Look around a working cleaning business

Everyone who asked and has not booked, on a five stage board, with the four ways a lead gets written. Three things the product does not do are printed on the screen rather than quietly left out.

Every text and email a client sent, in one list, and the missed call the AI receptionist answered: what it quoted, the times it offered, the booking it made, with its own transcript underneath.

The other half of that conversation: the three voice presets as the prompt actually receives them, the five tools the model can call and the fact that there is no sixth, the two checks every outgoing message passes, when it hands the conversation to a human, and the daily spend cap.

Every client the workspace has, searchable, filtered by tag and exportable, with lifetime value summed from that client's own invoices and last and next visit read off the same week the calendar shows. Two limits are printed on the screen rather than glossed: the demo week is all the history there is, and every client in it has exactly one property.

The day board, the route as the office booked it, the route the optimizer returns for the same four stops, the measured minutes between those addresses, and where the optimizer stops short.

Today's route, one visit open with its checklist and photos, the schedule for the fortnight, and what the app does when the signal drops halfway through a job.

The builder, the pricing rule that produced the number, the quote sitting with a client, and the page that client opens to accept it.

A Pro feature nobody without an account has seen: the prompt the model is given word for word, the six fields it is allowed to return, the office review gate between the reading and the sending, and both ways it says no, the reply that failed the schema and the request the pricing rule could not price. The model never returns a number; your own rule prices what it described.

Every invoice in the sample company, what has been paid, and how old the unpaid money is in four aging buckets that add back up to the outstanding figure above them. Also the draft the product builds from completed visits nothing has billed yet, and the CSV exactly as the export route would write it, blank cells included.

Four cleaners on four different pay bases, commission, hourly, hybrid and per visit, every line written by the payroll engine, and the file that goes to the payroll provider.

Rosa Delgado's slice of that same period, on the one page an office prints and hands over: every line the engine wrote with its basis, quantity and rate, the five buckets rolled up over those lines, and the gross. Read it against her row on the pay period and the number is the same one.

What the work is worth, how loaded the crew is, who has not paid, who looks like they are leaving, and what referrals, gift cards and reviews are doing.

The bill going the other way: the plan prices the product would really charge, the trial clock and the day the two roundings disagree, and the SMS credits the workspace spends on its own reminders.

What a household sees on their own link, with no password: their visits, their invoices including one that is part paid, and the few things they are allowed to change themselves.

Who gets asked after a visit and who does not, what the ask says, where each star rating is routed, and the reviews widget on the company website. It also says plainly that nothing in this product writes the words in a five star review.

Two contracted accounts across three sites, a monthly flat contract billing the same whatever the visit count was, and the scored walkthrough their facilities contact reads.

The three questions a new owner answers minutes after signing up, and the empty workspace on the other side. These are the answers that produced the company the rest of the demo shows.

The four step import wizard against a sample export with the mess left in: the columns it guesses, the one it cannot, the duplicate it catches on a phone number, and the rows it refuses.

What is sold and how each service is priced, when clients may book, what a late cancellation costs, who is on the team, and the wording a client receives.

The Stripe connection, shown connected and disconnected at once because the real page only ever renders one of them and a visitor should see both. What BroomDesk stores is two columns on one row; the rest lives in Stripe. The fees are Stripe's own published rates and BroomDesk adds nothing to them, and the screen names the four settings it does not have.

The sites under a commercial contract, what each one bills every month whatever the visit count was, and the paused site that holds no seat and bills nothing. A Business plan feature, so the plan limit and the two refusals it produces are on the screen, composed by the same rules the real action uses.

The form a stranger fills in on the cleaning company's own website, priced by the same engine the office quotes with, five steps to a visit on the schedule.

The owner's side of that form: four widgets, the public key each one carries, the domains it may be framed on, the two shapes of the snippet, and the three ways to take one down.

QuickBooks Online, Google Calendar and the public API. All of them push and none of them read back, which is stated as the conflict rule rather than hidden behind the word sync, and the log keeps the pushes it decided to skip.

The applicant board and its six stages, plus the three things it does not do: nothing feeds it, a trial clean is a stage rather than a booked visit, and moving somebody to hired does not create a team member.

Revenue, churn and running costs across every organization at once. No cleaning company can reach this: the real route returns a 404 to anybody who is not a platform superadmin. The tenants in it are invented, so the figures are invented, and the screen says that before it shows a number.