimization or pristine digital twins can solve this.

00:18:58: Fixing a culture of silence requires a foreman having the situational awareness to look a worker in the eye, ask how they're actually doing and then have patients wait for their real

00:19:07: answer.".

00:19:11: And if we synthesize everything we have explored today from Victor's retiring estimators to Florian's dual pipeline twins, it is clear that data foundations and AI will completely rewrite the mechanics of how we build.

00:19:22: The tools are scaling exponentially... They

00:19:24: really are!

00:19:24: But your ultimate competitive advantage in next decade won't be the software license you purchase.

00:19:29: Exactly

00:19:30: because when the dust settles every single firm has access to exact same generative AI and the exact same rendering engines ...the only variable left would be your people

00:19:39: Right

00:19:40: The most valuable asset your firm will possess is the human intuition required to guide those models, the culture of learning you foster and the frontline workers you choose to fiercely protect and empower.

00:19:52: At the end of day algorithms don't build projects.

00:19:56: people do.

00:19:57: I couldn't agree more.

00:19:58: If you enjoyed this episode.

00:19:59: new episodes drop every two weeks.

00:20:02: also check out our other editions on smart manufacturing in digital power tools.

00:20:06: Thank you so much for joining us on this deep dive.

00:20:08: Don't forget to hit

00:20:14: subscribe,