A true story from inside one of our BIM projects and the AI-powered tool it led us to build.
Sometimes a client hands you a problem nobody has solved before. That’s what happened to our team at Excelize. By the end of it, a computation that once took 140 hours almost six days of nonstop processing was finished in 9 minutes with the help of AI.
This is a story about a team that refused to accept “it can’t be done,” and used AI to build something that didn’t exist before to prove it.
The Problem: Finding Every Element Inside Every Room
The client’s requirement sounded simple on paper: identify every element associated with each room in a large BIM model, including elements hidden above ceilings and elements inside rooms having slanted walls.
Anyone who has worked in Revit knows why that’s harder than it sounds. Rooms in Revit aren’t neat boxes; they’re volumes shaped by walls, slabs, and ceilings. And Revit’s room-boundary logic is really only reliable for plain, straight walls forming simple rectangular rooms.
In this project, every single one of the 2,000 rooms was bounded by slanted walls. The very foundation Revit relies on to define a room’s shape was already working against the standard logic, before anyone even got to the harder part.
The first instinct was reasonable: “We can solve this with Dynamo.” And on a small R&D model, it worked beautifully.
A Clean Idea That Broke at Scale
The logic was simple: if an element intersects a room’s boundary, associate it with that room. Using Dynamo and Boolean geometry operations, the team built a workflow that checked every element against every room fast and accurately, on the demo model.
Then the real project arrived.
| Factor | R&D test model | Real project model |
| Rooms | A few dozen | ~2,000 |
| Elements | A few thousand | 250,000+ |
| Wall geometry | Plain, straight walls | Slanted, in every room |
| Special cases | None | Elements above ceilings |
Because every room broke Revit’s plain-wall assumption, the workflow had to fight the software’s own defaults just to get a usable boundary. Above-ceiling elements added another layer entirely. And checking 250,000+ elements against 2,000 slanted rooms turned a quick script into a computational marathon.
The 140-Hour Wall
The team eventually engineered a working method for slanted walls, ceiling voids, and all. But the core computation took approximately 140 hours, nearly six days of continuous processing.
In a live project, that’s not just slow; it’s also risky. What if the system crashes on the last day? If the model changes mid-run? If the process fails at 95%? Any one of these problems means starting over. A technically correct answer that takes six days to run isn’t really a solution; it’s a liability.
This kind of hidden inefficiency is more common than most firms realize. A joint Autodesk and FMI study of nearly 3,900 construction professionals found that bad or unusable project data caused an estimated $88.69 billion in avoidable rework in a single year.
The Real Shift Wasn’t Technical. It Was the Question.
Instead of asking “How do we make this faster?” the team asked something different:
Why are we solving this problem this way in the first place?
A script that had to be babysat for six days wasn’t the answer. What the problem actually needed was a proper tool, built once, running in minutes, on any room geometry, every time the model changed. So, the team set out to build one this time, with AI as a partner in the process.
Building Room Data Extractor
Building a dedicated tool from scratch requires specialized software development expertise that we don’t have in-house. Rather than waiting for someone else to do it, our team used AI as a hands-on collaborator throughout development to move faster on the parts that would normally take the longest:
- Exploring alternative algorithms for associating elements with slanted, irregular rooms
- Designing the overall structure of the tool
- Writing and refining the core logic
- Debugging issues as they surfaced on real project data
- Spotting performance bottlenecks and optimizing for speed at scale
The domain knowledge cannot come from AI. Understanding what the client needed, how Revit’s room boundaries behave, and what a correct answer even looked like on a 250,000-element model, that came entirely from the team’s own BIM and civil engineering expertise. AI gave that knowledge a fast, capable way to become real, working software.
The result is a tool; our team named it Room Data Extractor. The name itself suggests what it does: given a room, extract all the elements associated with it, regardless of shape.
Square room, slanted room, or any irregular-shaped room with elements tucked above the ceiling it doesn’t matter. Room Data Extractor was built to handle any geometry, because the real project had already proven that “plain walls only” wasn’t good enough.
It took roughly a week working with AI at every stage of development, testing, and validation to get from idea to a solution the team could trust on a live, 250,000-element model, tested repeatedly against real project data rather than the clean R&D model that had fooled everyone the first time.
The Result: 140 Hours Became 9 Minutes
| Metric | Before | After |
| Computation time | ~140 hours | ~9 minutes |
| Rooms processed | ~2,000 (all slanted) | ~2,000 (all slanted) |
| Elements checked | 250,000+ | 250,000+ |
| Geometry supported | Plain walls only | Any geometry |
| Reusability | One-off script | Reusable tool, delivered to client |
Even a conservative estimate of 10 minutes of manual review per room, across 2,000 rooms, would total roughly 333 hours before accounting for human error or model revisions. A nine-minute run, repeatable every time the model updates, replaced what could have been weeks of manual coordination.
Why This Matters More Than the Time Saved
Room Data Extractor wasn’t a one-time trick; it became a reusable capability. If the model changes, the tool runs again. If a similar requirement appears on another project, the capability already exists.
And the team gave the tool to the client, so they can run it themselves on future projects, and they are using it very happily. That decision wasn’t really about the tool. It was about the kind of relationship worth building: one where a client can bring a genuinely hard, unsolved problem to the table and trust that someone will build what’s needed to solve it.
The Real Lesson
It’s tempting to read this as “AI made things faster.” That’s true, but it’s not the real point. The real lesson: AI helped the team solve a problem it initially didn’t know how to solve by building something that didn’t exist before.
That required two things most teams have only one of: deep domain knowledge of the problem, and the ability to build the solution. AI didn’t replace any of them; it closed the gap between them, letting the team’s own expertise become working software in days instead of months.
McKinsey Global Institute estimates automation, and modern tooling could unlock roughly $228 billion in annual value for the U.S. AEC industry by 2030. The same shape of opportunity this project revealed at a much smaller, human scale: less time lost to rework, less risk from fragile processes, and more capacity to take on problems nobody else wants to touch.
We may not have the answer or the tool on day one. But we’re willing to stand in front of the hard problem, build what’s needed, and find one.
If at any time your team has a BIM requirement that feels like it’s never been done before, it’s usually an opportunity for you to make something most valuable. Excelize always loves taking on these kinds of challenges.
Frequently Asked Questions
How can a BIM computation be reduced from 140 hours to 9 minutes?
By rethinking how the process works, not just speeding up the existing method. A 140-hour Dynamo script was replaced with a purpose-built tool, Room Data Extractor, that handles room geometry directly instead of relying on repetitive, brute-force checks.
Why does Revit struggle with slanted or irregular room geometry?
Revit’s native room-boundary logic is optimized for plain, straight walls forming rectangular rooms. Slanted walls or irregular shapes make its default boundary detection unreliable, which is why standard workflows can break down or slow dramatically at scale.
What is Room Data Extractor?
A tool built to identify and extract every element associated with a room in a BIM model, regardless of geometry: square, slanted, or any irregular shapes. It was developed to solve a real project where Revit’s standard room logic couldn’t handle 2,000 rooms bounded by slanted walls.
How much does poor BIM coordination cost the industry?
A joint Autodesk and FMI study estimated that bad project data caused approximately $88.69 billion in avoidable rework in one year, while McKinsey found construction productivity grew only 10% between 2000 and 2022; it is far behind other industries.
Can a custom BIM tool really replace weeks of manual coordination work?
Yes. In this project, even 10 minutes of manual review per room across 2,000 rooms would total roughly 333 hours. A reusable tool completing the same work in minutes, re-runnable whenever the model updates, removes that recurring burden entirely.