Every year, DBIA draws the people who run pre-construction for the most complex design-build projects in North America. In 2026, the conversation has shifted. AI isn't a side topic anymore. DBIA 2026 in Dallas features the first dedicated AI in pre-construction track in the conference's history.
That's worth paying attention to — not because AI is new, but because the industry is finally asking the right questions about it: How accurate is it on real specs? What does it actually replace? And where does it still fall short?
This guide breaks down what GC pre-construction teams need to get out of DBIA 2026's AI sessions — and what to bring back to the office that actually moves the needle.
Design-build projects are pre-construction-intensive by nature. Scope is negotiated, not just read off a drawing set. The GC is often in the room when scope is still being defined. That makes scope discipline — and the tools that support it — a competitive differentiator.
The industry data backs up the urgency. According to the Arcadis 2025 Global Construction Disputes Report, the average U.S. construction dispute was worth $60.1 million in 2024. Errors and omissions in contract documents have been the number one dispute cause for six of the last nine years.
At the same time, FMI's Construction Disconnected report put the annual U.S. rework cost at $31 billion. Twenty-six percent of that rework comes from communication breakdowns. Twenty-two percent from bad project data.
Pre-construction teams don't cause all of that. But they're the last checkpoint before a scope gap becomes a field problem. That's why DBIA's decision to dedicate a full conference track to AI in pre-construction is significant. The industry is starting to measure what AI can actually do — not just demo.
DBIA 2026's AI track spans three primary topic areas. Each one maps directly to a workflow problem GC pre-construction teams deal with on every pursuit.
This is the highest-leverage conversation happening at the conference. Scope gaps — items missing from subcontract scopes, conflicts between drawings and specs, trades that assume the other party owns a line item — don't show up as surprises. They're produced by specific habits during bid and buyout.
A Pre-Construction Lead at a top-ENR Canadian GC put it plainly: "If you miss anything, they'll bill it." That's not new. But the scale of what gets missed is still underappreciated.
Consider a few real examples from GC interviews conducted for the Scope Gap Playbook:
These aren't rounding errors. They're the result of bid-stage habits that AI is now being applied to fix — and DBIA's 2026 sessions are walking through both the habits and the tools.
If you want the full framework, the Scope Gap Playbook covers the eight habits that separate firms with clean scopes from firms that bleed margin in the field.
This is the session most chief estimators should be in. The room is full of vendors claiming AI accuracy. Very few of them publish benchmarks on real construction documents.
The distinction matters. A large language model trained on general text will hallucinate on a 2,000-page project manual. It doesn't know that "gypcrete" and "lightweight concrete topping" might refer to the same scope item with different cost implications. It can't reconcile a drawing note with a spec section that contradicts it.
Purpose-built construction AI is different because it's trained and validated on actual project documents — drawings, specs, addenda, RFIs, and contracts together. Provision's Risk Review, for example, achieves 99.5% accuracy on pre-built risk checklists and 97%+ on custom ones. That's not a marketing number — it's measured against real project documents, not test sets built to make the tool look good.
Provision has processed over 66,000 documents and found more than 1,000,000 risks across $100 billion in project value. That's the training surface that makes accuracy numbers meaningful.
At DBIA 2026, expect the accuracy benchmarking sessions to draw a hard line between tools that surface answers and tools that surface correct answers with citations. That line matters when you're deciding whether to trust AI output on a $50M bid.
The third major topic is workflow integration — specifically, how AI fits into existing pre-construction processes without breaking them.
The honest answer is that the biggest time savings are in document search and first-pass review. A project set for a mid-size commercial job can run 1,500 to 2,000 pages of specs alone. An estimator doing a trade-specific read-through to check for scope inclusions, alternates, and owner-furnished items can spend 30 to 40 hours per bid on that work alone.
Provision's Chat Agent answers questions directly from the project set — drawings, specs, contracts, addenda — with cited answers in under 20 seconds. That's not a chatbot experience. It's a tool that knows where the answer is in the documents and shows you exactly where it found it.
DBIA's sessions are likely to surface the same finding GC teams are reporting in the field: AI doesn't replace the estimator's judgment. It eliminates the part of the job that shouldn't require judgment — digging through documents to find information that's already there.
Design-build creates scope risk that hard-bid projects don't have. When the GC is involved early, scope evolves. What was a conceptual allowance in month two becomes a fixed-price subcontract in month six. The gap between those two documents is where margin disappears.
A Director of Pre-Construction at a mid-market Southeast GC described the disconnect this way: "Pre-con is working in the scope sheet world and project management is working in the scopes of work." That handoff — from pursuit language to subcontract language — is where scope gaps are created at scale.
Design-build amplifies this because the design is still moving when early trade packages go out. Drawings and specs aren't final. That means the scope package can't just reference documents that don't exist yet. It has to define scope in language precise enough to survive document evolution.
