What 1,494 Change Orders Reveal About Construction's Real Problems

Change orders have a bad reputation, and most of what gets written about eliminating them is aspirational. A recent data analysis by Alim Uderbekov, CEO and co-founder of SURF(AI)CE, took a different approach: he looked at 1,494 change orders across 25 retail and quick-service restaurant brands to see what actually drives them. The findings line up with what we see on our own projects, and they're worth sharing with clients who want fewer surprises during construction.

The headline number: change orders added about 4% to contract value across the dataset, or roughly 7% if you include one large outlier. That's consistent with the best public benchmark available, the AIA/Catina study's 3-5% range across more than 18,000 U.S. projects, not the inflated 8-14% figure that gets repeated without a real source behind it.

Break the numbers down and a clearer picture emerges. A small number of large change orders account for most of the dollar impact, while the bulk of the paperwork comes from smaller ones. Projects in existing buildings ran about 2.5 times more change order cost than ground-up builds, which tracks with something every GC already knows: new construction drawings describe what you're going to build, while drawings for an existing building describe what someone hoped was behind the wall.

By cause, weather and ground conditions are rare but expensive when they hit. A handful of items keep reappearing project after project: ceiling tile, drywall-to-deck, slab leveling, roof patches, and electrical service, alongside local code requirements a jurisdiction adds late and owner-furnished equipment that shows up incomplete.

The most useful classification isn't the reason category, it's whether the change order could have been known before the job was priced. Two out of three change orders in the dataset came from information that already existed somewhere before the bid: in a site walk, a local code, a utility's standard practice, an owner's equipment spec, or the last project built for that same client. And 86% of those "knowable" change orders were still approved and paid for, at markup, on top of the schedule impact.

That points to what the analysis calls a memory problem more than a construction problem. Individual change orders get processed and filed away, but the underlying lesson, the code requirement, the equipment quirk, the site condition, rarely makes it into the next bid. A $3,000 surprise on one store is a rounding error. The same $3,000 surprise repeated across 50 stores of the same building vintage is $150,000, and it was avoidable after the first one.

The piece also makes a point worth sitting with as more of the industry starts using AI for estimating: it only helps if a GC brings the right context, and it can make things worse if it doesn't. A model with no memory of past change orders, local requirements, or a client's equipment history will still produce a clean, confident number, wrong in exactly the same places a person without that history would miss. The value is in the discipline behind the tool, not the tool itself.

It's a good reminder for anyone managing a multi-site program: the projects with the fewest change orders usually aren't the ones with the least risk. They're the ones where somebody remembered what happened last time.

Source: This post summarizes original research and analysis by Alim Uderbekov, CEO and co-founder of SURF(AI)CE, published on his Substack. Read the full piece, "Demystifying Change Orders," here: https://surfaice.substack.com/p/demystifying-change-orders