1 · The shift
By the end of 2026, you’ll have your first AI estimator on staff.
Not a copilot. Not a plug-in. An always-on teammate that runs the routine preconstruction loop while your humans focus on the bids that matter.
AI Estimator
Bid due tomorrow? Get your takeoff tonight.
Drop your plan set into Slack. Get a quantified takeoff back in 30 minutes — not 30 days. Onboarding is one message, not a procurement cycle.
Async path: drop a plan set in the email. Gandalf reads it and sends the takeoff back within 30 minutes. No call required.
Real takeoff, real plan set, real time — electrical fixtures counted live.
What an AI estimator actually does
One workflow, four steps. Drop a plan set into a Slack channel and a real bid package comes back — not a chatbot answer, not a half-finished draft.
- 1
Slack message
An estimator drops a plan set into a project Slack channel and asks for what they need. No new tool, no new login.
- 2
Reads drawings & specs
The AI estimator reads every sheet, every spec section, every schedule. It treats drawings and specs as one document, the way an estimator does.
- 3
Flags spec ↔ drawing mismatches
Where the drawings say one fixture and the specs say another, the AI estimator surfaces the conflict before it becomes a change order.
- 4
Takeoff + bid summary
Quantified takeoffs across trades and a bid-ready summary — delivered back to the same Slack channel, ready to send out for pricing.
Why an AI estimator, why now
Four reasons preconstruction teams are putting AI estimators on staff this year.
2 · The proof
See it in 60 seconds, not 60 minutes.
A real plan set, a real Slack message, a real takeoff. No staged screens. No edited cuts. Watch the hero above.
3 · The receipts
Task-level accuracy, published.
Electrical fixtures and conduit ≥ 0.90, architectural ≥ 0.85, and mechanical ≥ 0.74 are published floors. Task-level results include electrical device F1 0.64, straight/rigid mechanical ductwork F1 0.85, flex duct F1 0.72, and HVAC fitting F1 0.84. Ask any vendor for theirs. Most won’t answer.
4 · The model
One model beats thirty-seven.
The Construction Vision Model is one foundation model trained across every trade, with specialized heads on top — not a zoo of narrow models that hit a ceiling.
Task-level F1 receipts
F1 score is the standard accuracy metric for object detection. We publish task-level results and separately stated floors, so buyers can compare the task that matters to their own plan set.
| Task | Published result | What it covers |
|---|---|---|
| Structural steel beam detection | F1 0.93–0.95 | End-to-end beam line detection, labels, centerlines, and braces. |
| Architectural door detection | F1 0.88 | Door symbol detection across architectural floor plans. |
| Mechanical straight ductwork detection | F1 0.85 | Straight duct segments with width measurements; flex duct detection is F1 0.72. |
| Mechanical HVAC fitting detection | F1 0.84 | Elbows, dampers, GRDs, and flex ducts in the combined pipeline. |
| Electrical device detection | F1 0.64 | The known weak area, reported separately from fixtures and conduit. |
Separately, the published floors for work we have not seen before are electrical ≥ 0.90 (fixtures and conduit), architectural ≥ 0.85, and mechanical ≥ 0.74. Floors are not replacements for task-specific F1, and electrical device detection is F1 0.64. If a vendor won’t publish comparable evidence, that’s the answer.
Put Gandalf on a real bid
Try it on your own plans. Then put it to work.
Book a 20-minute call to get your workspace provisioned and your first takeoff back in the same call — or drop a plan set into plans@getboon.ai and Gandalf returns a quantified takeoff in 30 minutes.
Advancing Preconstruction 2026
From Advancing Preconstruction 2026
What we said on stage, the demo we ran in the booth, and the library.
Five-minute best-of from the keynote — the construction vision model thesis, the F1 receipts, and the offer.
Want the full 30-minute keynote, the 60-second shorts, or the BBB Ep 77 cuts? They’re on our YouTube channel.
Drop a plan set. Get the takeoff tonight.
Book a 20-minute call — we’ll provision your workspace and run your first takeoff on the call. Or email a plan set to plans@getboon.ai and Gandalf returns the takeoff in 30 minutes.