Hotel operations, defined plainly
Fourteen terms that come up in every conversation about hotel demand forecasting, labor cost and scheduling, each defined in two or three sentences with a worked hotel example. Written for operators, quotable by anyone.
RevPAR
Revenue per available room: rooms revenue divided by rooms available, whether sold or not. A 200-room hotel earning US$24,000 in rooms revenue tonight has a RevPAR of US$120. It blends occupancy and rate into one demand-side health number.
GOPPAR
Gross operating profit per available room: GOP divided by available rooms. Unlike RevPAR it feels costs, labor above all, which is why two hotels with identical RevPAR can be US$30 apart in GOPPAR. Scheduling quality shows up here.
ADR
Average daily rate: rooms revenue divided by rooms sold. ADR tells you what you charged; RevPAR tells you what you earned across the whole house.
CPOR (labor cost per occupied room)
Total labor cost divided by occupied room-nights. The single best scheduling KPI: at 78% occupancy, a 250-room hotel with a US$95,000 labor week runs about US$70 labor CPOR. US hotels saw labor CPOR rise 12.8% year on year (CBRE, US benchmark).
Hours per occupied room
Paid labor hours divided by occupied rooms, CPOR's wage-neutral sibling, useful for comparing properties across markets with different pay levels. Housekeeping alone typically runs 0.4-0.7 hours per occupied room depending on segment.
MAPE (mean absolute percentage error)
The standard forecast-accuracy measure: the average size of forecast misses, as a percentage of actuals. A 90-day manual hotel forecast typically runs ~28% MAPE; 7-day AI forecasts run near 8%. Every point of error becomes over- or under-staffing.
Demand forecasting
Predicting future hotel demand, arrivals, departures, occupancy, covers, from reservations on the books, pickup pace, seasonality and events. Forecasting is only operationally useful at department-and-hour granularity: "82% occupancy Saturday" cannot staff a breakfast shift; "310 covers between 07:00-10:00" can.
Forecast horizon
How far ahead a forecast looks. Error grows fast with horizon (8% MAPE at 7 days to ~28% at 90 days), which is why rosters locked weeks in advance are structurally wrong and need demand-triggered updates.
OHIP (Oracle Hospitality Integration Platform)
Oracle's API platform for OPERA Cloud, exposing 3,000+ REST APIs to 1,200+ integration partners. It is how third-party systems read reservations, arrivals and occupancy from OPERA Cloud without touching the PMS database. HotelCadence connects read-only via OHIP in 1-2 weeks.
PMS (property management system)
The hotel's system of record for reservations, guests, rooms and billing, OPERA Cloud, Mews and Cloudbeds are the three HotelCadence connects to. The PMS already contains tomorrow's demand; most hotels just never route it to the roster.
Labor standard
The agreed ratio between workload and staffing that scheduling optimizes against, for example "one front-desk agent per 30 hourly check-ins" or "28 minutes per room-turn". Standards encode the service level, so cost optimization never silently degrades service.
Roster (rota, schedule)
The department-level plan of who works which shift. A demand-matched roster is built against the hourly forecast; a calendar roster repeats last week. The difference between them, priced out, is typically 6-12% of labor cost.
Overtime leakage
Overtime paid not because demand required it, but because the base roster missed the demand curve, under-staffed peaks patched at premium rates. Manual scheduling drives roughly 30% of hotel overtime cost.
Service coverage
The share of demand hours where staffing met the labor standard. HotelCadence pilots report labor cost and service coverage side by side, so a saving that quietly cut service would be visible immediately, the reassurance metric behind "without cutting guest service".
See the numbers behind these terms in the APAC hotel labor statistics, or put them to work in the labor savings calculator. Deep dives: hotel demand forecasting software and AI hotel staff scheduling.