Spatial AI is a different category of technology than the chatbots and text generators that dominate AI headlines. Built for geometry, physics, and building systems, it automates the preconstruction coordination work that currently costs contractors hundreds of hours per project. Real numbers from live projects show 50-60% reductions in modeling time. This piece breaks down what spatial AI actually does, why general-purpose AI can’t do it, and what contractors need to know before adopting it.
Construction’s productivity problem isn’t a secret. According to a 2024 McKinsey report, construction productivity improved by just 0.4% annually between 2000 and 2022, while construction prices rose 52% between 2015 and 2023. Two decades of near-zero productivity growth, in an industry building some of the most complex structures humans have ever attempted.
A lot of that drag lives in preconstruction – specifically in the coordination work that happens after the architects hand off drawings but before anyone breaks ground. Structural, mechanical, electrical, and plumbing systems all need to fit into the same physical space. They don’t always agree on who gets which ceiling cavity. Resolving those conflicts falls to VDC teams and trade contractors, and it eats time that most projects can’t afford to lose.
AI is now entering that coordination process. But not the AI most people are thinking of. Francesco Iorio, CEO of Augmenta, published a piece in Forbes this month that cuts through the noise and makes a clear case: the AI that works for construction coordination is spatial AI, and it requires a fundamentally different technical approach than the large language models powering tools like ChatGPT. Understanding that distinction matters if you’re evaluating where AI actually moves the needle on a project.
General-purpose AI struggles with geometry, physics, and the deterministic guarantees that physical structures demand. Spatial AI is purpose-built for exactly those constraints, and it’s already cutting hundreds of hours of modeling work from real projects.
Why Preconstruction Is Where Productivity Goes to Die
The coordination problem that software hasn’t solved
Every building system – structural, mechanical, electrical, plumbing – has to share the same physical space. The routing decisions those systems make are interdependent in ways that aren’t immediately obvious. What a structural engineer does to a beam affects where a plumber can run a pipe. Where a mechanical contractor routes ductwork determines what’s left for electrical raceways. These systems compete for limited room in ceiling cavities, wall chases, and vertical shafts, and the teams designing them often work in separate silos until late in the process.
The detailed coordination work – exact routing paths, clash resolution, optimizing for constructability and prefabrication – typically gets pushed downstream to contractors and field teams. That’s where the real manual modeling load accumulates. VDC teams, estimating, project management, and shop teams often end up doing redundant work on the same coordination problems because the handoffs between them are fragmented. Add larger, more complex projects with tighter budgets and shorter timelines, and the traditional process starts to buckle under its own weight.
Software tools built for this phase have struggled to make a dent. The coordination problems are too geometrically complex, too dependent on experience-based rules that are hard to codify, and too interconnected across systems for conventional software to handle at scale. That’s the gap spatial AI is trying to close.
What Makes Spatial AI Different
Why you can’t just point a large language model at a BIM file
Large language models are extraordinary at working with text. They can summarize, draft, translate, and reason over written content at speeds no human team can match. What they can’t do reliably is reason about three-dimensional geometry, respect the laws of physics, or make the kind of deterministic guarantees that a physical structure requires. When an LLM hallucinates a fact in a document, you can catch and correct it. When an AI hallucinates a routing path in a building system, you potentially have a collision between a structural beam and a mechanical duct that doesn’t get caught until it’s already built.
Traditional computational methods have the opposite problem. They’re mathematically precise but can’t handle the scale and nuance of a modern building’s coordination challenges. They don’t capture the fuzzy, experience-based rules that govern how trades actually work – the kind of judgment that a senior VDC manager has built up over 20 years on complex projects.
Spatial AI, as Iorio describes it, combines the probabilistic reasoning of modern machine learning with the deterministic guarantees required for physical structures. It’s trained on domain-specific datasets that encode deep understanding of building systems, construction physics, and trade intelligence. That combination – holistic reasoning plus mathematical correctness plus domain expertise – is what makes it useful for preconstruction coordination in a way that general-purpose AI simply isn’t.
