Good Demos Are Easy. Good Tools Are Rare.
NASCC 2026 has more AI vendors on the show floor than any previous year. At least three companies are showing AI-powered takeoff tools, Trimble is embedding AI into Tekla, and ConnectionAI is demonstrating AI-driven connection design. The pitches will be polished. The demos will be impressive.
The problem: demo PDFs are cherry-picked. Every vendor shows drawings their tool handles well. The drawings that matter are yours — the messy ones, the raster scans, the international standards, the 40-page packages with mixed detail levels.
Here are five questions that cut through the demo and tell you whether an AI tool will actually work in your shop.
1. "Can I Run This on My Own Drawings Right Now?"
This is the single most important question. If the answer is anything other than "yes," be cautious.
Why it matters: AI takeoff accuracy depends heavily on drawing quality, notation style, and how the tool handles edge cases. A demo on a clean, vector-text PDF with standard AISC callouts proves nothing about how the tool handles your actual drawings — which might be scanned, rotated, hand-lettered, or use non-US notation.
What to bring: Three recent bids on a USB drive. Pick drawings that represent your real work — not your cleanest jobs, but your typical ones. Include at least one that gave your estimator trouble.
Red flags:
- "We'd need to set that up on our end" (they can't run it live)
- "Our demo set shows the full capability" (they're avoiding your drawings)
- "We'll process those and get back to you in 24-72 hours" (it's a service, not a tool)
Green flags:
- They hand you the keyboard and let you upload
- Results come back in minutes, not days
- They're confident enough to show raw results, not just the summary
2. "What Happens When the AI Gets It Wrong?"
Every AI tool will make mistakes. The question is how the tool handles errors and how easy it is for your estimator to correct them.
Why it matters: An AI tool that's 90% accurate but gives you no way to verify or fix the other 10% is worse than a manual takeoff. The value of automation isn't eliminating human judgment — it's reducing the tedious identification work so your estimator can focus on the decisions that require experience.
What to look for:
- Can you see exactly where on the drawing each detection came from? (bounding boxes, page references)
- Can you reject false positives with one click?
- Can you add members the AI missed?
- Is there a confidence score so you know which detections to double-check?
Red flags:
- "Our accuracy is 98%" (accuracy claims without methodology are marketing, not data)
- You get a spreadsheet with no link back to the source drawing
- There's no way to override or correct individual detections
Green flags:
- Every detection is traceable to a specific location on a specific page
- The tool explicitly flags low-confidence items for review
- Your estimator stays in control of the final numbers
3. "Does It Know What a W12X26 Weighs?"
This sounds like a trick question, but it's the fastest way to find out whether a tool has real steel intelligence or is just a generic measurement platform with AI bolted on.
Why it matters: Detecting that something on the drawing says "W12X26" is step one. Knowing it's a wide flange beam that weighs 26 lbs/ft, classifying it correctly in a bill of materials, and looking up its depth, flange width, and web thickness — that's the intelligence a steel fabricator actually needs.
What to ask specifically:
- Does the tool have a built-in steel section database?
- How many sections are in it? (AISC alone has 550+)
- Does it include section weights, or just identification?
- What happens with less common sections — does it flag unknowns or silently skip them?
Red flags:
- "You can add weight data in post-processing" (the tool doesn't know steel)
- The tool identifies W12X26 but can't tell you its weight or category
- No mention of section properties anywhere in the product
Green flags:
- Built-in weight lookup that fires automatically at detection time
- Sections categorized by type (beam, column, HSS, channel, angle, plate)
- The tool generates a BOM with weights, not just a list of labels
4. "What Standards Do You Support Beyond AISC?"
If your answer to this question is "we only bid US work, AISC is all I need" — fair enough, skip this one. But if you've ever seen an HEA200, a UC305×305×158, or a 310UB40.4 on a drawing, this question matters a lot.
