Twenty years of running projects has taught me that the most expensive document on a job is rarely the contract. It is the bid tab nobody leveled properly. Three electrical subcontractors quote your custom home. One says $68,000, one says $74,000, one says $61,000. The $61,000 bid wins, obviously, because the owner sees a number and the GC sees margin. Six weeks into rough-in, the electrician mentions that lighting controls were not in his number. Neither was temporary power. The change orders arrive at $9,400, then $6,100, and suddenly the cheap bid is the most expensive one on the tab. I have watched this movie enough times to know the ending by heart.
Bid leveling is supposed to prevent exactly this. You take every sub quote, normalize them into a single scope sheet, line items as rows and bidders as columns, and you find the gaps before anyone signs anything. In practice, leveling is the most grueling step in preconstruction. An estimator spends days transposing numbers out of PDF quotes into a spreadsheet the size of a bedsheet, squinting at exclusions written in eight-point font, trying to determine whether Subcontractor A's "electrical package" contains the same universe of work as Subcontractor B's. It is meticulous, exhausting work, and it is where the mistakes that cost real money are born.
Where the money actually leaks
Change orders are not bad luck. They are what happens when the planning phase does not do its job, and a meaningful share of them are scope gaps that careful leveling would have caught. Dodge Data & Analytics puts the average change-order load at 10 percent of contract value, with troubled projects hitting 25 percent. PlanRadar's 2025 synthesis of rework research lands in the same neighborhood: 5 to 10 percent of project value lost to rework on typical jobs, up to 20 percent on poorly controlled ones, enough to erase a contractor's entire margin.
Run the arithmetic on an $800,000 residential GC contract. Ten percent in change orders is $80,000. If even a quarter of that traces back to scope gaps missed during bid leveling, and I am being conservative because nobody has cleanly measured this split, that is $20,000 per project walking out the door for want of a careful spreadsheet. On a builder doing six homes a year, that is a full-time estimator's salary, lost to exclusions nobody read.
What the new tools do
The AI wave has finally reached the estimator's desk. STACK Construction Technologies launched STACK IQ on September 1, a conversational AI layer for estimating and preconstruction that audits an existing estimate for missing items before a bid goes out, builds takeoff libraries from spreadsheets in seconds, and reformats proposals without the estimator learning a new workflow. Togal.AI auto-detects and measures plan areas for rapid estimates. ConWize and its competitors have been grinding at the bid management side for years.
These tools parse unstructured PDF quotes, line items arriving in five different formats from five different subs, and normalize them into a single comparison. Your estimator stops being a data-entry clerk and goes back to being a judge of scope. On paper, the days of transposing numbers by hand are ending.
A quiet process note that costs nothing
Before anyone buys software, there is a discipline that costs zero dollars and catches half the gaps: require every sub to bid against your scope sheet, not their own proposal format. Send the line items. Make them fill in your columns. Subs who refuse, or who return a lump sum with a page of exclusions, have told you something valuable. The best estimators I have worked with kept a wall of shame, a binder of bid tabs where the low bidder's exclusions were highlighted in yellow, and they consulted it the way a pilot consults a checklist. AI can now do the highlighting. It cannot do the remembering. That part is still yours.
The part nobody wants to hear
Now the honest counterargument, because a tool review that skips it is a brochure. Bid leveling software does not fix the incentive problem underneath the spreadsheet. Owners still award to the low bidder, and the low bidder is frequently low precisely because something was left out. An AI that perfectly flags every scope gap does not stop a GC under margin pressure from taking the cheap bid anyway and planning to "manage it in the field," which is estimator slang for hoping the problem belongs to someone else by the time it surfaces. I have seen beautiful leveled bid tabs get overridden by a principal who liked the low number. The spreadsheet was right. The decision was wrong. No software update fixes that.
There is a second-order effect worth naming. Perfect leveling transparency teaches subs exactly which line items get scrutinized, and the clever ones adapt by moving exclusions into allowances, alternates, and qualifications phrased just vaguely enough to survive a side-by-side. That arms race does not end; it moves. And for the small residential GC running three projects a year, the subscription math is genuinely questionable. These tools are priced and trained for bid volume, and a builder who levels twelve trade packages a year may be better served by the scope-sheet discipline and a careful afternoon than by another monthly invoice.
What to do Monday morning
If you run a GC or build on contract: before awarding any trade package over roughly $50,000, produce a leveled bid tab with line items as rows, bidders as columns, and every exclusion flagged in plain language. AI tools cut the transposition labor to a fraction of what it was. The remaining work is judgment, and it was always the valuable part.
If you are a homeowner signing a fixed-price contract: ask your builder whether the sub bids were leveled and whether you can see the exclusions list. A bid $30,000 under the next competitor, carrying six exclusions, is not cheaper. It is a $30,000 change order waiting for your signature.
If you are a subcontractor: bid against the GC's scope sheet in the GC's format. Estimators remember who made their Tuesday easy and who made it miserable, and the easy ones get invited back. In a market where AI is doing the comparing, the most quotable sub is the one whose numbers drop cleanly into the columns.
Limitations of this analysis
STACK IQ launched four days before this article was written. No independent verification of its estimate-auditing accuracy exists, and the customer quotes in its launch materials are vendor-supplied. The "one quarter of change orders from scope gaps" figure is my assumption for the worked example, not a measured statistic; no published study cleanly separates scope-gap change orders from owner-directed changes, unforeseen conditions, or design errors. Pricing for AI estimating and leveling tools is mostly quote-based and varies widely, so treat any per-seat figure in trade press as directional. Field data on adoption among small residential GCs does not exist. Finally, the $20,000-per-project illustration assumes the Dodge 10 percent average applies to residential GC work specifically, which the underlying study does not break out; your trade mix and contract type will move the number.