An estimator at a GC ran one of our early conflict reviews on a single project and got 738 flagged items. His reaction was the honest one: discarding the irrelevant ones is easy, but there were a lot of them.
That's the real problem with document AI in preconstruction, and almost nobody writes about it. Recall's the easy half. A tool that reads every drawing against every spec section will find contradictions, because real project sets are full of them. Most of them don't cost you a thing.
So the skill isn't finding conflicts. It's sorting them, and here's the method we've watched work.
A flagged conflict has a review cost whether or not it matters. Somebody opens it, reads both sources, decides, and moves on. Call it 90 seconds each when the citation is right there in front of you.
738 items at 90 seconds is about 18 hours. That's most of a bid week gone to triage, and you haven't priced a thing yet.
Cut that list to the 60 items that carry money and the same review takes 90 minutes. The findings didn't change. The order did, that's all.
Ranking by severity sounds right and doesn't work, because severity's a property of the drawing and cost is a property of your contract. A dimension mismatch on a millwork detail is a serious documentation error and costs you nothing if millwork is a lump-sum package with a tight scope letter.
Sort by who ends up paying instead. Four buckets, in this order.
The item appears in the documents and lands in no trade package. These are scope gaps, and they're first because they have no ceiling. There's no wrong number to argue about, because there's no number.
The Scope Gap Playbook interviews are full of these. A $200K wood-flooring gap on a luxury condo. Recurring disputes over generator field conditioning that one operator put in the millions, project after project, because nobody named who was responsible for it.
Interface conflicts. The trench, the embeds, the blocking, the flashing at a storefront-to-masonry junction. You'll pay for these twice, or you'll fight about them once.
Sort these by the dollar value of the two packages that touch the interface, not by how badly the drawings disagree.
Drawings say one thing, specs say another, and a sub has already bid one of them. Real exposure, but bounded: the gap between two known positions. Resolve by RFI before bid close and you've spent a phone call.
Stale references, superseded detail callouts, an addendum that renumbered something. Worth fixing so the record's clean. Worth zero on bid day.
Most of a 738-item list is bucket four. That isn't a failure of the tool. It's what a complete read of a real document set looks like.
Two passes, and the first one is fast.
Pass one, by trade. Group every flag by the package it touches and go trade by trade with the person who owns that buyout. They'll clear bucket four at a glance, because they know which callouts don't matter on their scope. Do not do this alone at 11pm before a bid.
Pass two, by dollars. On what survives, put a number next to each one. Not a precise number, just an order of magnitude. Anything above your change-order threshold goes to an RFI. Anything below goes in a clarifications list and into the scope letter.
What you want out of this isn't a clean conflict log. It's a shorter list of RFIs and a longer list of clarifications, which is the same shape as good scope practice generally.
Being straight about it: prioritization's manual today. Risk Review gives you severity levels and cites every finding to the exact clause, section or page, which is what makes a 90-second decision possible instead of a re-read. It runs at 99% accuracy on pre-built checklists and cuts review time by about 80%.
What it won't do is rank by your cost impact, because your cost impact depends on how you package trades and what your scope letters say. We can hand you every finding with its source attached. Deciding which ones carry money is still the estimator's call, and honestly that's the part you'd want to own anyway.
The teams that get value from conflict detection are the ones that built a triage habit before they turned it on. The teams that bounce off it are the ones who opened a 700-item list with no method and concluded the tool was noisy.
That last one is worth more than any demo. Bring a finished project and we'll run it with you.
Because real project sets contain many genuine contradictions, most of which carry no cost. High recall means you see all of them, including the ones your team would have ignored.
Severity is a property of the documents. Cost is a property of your contract and packaging. Rank by who ends up paying.
Budget about 90 seconds per flagged item when each one is cited to its source. Sorting by trade first removes most of the list before anyone reads closely.
No. It surfaces findings with severity levels and a citation to the exact clause, section or page. Cost ranking depends on your trade packaging and scope letters, so that judgment stays with your estimator.
Roll them into a clarifications list that goes into the scope letter, rather than an RFI. They still protect you, without spending owner or design-team time.
Risk Review flags conflicts with the exact clause, section or page attached.
See Risk ReviewMore Articles