“AI has the ability to do the routing and coordination across all systems, freeing human expertise for the complex engineering decisions that require experience and judgment.”Francesco Iorio, CEO of Augmenta, Forbes Technology Council
The Numbers That Matter
Real project data from early adopters
Iorio’s piece includes two concrete project examples from Augmenta customers. On a healthcare facility project, 216 hours of manual modeling were cut to 103 hours. On a science lab project, 125 hours came down to 57. That’s roughly a 50-55% reduction in both cases, with Iorio noting a clear path toward 70% as the technology matures. These aren’t projections from a product demo – they’re actual hours logged on live projects.
Put those numbers in context. The construction industry is already working through a workforce deficit that isn’t going away quickly. 41% of the current workforce is expected to retire by 2031. Teams are regularly putting in 14-hour days. The labor pool that does VDC coordination work is skilled, scarce, and expensive. Cutting 50-60% of the manual modeling load on complex projects isn’t just an efficiency win – for a lot of firms, it’s a capacity unlock that lets them take on more work without hiring staff they can’t find.
- MEP coordination. Automated routing across mechanical, electrical, and plumbing systems, with clash detection and resolution built into the process from the start rather than as a downstream fix.
- Prefabrication optimization. Outputs that are constructability-aware and ready for shop fabrication, reducing the rework that happens when VDC models get handed off to fabrication teams.
- Workforce leverage. VDC leads and trade experts spend less time on manual modeling and more time on the high-judgment coordination decisions that actually require their experience.
- Project scalability. Firms can take on larger and more complex projects without proportional increases in VDC headcount – a meaningful competitive advantage in a tight labor market.
The Bigger Picture: Infrastructure Demand Versus Labor Supply
Why this problem gets harder before it gets easier
The demand side of the equation is accelerating. By 2055, the US population is projected to reach 372 million, requiring substantial new infrastructure – hospitals, data centers, schools, climate-adapted buildings. That demand is going to be politically and economically difficult to ignore. The supply side, in terms of skilled construction labor, is moving in the opposite direction. More retirements, persistent recruitment challenges, and wage inflation that compresses margins even as order books grow.
That gap between demand and available human capital is what gives spatial AI its urgency. This isn’t a technology looking for a problem – it’s a tool being built to address a structural constraint that the industry can’t resolve through hiring alone. Iorio makes the point that spatial AI can also expand into on-site applications: digital twins for site mapping, robotics support for manual tasks, and safety monitoring. The preconstruction coordination use case is the most mature right now, but it’s part of a broader capability set that’s still developing.
For contractors thinking about where AI fits in their technology roadmap, the preconstruction coordination layer is worth serious attention. The productivity gains are measurable, the labor math is straightforward, and the technology is past the theoretical stage. We’ve written before about how AI adoption in construction tends to target incremental improvements rather than systemic change – spatial AI applied to coordination dysfunction is one of the more credible cases for genuine systemic impact.
How to Actually Adopt This
A practical starting point for VDC and operations teams
Iorio outlines a three-step framework for contractors looking to adopt spatial AI without getting burned by a poorly scoped first deployment. The starting point is identifying where coordination dysfunction hits margins hardest. That varies by project type and trade mix, but the principle is consistent: solve the hardest, most expensive coordination problem first. Proving the technology’s value on a genuinely difficult problem builds more internal credibility than a pilot on something that wasn’t causing much pain to begin with.
The second step is framing the adoption correctly for VDC teams. The risk with any automation tool is that experienced staff read it as a threat to their jobs. Spatial AI doesn’t replace the judgment of a skilled VDC lead – it removes the manual modeling work that sits beneath that judgment. The framing that matters is capacity and reach: the same team can handle more projects, more complex coordination, and more demanding clients. That’s a different conversation than “the AI does your job now.”
The third point is the most technically important: ensure the AI’s outputs are optimized for how construction actually works, not just for geometric correctness. That means procurement logic, prefabrication constraints, and constructability requirements need to be embedded in the model’s output from the start. A routing solution that looks clean in BIM but can’t be prefabricated efficiently hasn’t solved the problem – it’s just moved it.
Get the ConTech intelligence your competitors are reading
AI, M&A, funding rounds, and the deals reshaping construction – straight to your inbox every week.
