Field notes.

Continental's margin case leans on three levers, and they are not equally supported. A pitch deck can outrun the work unless somebody separates the measured from the modeled. This note does that for robotics, the lever with the most numbers behind it.

Two things. First, an operating independent's real P&L: a good one, already at about 12 percent net operating income on a run-rate near $19 million, with direct labor and burden at 60.9 percent of revenue and a 34.3 percent gross margin. Second, the bids and ERP records of a second operator: 415 bids, 147 of them Class A, with the hours on hard-surface rows separated from carpet and fixtures and multiplied out by frequency.

The reproduction against the locked workbook came within half a percent of the original full-capture lift. That gate passed.

Seventy percent capture of hard-floor hours is a planning case; no live robot route has demonstrated it yet, and dust-mop work, edges, obstacles and mats stay human. Two dollars a machine-hour is a placeholder with no manufacturer quote behind it, and three dollars an hour of teleoperation is an input. Ten percent exception labor is an input too, and a harsh one: it is ten percent of the labor dollars removed. The savings only print after each subcontractor agreement is re-cut by the hard-floor share, which is assumed and has not been executed. And carrying a nine-point lift measured on one operator's contracts across to another operator's employee labor is one analogical step: the two sit at similar field intensity, but nobody has parsed the second operator's bids.

On accounts that already bill a scrubber, and where the bid matches the ERP, robotics takes about nine points of cost out of the contract at 70 percent capture. Applied to a P&L that is already at 12 percent NOI, field labor goes from 61 to 52 percent of revenue. Pilot sites with a company machine on the floor can print closer to fourteen points. Those pilot accounts are a slice of the book, and the savings only print after each subcontractor agreement is re-cut.

That profit goes from 12 to 71 per hundred dollars of today's revenue. Twelve to about thirty is the supported compound of robotics and the software overhead cut; the data layer on top of that is a thesis overlay, and it is presented as one.

The National Human Activity Pattern Survey, run for the U.S. Environmental Protection Agency, found that Americans spend about 90 percent of their time indoors: in homes, offices, schools and clinics. Most of the air a person breathes in a lifetime is indoor air, and most of the surfaces they touch are indoor surfaces.

Healthy indoor spaces are clean ones; the hygiene of a building starts with the crew that cleans it. The Energy Information Administration counts about 5.9 million commercial buildings in the United States, and almost all of them are on a weekly service cycle: somebody comes at night, empties the bins, scrubs the floor and leaves before the occupants arrive.

Continental's mission, secure the health of human spaces, is the plain reading of those two facts. If people live indoors, the condition of the indoors is a health question, and the work that keeps a building in condition deserves the same seriousness as any other health-adjacent trade. That is why the scope is written down and the inspections carry scores.

Walk into the office of a good commercial cleaning company and you will find the operation on the desk: clipboards of inspection sheets, floor plans marked up in highlighter, a spreadsheet of accounts with red and green cells, and two or three pieces of software that do not talk to each other. The scope of work lives in a contract PDF and the schedule lives in the supervisor's head. Inspections stay on the clipboard until somebody types them in, which is sometimes.

The operators did not choose this; a field with thin margins and 200 percent turnover never got purpose-built tools. The client feels it anyway. A tenant finds the missed restroom before the supervisor does, and a subcontractor's insurance lapses with nobody noticing until the claim.

The contract's scope becomes the system's scope, with each surface held as a record that carries its frequency and its acceptance standard. Schedules are generated from those records, and inspections are scored against them on a phone the same night. Work orders, subcontractor compliance, quotes and leads sit in the same place, so the client's report is drawn from the data the operator actually runs on. The software already runs in operating cleaning companies.

The margin case for software is still a direction: the deck's admin line goes from 22 to 12 per hundred dollars of revenue, and that size stays a claim until an acquired company has run on the system for a full year. Continental presents it as a path.

In a 2024 study of perceived occupational status across the 22 O*NET job families, building and grounds services came last. The Bureau of Labor Statistics puts the field's median wage near the bottom of the table, and industry bodies (ISSA and BSCAI) report annual turnover in cleaning labor of around 200 percent. A night crew that starts in January is, on average, a different crew by summer.

Turnover sits behind the field's other problems. Training does not compound when the trainee leaves in April. And quality cannot be inspected into a building by a supervisor who is also covering a route on paper forms.

Continental's sequence is deliberate. Software gives the crew a schedule and a checklist, which is most of what turns a shift into a job, and the records it leaves make the work visible to the client who pays for it. Robotics then takes the hardest hours (the scrubber on the concourse at two in the morning) off the crew and onto a machine an operator drives from a chair. The people who stay get better work. The turnover number is the one to watch.

Run the same hard-floor hours through two models: a scrubber driven remotely by an operator at three dollars an hour, and an autonomous scrubber at two dollars a machine-hour plus ten percent exception labor. The teleoperated case comes out slightly ahead at those defaults. Autonomy only wins below about $1.30 a machine-hour, or with the exception labor cut.

So Stage 1 of Continental's robotics program is teleoperation from an operations center. A machine on the floor with a person in a chair is a solved problem today, while autonomous navigation of an occupied lobby is still research, and each teleoperated route leaves behind the data autonomy will eventually learn from. The nine points in the robotics note come from moving hours onto machines; at today's inputs the model actually prices the remote operator slightly cheaper than autonomy.

Stage 1, the plan and the drawings, is in progress. Stage 2 is the prototype and Stage 3 is deployment on client floors; neither has started. The status is on the home page and will change there first.

Most cleaning contracts read like marketing copy: "high standards" and "your complete satisfaction." A scope of work reads like a checklist. For each area of the building it names the surfaces, the task, the frequency and the criterion for judging the task done, and it is priced per square foot, so the client can see what a change in frequency costs.

Restrooms, floors 1–14 · hard-surface floors · damp mop with neutral cleaner · nightly, Monday to Friday · acceptance: no visible soil, no standing water, corners and edges included · inspected weekly, scored 1 to 5, reported monthly.

Multiply that across the building and the contract becomes a document a supervisor can inspect against and a new crew member can be trained from. It is also the record the next contract is priced on. Continental writes each contract this way, and the software stores every line as a record.