Arbitration promised efficiency. Construction data stole it. GenAI can recover it.

I once worked a $25 million construction dispute that cost about $15 million to litigate. 1.2 million documents. A partner, a paralegal, and I all relocated out of state for a month to arbitrate, billing 12 hours a day. A panel of three arbitrators, one of whom charged $7,500 per day. Extensive post-arbitration briefing that dragged on for years.

The parties had chosen arbitration, of course. The pitch is well-worn: arbitrate, and you get a faster, more-proportionate path to resolution. You escape litigation’s procedural sprawl, its jurisdictional gamesmanship, its discovery free-for-all, plus a decision-maker who understands a schedule delay and can read an MEP drawing.

What the pitch doesn’t account for is the data. Oh, the data.

Construction Dispute Data Is a Double-Whammy Problem: It’s Volume and Diversity

Here’s what differentiates construction disputes: the documents aren’t documents. Not in the sense that a contract, a memo, or an email is a document.

A construction project generates RFIs—requests for information—that chain across months and reference submittals that reference drawings that reference specifications that reference a change order that was never formally executed but everyone acted on anyway and that three people texted about. It generates daily reports, hand-marked field drawings, schedule files in Primavera or Aconex, BIM models, cost-loaded schedules, critical path planning, subcontractor correspondence scattered across Procore, email, and text. It generates years of data because projects run long and disputes run longer—and the documents accumulate cross-referentially, across platforms and timelines that only make sense if you understand the project.

1.2 million documents isn’t unusual. That volume alone is manageable.

The double whammy: those 1.2 million documents speak fifteen different languages, and standard review workflows were built for one or two. A commercial litigation review attorney can parse a long email chain faster than you can say “critical path.” But they probably can’t parse the relationship between an RFI response, a marked-up drawing, and a change order directive—and understand, intuitively, that the absence of a formal change order despite an undisputed direction to proceed is itself a fact of consequence. The document type is alien. The cross-referential logic is alien.

Previous Tech Was Good, but Not Right-Fitted

Technology assisted review—TAR, predictive coding, call it what you like—cut volume with reasonable accuracy. That ain’t nothing.

But here’s what TAR couldn’t do: understand that an RFI chain, a change order directive, a set of hand-marked drawings, and a project schedule update are in conversation with each other. It treated your construction documents like emails, processed them accordingly, and offloaded a smaller pile that was still, fundamentally, opaque. That limitation has long defined construction eDiscovery workflows.

And it’s ironic: construction companies build projects on AI-native platforms. Procore. Primavera. BIM environments that make structural relationships explicit and searchable. Then a dispute arises, and we strip all that intelligence out and flatten the data into a review queue. We take a sophisticated, interconnected data architecture and review it like it’s a box of photocopies someone scrounged up from a storage unit.

Once upon a time, technological limitations forced us to smoosh our square-peg data into round-hole eDiscovery tools. ’Twas the best we could do. But then GenAI arrived. So why are we still self-sabotaging?

A fable comes to mind: once there was a flea. Fleas are prodigious jumpers and can launch themselves many times their own height. Put one in a jar with a lid, though, and it learns. It jumps, hits the lid, adjusts. After enough collisions, it caps its own jump at lid-height. The interesting part: remove the lid, and it still won’t jump higher. The ceiling is gone. The flea doesn’t know that. It internalized the constraint so thoroughly that the constraint no longer needs to exist.

The legal industry is the flea. The lid was technological limitation, and the habits calcified. The workflows ossified. We’re still jumping lid-height out of sheer institutional memory and billing accordingly. But GenAI blew the lid off: we can jump higher now.

Earlier Tech Sorted; GenAI Interrogates

First: construction-aware categorization before review begins

GenAI doesn’t just cull for relevance. It categorizes documents at a granularity that previous tools couldn’t touch—by trade, by building system, by document type, by project phase. Before a single human reviewer breaks ground, you have a bird’s-eye view of the data landscape—what you have, where it came from, and where the dispute concentrates.

I once litigated a matter that hinged on metal shavings and damage to a waterproofing membrane. (This was in the days of yore, before GenAI.) I combed hundreds of photos hunting metal shavings; I bored through spec sheets; I sifted thousands of emails for the needle—all at northwards of $500 an hour. Now, I litigate it differently in my head. Armed with GenAI, I’d run the dataset through a platform like eDiscovery AI, drill down on the 600 or so that dealt with metal shavings, and understand the timeline—all within a day or two.

GenAI tells your outside counsel where to concentrate their efforts. That gives you control over the data they’ll review, and that gives you control over the budget. Instead of a budget built on assumptions and professional optimism, you get one built on actual data—document counts, document types, review scope. Now, you get the picture before you start the puzzle.

Second: natural language interrogation of the full document set

Ask a retrieval-augmented generation (“RAG”) system: “Is there any evidence the concrete subcontractor used an unapproved grade?” Answering that the old way meant weeks of targeted review across fragmented document sets—daily reports, test logs, submittal records, field correspondence—with no guarantee you’d surface the right documents even then. Now it’s a query; an answer within minutes.

RAG tools, however, don’t just find documents with the words “concrete” and “unapproved.” They find documents where the relationship between a submittal approval status and a daily report note implies the answer—across document types, platforms, and the project timeline.

I’ve seen this surface evidence that linear review would have found late, or not at all. One matter: a contractor believed their engineer had stolen project secrets. A natural language query returned a document containing the sentence “the key is to have the plans in the bag.” That sentence wasn’t on anyone’s custodian list. No keyword search would have flagged it. RAG found it because it understood the query and the parlance.

That’s what wins cases: not a smaller review pile, but the right document, at the right moment, before the other side finds it first.

Third: the timeline your project never kept

Delay claims are the engine of construction litigation: who was responsible for which delay, when, in what sequence, with what notice, affecting which critical-path activities? Traditionally, that meant a forensic-scheduling expert, weeks of document review, and a hefty bill.

Enter GenAI. Feed it the project schedule files, the RFI logs, the daily reports, the change order directives—and ask it to surface the sequence of events on a particular work scope. It won’t replace your expert. But it will gift them a map instead of a blank page, find the document they needed but didn’t know existed, and excavate the gap in notice that rewrites the damages calculation—in days, not weeks.

Two Questions to Ask Before the Next Matter Begins

The goal isn’t to quiz your outside counsel on AI literacy. It’s to signal that you know enough to hold them accountable.

• How are you using GenAI to excise irrelevant documents before review begins? Not during. Before. If the answer is “we’ll see what the data looks like when we collect it,” that’s a concern.
• How are you using GenAI to prepare for depositions? Deposition prep is a document problem. Preparing without GenAI is archaeology with a toothbrush—possible, technically, but you’re paying for the methodology. GenAI interrogation compresses that preparation timeline.

These aren't hypothetical questions. They're the difference between a matter that runs on assumption and one that runs on intelligence.

GenAI is the first tool capable of making arbitration's original promises real: speed, proportionality, and a decision-maker who understands your data. The smarter path parties always believed arbitration could offer is now actually available. The tools finally caught up to the complexity. GenAI construction arbitration is no longer theoretical—it’s operational.

If you are heading into a construction dispute, or want to pressure-test whether your outside counsel is putting these tools to work, contact our team or explore how TransPerfect Legal approaches GenAI for eDiscovery. To learn more about the depth of experience behind this work, visit our Construction Disputes Practice Group.