Housekeeping schedule optimizer

A housekeeping schedule optimizer only earns trust if it reads one frozen arrival list. Lock that list first. Until arrivals are frozen for the day-part you are staffing, every room order and every attendant count is a draft.

The property management system is the source. Hotels that run Oracle Hospitality, Mews, Cloudbeds, Agilysys, or another PMS in that class all expose due-in and in-house records. Pull the extract for the floors you own, stamp the lock time, and refuse later PMS chatter until the next lock. Walk-ins, early check-ins, and room moves that land after the lock wait for the next cycle. They do not rewrite a live board.

Locking is a quality step. Arrival time, room type, and occupancy vintage only mean something if they refer to one snapshot. If the file is still moving, a cite that looked true at 06:00 can describe a different stay by 08:00.

Do not treat an unlocked forecast as arrivals. A pickup curve, a group block, or a same-day cancellation list is not a stay. If the record does not name a room and an expected arrival window, it does not earn a slot.

Departures are a separate extract. When you need due-out timing to know when a room can enter the turn, read it from departure time optimizer alongside this board. Do not let the housekeeping model infer a checkout from a gap in the arrival file.

Once the file is locked, the model proposes a cleaning order and a staffing level. Every proposed slot must carry three cites: arrival time, or a stated due-in window from the locked extract; room type as the PMS coded it; and occupancy vintage, meaning when the room last held a guest, or that it is vacant with no last-stay date in the extract.

An assignment without those three cites is not an assignment. It is a guess. Reject it.

Arrival time tells you whether the room is a same-day due-in or a later wave. Room type tells you linen, crib, and setup load, and it is the bridge to guest preference-based room setup once the board is published. Occupancy vintage tells you whether you are turning a room that just checked out, a stay that has been empty, or a unit with no last-stay record at all.

Staffing follows the same cites. Headcount for a floor is the count of cited due-ins plus cited vacant turns you have evidence for, not a target the model invented to fill a roster. If the locked file is thin, the roster is thin. Do not pad attendant lines to match yesterday's crew. Room type mix belongs in the cites too: a board of kings and suites is not the same work as a board of twins, even when the room count matches.

Sequence the cited rooms by the work the cites imply. Earlier arrival windows go earlier. Heavier room types go where the attendant mix can absorb them. Older occupancy vintage, a room empty for several nights, is not automatically first or last. It is a cite the manager uses to decide inspect versus full turn. The model does not get to hide that decision inside a rank.

A slot with no arrival cite fails this test even if the cell looks complete. If the cell shows a room number and an attendant name but no arrival time, and no explicit vacant-turn vintage from the extract, strike the cell. Do not send someone there on the model's say-so.

Empty stays empty. If the locked extract has no arrival for a room, the optimizer leaves that room's slot blank. It does not invent a stayover. It does not invent a checkout. It does not borrow a neighbor's due-in to keep the column looking full.

A blank is a quality signal. It means there is no arrival time, no room-type-backed stay, and no occupancy vintage that justifies a turn on this board. Housekeeping can still inspect a vacant that operations flags by hand. That walk is a manager add, recorded as a manager add, not as a model assignment.

Inventing a checkout is the failure that looks most like diligence. Checkout is a departure event. If the locked arrival file does not also carry a verified due-out for that room, you do not assume the room will be vacant for a full turn. Do not let the housekeeping optimizer mint due-outs to justify work.

Stayover is the same rule in the other direction. A stayover exists only when the in-house record says the guest remains. Silence is not a stayover. Filling a blank with a probable stayover so an attendant has a line is fabricating occupancy vintage.

Take a Thursday lock at 06:30 for floors 4 and 5. The extract shows room 412 as a king due in at 15:00, last occupied two nights ago; room 418 as a twin due in at 16:30, last occupied last night; room 501 as a suite due in at 14:00, vacant since Sunday; and room 414 with no due-in, no due-out, and no in-house stay.

The optimizer sequences 501, then 412, then 418. Each line cites the arrival window, the room type, and the occupancy vintage above. Room 414 stays blank.

A manager might still walk 414 if the floor supervisor wants a vacant inspect. That walk is not on the model board. The model does not fill 414 with a probable stayover or a likely checkout to keep attendant 3 busy. Attendant 3 either picks up overflow from the three cited rooms or is not rostered for that wave.

This illustration is not a measured result. It shows the cite rule and the blank rule on one board. If 414 later gets a walk-in, that stay waits for the next lock. Putting it on the live Thursday morning board would be treating a draft as published work.

The optimizer proposes. The housekeeping manager publishes. Until that person accepts the board, attendants do not run it.

Publishing means the lock timestamp is visible, every filled slot has the three cites, blanks are still blank, and the manager has scanned for the remaining failure modes. No slot without an arrival cite or a documented vacant-turn vintage. No copy of the board already in attendants' pockets as if it were live.

Treating the board as published is the operational failure. A printout on the desk, a screen in the locker room, or a message in the attendant chat does not publish the schedule. Those copies go out after the manager signs, and they carry the lock time so a later PMS change cannot masquerade as the signed board.

After publish, the floor can consume the order. Preference setup for the cited room types waits until the board is signed. Rooms whose vintage or condition looks off can be handed to maintenance issue predictor instead of being forced into the same turn as a clean due-in. Front office queue pressure is not a reason to un-blank a slot. If arrivals bunch at the desk, that is a queue wait time predictor problem, not a license to invent rooms for housekeeping.

Re-locks happen on a stated cadence: mid-morning, after the first wave, before the evening due-ins. Each re-lock produces a new draft. Each new draft needs a new publish. Yesterday's signature does not cover today's extract.