Why it matters: Structural steel is an international industry. UK engineers specify UC and UB sections. European firms use HEA, HEB, and IPE profiles. Australian fabricators see 310UB and 150PFC. A tool that only speaks AISC locks you out of any project designed outside the US.
What to ask:
- Which steel naming standards does the tool detect? (AISC, EN, BS/IS, AS/NZS)
- Does it auto-detect the standard, or do you have to configure it manually?
- Does it have weight databases for international sections, or just AISC?
- Can it handle mixed-standard drawings (e.g., UK drawings with both UC and IPE sections)?
Red flags:
- "We focus on the US market" (they only support AISC)
- "Our tool is standard-agnostic" (it's a manual tool with no detection)
- No answer at all — international standards aren't on their radar
Green flags:
- Multiple standard support with auto-detection
- Standard-specific weight databases (not just AISC converted)
- Tested on real international drawings, not synthetic examples
5. "What Does It Cost Per Takeoff at My Volume?"
AI tools price differently — per takeoff, per sheet, per seat, flat-rate monthly, or annual enterprise contracts. The sticker price tells you nothing. You need to calculate cost per takeoff at your actual monthly volume.
Why it matters: A tool that costs $199/month but limits you to 5 takeoffs is $40 per takeoff. A tool that costs $599/month with 10 takeoffs is $60 per takeoff. A per-takeoff service at $200/takeoff with no monthly commitment might be cheaper at low volumes and wildly expensive at high volumes.
What to calculate:
- Your typical number of steel bids per month
- The tool's pricing at that volume (monthly, annual, per-takeoff)
- Your current estimating labor cost per takeoff (hours × hourly rate)
- Net savings: labor cost saved minus tool cost
Red flags:
- Pricing requires a demo call (they're qualifying your budget, not being transparent)
- Per-takeoff pricing with no volume cap — costs scale linearly with your bid volume
- "Contact sales for enterprise pricing" on features you need for basic usage
Green flags:
- Pricing on the website, no demo required
- Flat-rate plans with clear volume tiers
- Free trial so you can test before committing
The Evaluation Framework
After NASCC, you'll probably have 2-3 tools worth serious evaluation. Here's how to run the comparison:
- Same three drawings through each tool. Pick one clean drawing, one messy one, and one large package.
- Compare against your own takeoff. You already know the right answer — how close does each tool get?
- Time the full workflow. Not just detection time — include review, correction, and export.
- Calculate total cost. Monthly fee plus your estimator's review time. The tool with the lowest total cost per takeoff wins.
- Check the audit trail. Can you defend every quantity to a GC by pointing to a specific location on the source drawing?
The vendors who are confident in their product will encourage this process. The ones who aren't will try to keep you in the demo environment.
What AI Takeoff Tools Are at NASCC 2026?
At least three: Beam AI (Booth #532), SketchDeck.ai LIFT (Booth #1127), and Ferra (private beta). Each has a different model — AI-assisted service, PDF detection, and steel-native platform respectively. See our full NASCC 2026 AI preview for details on every AI vendor on the floor.
Should I Test AI Tools on My Own Drawings?
Absolutely. Bring three recent bids on a USB drive — a clean one, a messy one, and a large package. Any vendor confident in their product will run your drawings live. If they won't, that tells you something.
How Do I Compare AI Takeoff Tools?
Run the same three drawings through each tool. Compare results against your own takeoff. Time the full workflow including review and correction. Calculate total cost per takeoff (tool fee plus estimator review time). Check whether every detection links back to the source drawing.
What's the Most Important Feature in an AI Takeoff Tool?
Source traceability — the ability to see exactly where on the drawing each detection came from. Without it, you can't verify the numbers, you can't catch errors, and you can't defend the quantities to a GC. A spreadsheet of numbers with no source link is worse than useless.
Are Any AI Takeoff Tools Free to Try?
Steelflo offers a free trial with one AI takeoff — no credit card, no demo call. Beam AI requires a demo. SketchDeck.ai and Ferra availability varies. Always test on your own drawings before committing to any tool.