Computer Vision Is the Andon Cord Construction Has Always Needed

LCI Conference 2025  ·  Reid Senescu, Doxel  &  Mike Miller, DPR Construction

‍

The Problem With Construction Schedule Data

Every superintendent knows the feeling: a trade partner reports they're on track, the schedule says green, and then, one week before a milestone, the reality hits. The work isn't there. The cascade starts. The conversations get harder.

The problem isn't that anyone is lying. The problem is that construction progress has always been measured the same way: someone walks around, someone asks, someone estimates. By the time a deviation surfaces through normal reporting channels, weeks have passed, and the cost of recovering has multiplied.

Data-driven construction executives are disciplined in leveraging technology and processes to capture crucial information about their projects. At LCI Conference 2025, DPR Project Executive Mike Miller described how automated progress tracking changed that dynamic on his hyperscale data center project, starting with a piping trade partner whose slip he could see forming in real time.

‍

The Piping Story: When the Data Sees What Manual Processes Miss

Starting in December, Doxel's computer vision system began showing a divergence between the piping trade's actual installed quantities and their planned schedule. Not a dramatic gap at first. Just a signal.

Under traditional reporting, that gap might not have surfaced until a formal schedule update meeting or until it cascaded into delayed downstream trades. With automated tracking, Mike could see it forming week by week, grounded in automated progress tracking rather than self-reporting.

"As a leader, you only have so much time to focus on certain things. I can't focus on everything all the time. I can see who's on plan and I can focus where we have the risk."— Mike Miller, Project Executive, DPR Construction

What followed was a textbook plan-do-check-act (PDCA) loop. The trade was brought to the table. The data was shared. Initially, there was resistance — new tools, new data, trust takes time. But by year-end, the team accepted the picture and decided to add resources. Check the result. Adjust again if needed.

Mike called it "bread and butter." The system is working exactly as it should.

‍

Why Early Detection Only Works if You Act on the Signal

KEY INSIGHT  Early detection gives leadership the visibility to have the right conversation at the right time.

The lean parallel is direct. In manufacturing, the Andon cord's value isn't the cord itself; it's the organizational culture that responds to it. Doxel's computer vision functions as a continuous Andon cord across the entire job site, flagging deviations the moment they're detectable, not weeks later when they've compounded.

Reid Senescu, Head of Product at Doxel, described the underlying principle: construction sites have historically lacked the sensor layer that lean manufacturing has always relied on. No factory would operate without real-time feedback on production output. Yet construction, which is far more complex and expensive, has run primarily on estimates and walking the site.

How Doxel's Tracking Actually Works

The system takes three inputs: 

• 360° video captured weekly using an off-the-shelf camera
• the project's BIM model
• and the P6 CPM schedule 

Computer vision compares the scan against the BIM to determine what has been physically installed at the component level across all trades and at every stage of construction. A large language model ties the schedule to the BIM, producing a real-time 4D view of the actual project state versus the planned state.

‍

The result: trade and project-level progress data grounded in physical observation rather than self-reporting, updated at least weekly, with no additional engineering work required for onboarding.

Doxel has captured over 3 billion square feet of construction. That dataset powers the "rules of credit" — weighted effort models that let the system aggregate component-level data up to an accurate project percent complete.

‍

What This Means for Owners and GCs

For owners, the implication of this story is straightforward: the capability exists to monitor capital investments in construction with the same rigor applied to financial data, inventory, or operational metrics. Requiring automated progress tracking in the RFP: specifying that it must cover all visible trades and at least 80 stages of construction is the most direct way to ensure this capability is on every project.

For GCs like DPR, the story is about competitive differentiation. The teams that build trust in objective data and act on it early catch problems before they cascade. They spend less on rework and recover faster.

The Andon cord is finally available for job sites. The question is how quickly the team can respond when it sounds.