Join 2,500+ ReadersSpatial AI vs. General-Purpose AI in Construction: A Quick Comparison
| Capability | Large Language Models (e.g. GPT-4) | Traditional Computational Tools | Spatial AI |
|---|---|---|---|
| 3D geometric reasoning | Weak – not designed for spatial tasks | Strong – mathematically precise | Strong – purpose-built for geometry |
| Physics / constructability awareness | Unreliable – prone to hallucination | Moderate – rule-based, limited nuance | Strong – trained on domain-specific data |
| MEP coordination and clash resolution | Cannot perform at required fidelity | Partial – handles rules, not judgment | Core use case |
| Prefabrication optimization | Not applicable | Limited | Embedded in output |
| Scale on complex projects | Can handle documents, not building systems | Breaks down at scale | Designed for large, complex buildings |
| Deterministic guarantees | No – probabilistic by design | Yes – but at the cost of flexibility | Yes – combines both approaches |
| Proven time savings on live projects | N/A | Marginal | 50-60% reduction in modeling hours reported |
Frequently Asked Questions
Spatial AI is purpose-built for tasks involving three-dimensional geometry, physics, and the kind of deterministic output that physical structures require. General-purpose large language models like ChatGPT are trained primarily on text and excel at language tasks – drafting, summarizing, reasoning over written content. They’re not designed for geometric reasoning and can produce errors (hallucinations) that are harmless in a document but dangerous in a building system. Spatial AI combines probabilistic machine learning with the mathematical correctness that construction coordination demands. (Source: Forbes Technology Council)
The primary application right now is MEP (mechanical, electrical, and plumbing) coordination – specifically the routing, clash detection, and resolution work that determines exactly how those systems fit within a building’s physical constraints. Spatial AI automates the manual modeling that VDC teams and trade contractors currently do by hand, producing outputs that are constructability-aware and ready for prefabrication. It doesn’t replace the high-judgment decisions that experienced engineers and VDC leads make – it removes the time-intensive manual work that sits beneath those decisions. (Source: Forbes Technology Council)
Augmenta, one of the leading spatial AI companies in construction, has published data from two customer projects. A healthcare facility project saw modeling hours drop from 216 to 103 – a reduction of roughly 52%. A science lab project went from 125 hours to 57, a reduction of about 54%. Iorio notes that many contractors are seeing 60% reductions in modeling and coordination time, with a clear path toward 70% as the technology develops further. These are live project figures, not projections from a controlled trial. (Source: Forbes Technology Council)
The more accurate framing is that it changes what VDC staff spend their time on. Manual modeling and clash resolution are tasks that consume a significant portion of a VDC team’s hours on complex projects. Spatial AI automates those tasks, which frees experienced VDC leads and trade experts to focus on the strategic, high-judgment coordination work that the AI can’t do. Given that 41% of the construction workforce is expected to retire by 2031, the industry’s actual problem is a shortage of skilled workers – spatial AI is more likely to be a tool for doing more with existing teams than a driver of job losses. (Source: Forbes Technology Council)
The most important question is whether the tool’s outputs encode real trade intelligence – procurement logic, prefabrication constraints, and constructability requirements that reflect how your trades actually work. A spatial AI that produces geometrically clean routing solutions but ignores how those systems get fabricated or installed has missed the point. Beyond that, look for evidence of performance on projects similar in scale and complexity to yours. Published case study data (actual hours, not percentage claims alone) is a reasonable minimum bar. We’d also suggest reading our analysis of where AI adoption in construction tends to go wrong before scoping a pilot.
Healthcare facilities, science labs, data centers, and other technically complex building types see the most immediate gains – these are the projects where MEP coordination is densest, clash resolution takes the longest, and manual modeling hours accumulate fastest. Simpler project types still benefit, but the ROI case is strongest where coordination complexity is highest. As the technology matures, the benefits are expected to extend further down the complexity scale. (Source: Forbes Technology Council)
Spatial AI sits in a different category from the general-purpose AI tools (chatbots, document summarizers, spec analyzers) that are getting most of the attention in construction tech right now. Those tools have real utility, but they don’t touch the core productivity problem in preconstruction. Spatial AI does. It’s also worth watching how it intersects with spatial data capture tools – we’ve covered the OpenSpace-Disperse merger and its implications for spatial AI in construction separately. The combination of rich site data and AI-driven coordination could be more impactful than either capability on its own.
Related Articles
“We Need a Moonshot” – Why AI in Construction Is Focused on the Wrong Problem
Inside OpenAI’s Construction Power Move
