An independent magazine about AI at work in construction. Every story starts on a real job site, with the people who used the technology and the results they will answer for.
Rogers-O'Brien Reports 350,000 Questions Answered in Compass's First Year | ConstructionMagazine.ai
Rogers-O'Brien Reports 350,000 Questions Answered in Compass's First Year
The Texas contractor's in-house AI assistant now reaches 705 people a month and reads Procore, scheduling, estimating and Microsoft 365, by company figures published Sept. 1–2, 2026. The counts are the company's own and unaudited.
Rogers-O'Brien Construction marked the first anniversary of Compass, the AI assistant it built for its own employees, with a company post on Sept. 1, 2026 that puts a number on the tool's use: more than 350,000 questions answered since launch. The Texas general contractor says 705 people used Compass in a typical month and that nearly a third of the company opens it on a given day. Those figures come from the company's own logs and are unaudited. ConstructionMagazine.ai has not reviewed the underlying data.
Compass draws on several of the company's systems at once. According to the post, it connects to Procore, the firm's scheduling and estimating software, an internal data warehouse, Microsoft 365 and Playbook. It runs under the company's enterprise security controls and respects the access permissions employees already hold, so a project engineer asking about a budget line should see only what that engineer could already open in the source system. Users come from field operations, quality, preconstruction, accounting, marketing and virtual design and construction.
The company amplified the post on X the following day, Sept. 2, and the anniversary framing is itself notable. Most contractor AI announcements describe a launch. This one describes a year of use, with counts, which is rarer and more useful to readers trying to judge whether such tools stick after the initial novelty wears off.
AI-assisted illustration.
Three workflows the company timed
The post leads with time. A weekly schedule narrative that a project team used to write across a Friday afternoon now takes about 15 minutes, the company says. Comparing a project budget against its estimate used to be dominated by the work of assembling files; the company says the time is now spent mostly on the comparison itself. On a healthcare project, a team reconciled budget and estimate in minutes rather than the afternoon it had taken before. None of these are controlled measurements. They are before-and-after estimates from the teams involved, reported by the company in its own post.
Attention is one of the most valuable things our builders bring to a jobsite. … Compass helps protect it by making information easier to access and guiding focus toward what matters most. That allows us to plan better, support healthier teams and strengthen critical thinking, not replace it.
Wynne's framing sets the standard the company wants to be judged against. The claim is that Compass gives attention back to builders, and the evidence offered for it is time saved on retrieving and assembling information.
In the field, a feasibility study becomes a schedule
Field Director Bob Britton describes using Compass to move from a feasibility study to a planning-level critical path method schedule. In that exercise, the company says, the analysis indicated that reusing an existing building could save five to six months compared with building new. Britton's account is the most ambitious use in the post, because it treats Compass as a contributor to schedule logic and constraint analysis rather than as a search tool over dates someone else already entered.
I've used Compass as a planning partner to develop schedules based on scope, field execution, regulatory requirements and contract risk, not just to review existing dates.
The post does not say whether the five-to-six-month finding was checked by a scheduler working independently of Compass, or whether the project went forward on that basis. Both are reasonable questions to put to the company, and both are the kind of check a reader would want before repeating the exercise on a live pursuit.
In quality, 80 emails a day
Quality Manager Christian Fowler offers the most ordinary example in the post, and probably the most transferable one.
I get about 80 emails a day. Instead of digging back through 50 of them, I ask what I might be missing today, and it gives me a list I can check off.
Quality managers on active projects receive a constant stream of submittal responses, inspection notices, RFI answers and superintendent updates. A tool that reads the inbox under the user's own permissions and produces a checklist does not need to be right about construction to be useful. It needs to be right about which messages still need a reply, and the user can verify that in seconds.
In marketing, a pursuit graphic the night before
Senior Marketing Manager Carolyn Haldeman describes a site logistics graphic that was not ready the night before a pursuit submission.
Our site logistics graphic just wasn't there. I used Compass to punch it up with drop shadows, a cleaner font, and a Google Maps background, and it came out far more polished than what we started with.
The example is small, and the company includes it deliberately. Marketing sits on the user list alongside field operations, and the post is making the point that the tool is general across the company rather than specialized for one department.
Voice, mobile and no-code tools
Compass is available as iOS and Android apps. A feature the company calls Compass Voice is built for a superintendent walking a floor who wants to ask a question without stopping to type. Compass Sites gives employees a no-code way to assemble their own small applications on top of the same data connections. Trevor Owen, Reality Capture Manager in the VDC group, says he has built about 10 Sites tools.
You can build your own interactive dashboard. … The biggest thing is to be curious.
Owen's count is worth pausing on. Ten tools from one reality capture manager, built without writing code, suggests that Sites is doing something different from the question-and-answer use that produced the 350,000 figure. A dashboard built once and opened daily does not show up as a question in the logs, and the post does not say how Sites use is counted, if at all. It also means the tool's reach inside the company depends partly on employees like Owen deciding to build for their own teams, which is why his advice ends with curiosity rather than with a feature. Whether the same permission model that governs a Compass question also governs a Sites application that a colleague shares is not described in the post.