A Pre-Construction Lead at a top-ENR Canadian GC described the standard they hold themselves to as the "peanut-butter test": "It's descriptive — bread, put it on a plate, use the open jar… You have to get to that level of detail or else they'll just be like, 'you didn't tell us that.'"
That level of specificity takes time. AI tools that help generate and validate scope language against evolving documents are directly applicable to design-build workflows. That's a key reason the DBIA AI track matters for GC teams specifically.
The exhibit hall will have AI vendors. Some are purpose-built for construction. Many are not. Here's a simple framework for evaluating what you see.
Generic AI — including ChatGPT and general-purpose copilots — fails on most of these criteria. They're not built for construction document ingestion. They don't produce structured scope outputs. And they don't have the construction-specific training needed to catch a trade gap in a division 09 spec.
Purpose-built tools like Provision's Scope Agent are built specifically for GC pre-construction workflows. Scope Agent generates complete scope-of-work packages from full project document sets in under 60 minutes — replacing 30 to 40 hours of manual work per bid. That's the kind of output estimators can actually use on bid day.
Whether you're in a panel, a product demo, or a hallway conversation, these three questions will cut through the noise faster than anything else.
Every AI tool will find risks. The question is whether it finds real risks. A tool that flags 200 items per spec but half are irrelevant trains estimators to ignore the output. Accuracy isn't just about catching things — it's about not crying wolf.
Ask the vendor to run a demo on a project set that resembles your work. ICI commercial, healthcare, institutional, industrial — the vocabulary and scope structure are different. A tool that works on a Class A office spec may not perform the same way on a complex MEP-heavy hospital package.
Summaries aren't useful on bid day. Ask to see the actual deliverable — the scope package, the risk checklist, the flagged document sections. If the output requires significant reformatting before it's usable, that's not a time-saving tool. That's a first draft generator.
DBIA 2026 will generate a lot of conversation about AI. The firms that get value from it won't be the ones most excited about the demos. They'll be the ones who leave with a clear picture of where AI fits in their current workflow — and where it doesn't.
For GC pre-construction teams, the highest-value applications right now are:
These aren't speculative use cases. GC teams are using Provision for general contractors on live pursuits right now. The EllisDon case study documents $1.8M saved on a single project. The NAC case study and Cleveland Construction case study show the same pattern: faster pre-construction, fewer scope gaps, better margin control.
Provision has answered over 50,000 queries from real construction documents, with 95% verified accuracy. That's the proof point to ask AI vendors for at DBIA 2026 — not a demo on a clean, curated dataset, but evidence of performance on real project documents at real scale.
If you want a structured framework before the conference, the Scope Gap Playbook is built on 200+ interviews with GC estimating and pre-construction teams. It gives you the vocabulary to evaluate what you're hearing in Dallas — and the benchmarks to know whether a vendor's claims hold up.
DBIA 2026 is the Design-Build Institute of America's annual conference, held in Dallas. It draws owners, architects, engineers, and general contractors working on design-build projects across North America. GC pre-construction and estimating leaders attend to track industry standards, procurement trends, and emerging tools.
It's the first year DBIA has dedicated a full conference track to AI in pre-construction. The move reflects how quickly AI tools are entering GC workflows — particularly for document review, scope package generation, and risk identification. The industry is ready to move past AI hype and into practical evaluation.
Purpose-built tools designed for construction document ingestion are the most relevant. Look for tools that read full project sets — drawings, specs, contracts, and addenda together — and produce structured, bid-ready outputs. Generic AI tools like ChatGPT are not designed for this workflow and lack the construction context needed for accurate scope and risk work.
AI helps by reading the full project document set and flagging missing trade inclusions, conflicts between drawings and specs, and items that fall into "readily inferable" gray zones. On design-build projects, where scope evolves through design development, AI tools can validate scope packages against the current document set — catching gaps before they reach subcontract buyout.
Ask for accuracy rates on pre-built and custom risk checklists, tested against real project documents — not curated demo sets. Ask for the false-positive rate on risk flags. And ask what volume of real projects the tool has processed. Provision, for example, has reviewed $100 billion in project value across 66,000+ documents, with 99.5% accuracy on pre-built checklists.
In design-build, the GC is often involved before drawings are complete. Scope evolves through design development, which means early subcontract packages may be written against incomplete documents. That creates additional exposure to scope gaps — especially when scope language doesn't evolve with the drawings. Precise, document-referenced scope writing is more important in design-build than in any other delivery model.
The Scope Gap Playbook is a free resource built on 200+ interviews with GC estimating and pre-construction teams. It covers the eight habits that prevent scope gaps, the anti-patterns that create them, and trade-specific examples from real projects. It's a practical read before the conference — or a useful framework for evaluating what you hear in Dallas.
Provision reviews full project sets and delivers bid-ready scope packages in under 60 minutes.
Book a demoMore Articles