Steel estimators spend days hand-counting beams and columns across stacks of drawings, and one missed item can wipe out the profit on a job. That problem is why I built SteelFlo.
It is also the clearest example of how I build with AI for clients. Here is the problem, how the first slice came together, and three lessons that apply to any AI product, including yours.
The problem
Steel takeoff is the work of reading a set of drawings and counting every beam, column, angle and plate needed for a job. It is careful, repetitive work, and estimators often do it by hand across dozens of sheets for every bid. Missing an item can cost a shop real money.
Starting with the estimator's workflow
The temptation with AI is to start with the model. I started with the estimator instead. What do they open first? What do they write down? What do they double-check? That workflow became the product's path.
The first slice was simple: upload drawings, extract members and quantities, review them, export them for pricing.
This is the same process I use with every client: real workflow first, AI second, review step always.
Book a free scoping call →Lesson 1: trust beats automation
An AI that is right most of the time is not useful if the estimator cannot tell which parts are wrong. The review screen, where every extracted item can be checked against the drawing and corrected, matters as much as the extraction itself.
Lesson 2: real inputs, early
Drawings vary a lot between engineers and firms. Testing on real drawing sets from the start exposed problems no clean sample would have shown.
Lesson 3: measure accuracy like a product feature
I treat accuracy as something to measure on every change, using a growing set of real examples. That way improvements are proven, not guessed.
What this means for your MVP
If your product uses AI, the same rules apply:
- Build the AI into a real user workflow, not a demo.
- Give users an easy way to check and fix results.
- Test on messy, real inputs as early as week one.
Read more on adding AI to your MVP.
Frequently asked questions
What is steel takeoff?
Steel takeoff is the process of reading construction drawings and listing every steel member and quantity needed for a job, so a fabricator can price and order material.
Can AI read construction drawings accurately?
AI can extract a large share of the information from drawings, but results need human review. The best tools make that review fast instead of trying to remove the human entirely.
Who built SteelFlo?
SteelFlo was founded and built by Audrey Graham, who also runs Thin Slice.
Build your AI product the way SteelFlo was built
If you have a workflow that is slow, manual and full of documents, AI can probably take most of it off your plate. Book a free call and walk me through it.
- Free 30-minute call
- Fixed price from $7,500
- Live in four weeks
- You own the code
I take on two to three builds at a time, so start dates fill up.