Joe Williams, Innovation Solutions Engineer on the Compass team, describes the goal in terms of where project staff spend their hours.
Our project teams should be able to spend their time acting on information instead of assembling it.
How a Compass question moves through Rogers-O'Brien's systems (company description; source: r-o.com, Sept. 1, 2026)
1An employee asks a question through the web app, the iOS or Android app, or Compass Voice. Compass applies the employee's existing access permissions before retrieving anything. It then queries the connected sources the company lists: Procore, scheduling and estimating software, the internal data warehouse, Microsoft 365 and Playbook. It returns an answer, a list or a draft, such as a schedule narrative, an email checklist or a budget-to-estimate comparison. The employee verifies the result and acts on it. This sequence is assembled from the company's description of the tool; Rogers-O'Brien has not published Compass's architecture.
Who built it and what comes next
Compass was built in-house. Wynne led the effort. Terry Mullis, Vice President of IT and Analytics, built the data warehouse that feeds it. James Wagner, AI Solutions Architect, led development with Chloe McKnight, Hagen Fritz, Yazen Barakat and Williams. Richard Anderson, Senior AI Enablement Program Manager, handles adoption across the company. The post names three development themes for the coming year: memory and personalization, agentic capabilities and proactive insights. Each would move Compass from answering questions toward taking actions or raising issues on its own, which raises the stakes on accuracy and on the permission model.
The sequence is worth noticing. The warehouse came first, and Compass sits on top of it. Companies that skip that step tend to end up with an assistant that reads scattered exports and answers inconsistently, because the underlying records disagree with each other.
The staffing tells a similar story. A development team of six under Wagner is a real commitment for a regional general contractor, and the separate enablement role under Anderson is the part most firms leave out. A tool that reads Procore, the warehouse and Microsoft 365 can be technically sound and still go unused if nobody is responsible for showing a superintendent why to open it. The 705 monthly users and the daily-use figure are, in that sense, as much a measure of Anderson's program as of the software. The three roadmap themes will test both sides: memory and personalization ask Compass to remember what a user cares about, agentic capabilities ask it to do work rather than describe it, and proactive insights ask it to speak first. The post offers no timeline for any of the three and does not say which projects will pilot them.
We strive to make Compass a little better every day.
The potential for Compass to have a positive impact on our daily work is almost limitless.
Compass activity counts after one year (company numbers, unaudited; source: r-o.com, Sept. 1, 2026)
Reading the numbers
The most quotable figure in the post is roughly 14 person-years of time returned. The arithmetic is simple: 350,000 questions multiplied by an assumed five minutes each. That assumption carries the whole result. It presumes every question would otherwise have taken five minutes to answer by hand, that none were asked idly or repeated, and that no time was spent checking the answers. The company presents the figure as an illustration, and readers should treat it the same way. The other counts, more than 25,000 specification searches this year and thousands of drawings reviewed, measure activity rather than outcome. The post contains no data on schedule adherence, RFI cycle time, rework or cost variance on projects where Compass was used, and no comparison with projects where it was not.
There is also a selection question. The people quoted are users who found value; a one-year review written by the company would not be expected to quote the superintendent who tried Compass Voice twice and went back to the radio. Daily use by nearly a third of the company is a strong number if accurate, and it also means roughly two-thirds of employees are something other than daily users.
Usage-only reporting is common across contractors describing internal AI tools this year, because usage is what the systems log and project outcomes take longer to attribute. What sets this post apart is specificity: named managers, counted workflows and a stated method for the person-year estimate. That gives an outside reader something concrete to ask about, which is more than most vendor case studies offer.
What should you measure before your firm builds one?
Rogers-O'Brien's post is most useful as a template for questions. If your company is evaluating an in-house assistant or a vendor's version of one, the one-year review suggests where to look.
- Which systems will the assistant read, and will it inherit each user's existing permissions? Rogers-O'Brien says Compass does. Ask a vendor to demonstrate a denied request, not just a successful one.
- Which single workflow will you time before and after? The weekly schedule narrative is a good candidate: it is bounded, recurring and already produces a document you can compare against the old version.
- Who verifies the answers, and how will you log the cases where the assistant was wrong? None of the examples in the post describe an error, which means either there were none worth mentioning or they were not tracked.
- Do you have a data warehouse to build on? Mullis's warehouse predates the assistant, and connecting to a clean store is easier than connecting to raw exports from six systems.
- Who owns the tool after launch? Rogers-O'Brien names seven builders plus an enablement program manager. A tool with no owner stops improving within a quarter.
Then run the same measurement on your own project. Pick the schedule narrative, time it for a month by hand, time it for a month with whatever tool you are evaluating, and compare against what Rogers-O'Brien reported. The company's numbers are its own; yours will be the ones that matter.