Occupancy is not enough: how to forecast workload department by department

Hotel Management by Pasquale Ascione 23 September 2026 15 min read

Payroll is almost everywhere the largest single line in a hotel’s P&L. The most quoted benchmark is a share of around 30% of revenue, but in the Italian context that threshold is increasingly hard to hold, because of the tax and social-security burden on labour. Which is precisely why the subject does not end with the question “how much am I spending on staff”. The useful question is a different one: how many people are needed, in which department, on which day, and how far in advance that can be known.

In most independent properties the answer is built on a single number, forecast occupancy. If the hotel is going to be 80% full, you roster the “80% team”. It is an understandable approximation, but a systematically imprecise one, because two days with the same occupancy can require very different amounts of work. This article proposes a different method, built on two distinct decision horizons and on one simple principle: every department has its own workload driver, and occupancy is only one of the ingredients.

Two horizons, two different decisions

The literature on service labour planning, starting with the series of articles by Gary Thompson published in the journal of the Cornell School of Hotel Administration, breaks the problem into four steps: forecast demand, translate it into staffing requirements, build the schedules, and control them as the day unfolds. In a hotel, though, that sequence applies twice, on very different time scales.

The first horizon is the long one, where the structural headcount is decided. For a seasonal property it is the team the season opens with, settled between winter and spring. For a year-round property it is the stable core that has to carry the low months as well. These are decisions taken months ahead, that commit contracts and that are not easily corrected.

The second horizon is the short one, where the flexible share is decided: the call-ins, the extra shifts, the additional hours, the reinforcements for a peak week or an event. Here the horizon is days, or a few weeks.

The most common mistake is handling both decisions with the same instrument, expected occupancy, and at the same tempo, that is, at the last minute. The first decision requires a reading of the whole season, the second requires a day-by-day forecast. Neither is solved well by looking only at how many rooms will be sold.

The heart of the method: every department has its own driver

The decisive step is recognising that a hotel’s work does not grow uniformly with occupancy. Each department responds to a different factor, and the forecast has to be built from that factor.

Housekeeping: departures and stayovers, not occupied rooms

Turning a room after a departure takes considerably longer than servicing a room where the guest stays on. Industry references, with wide variation between properties, indicate roughly 30 to 45 minutes for a departure and 15 to 30 for a stayover. The precise figure has to be measured in each property, but the order of magnitude is enough to see the problem.

Take a 40-room hotel with 32 rooms occupied on two different days, therefore at the same 80% occupancy. On the first day there are 20 departures and 12 stayovers; on the second, 4 departures and 28 stayovers. Assuming 40 minutes for a departure and 20 for a stayover, the first day requires about 17 hours of housekeeping work, the second about 12. The difference is worth more than half a room attendant’s day, at identical occupancy. The workload driver in housekeeping is the ratio of departures to stayovers, not the number of rooms sold.

There is a second element, which Thompson describes through the distinction between controllable and uncontrollable work. Turning rooms is partly controllable work: it can be done in a window that runs from the guest’s departure to the next guest’s arrival. If few guests arrive the following day, part of the work can be spread out; if arrivals are concentrated in the early afternoon, the window narrows and more people are needed at the same time. The useful forecast, then, is not only about how many rooms to turn, but by when they have to be turned.

Breakfast and F&B: guests present, not rooms

Breakfast service responds not to occupied rooms but to the people present. Twenty rooms occupied by families of four are not the same as twenty rooms occupied by solo travellers. Distribution over time matters too: a day with many early departures, or with a group coming down all at once, compresses service into a narrow band and requires more staff in the dining room and the kitchen even at the same total covers.

For the restaurant, where there is one, the driver is different again, because it depends on the share of guests dining in-house and on outside customers. In both cases the forecast has to reason in people and time bands, not in rooms.

Front desk: arrivals, departures and time bands

The front desk is the classic case of uncontrollable work: the guest has to be received when they arrive, not when staff are free. The load depends on the number of arrivals and departures and above all on how they are distributed across the day: departures tend to cluster in the morning, arrivals in the afternoon, with peaks driven by transfers, flights, trains and groups. A day with thirty arrivals is very different from a day with thirty stayovers, even if occupancy is identical.

This is why the front desk is planned by time-band coverage, and the useful forecast is the curve of arrivals and departures through the day. It is also the department where inexperienced staff cost the most, because a slow check-in at a peak hour produces queues the guest remembers.

Maintenance and services: the work that can be moved

Maintenance is the department where forecasting pays off most, because much of its work is controllable. Planned interventions, periodic checks and small jobs in the rooms can be placed on low-load days, when rooms are free and the rest of the team has slack. The corrective part, the breakdowns, is not predictable case by case, but it grows with how intensively the property is used. Knowing in advance which days will be light turns maintenance into intelligent filler rather than an emergency that piles onto the peaks.

Structural headcount: seasonal and year-round compared

Once each department’s load has been estimated day by day, the structural headcount decision becomes readable. The most useful instrument is a simple one: rank the days of the season, or of the year, from heaviest to lightest, and look at the shape of the resulting curve. A steep curve, with few very heavy days and many average ones, argues for a contained core and a substantial flexible share. A flat curve argues for the opposite.

The underlying criterion is that the fixed headcount should be sized to a load level that is actually exceeded on most operating days, with the excess covered by flexibility. Sizing it to the peak means paying idle people for weeks. Sizing it to the minimum means relying on call-ins for almost the whole season, with the costs and risks we will come to. The optimum lies between the two extremes and depends on the shape of the curve, not on a standard percentage.

For seasonal properties the problem has a particular dimension. In Italy demand is heavily concentrated: according to ISTAT, between June and September 2023 accommodation businesses hosted 58.6% of the entire year’s guest nights, and in 2024 the third quarter alone accounted for 45.4%. A seaside seasonal hotel sees an even sharper concentration. The headcount therefore has to grow and shrink within the season itself. The most effective answer is not a single team from opening to closing, but staggered starts and finishes: a core that opens the property, progressive additions that track the climb into high season, planned departures as the curve comes down. Seasonal employment law allows this, but it requires the dates to be decided in advance, and that is only possible if the load curve has been estimated. The right of first refusal seasonal workers hold on re-hiring also has an economic value: staff who come back every year know the property and reach full productivity immediately.

For year-round properties the knot is different. The stable core has to carry the low months without becoming excessive, and the peaks are distributed less predictably: trade fairs, events, long weekends, weekends, groups. Here the structural headcount is decided on the basis of the weakest months’ load, and the most effective lever for absorbing the swings is often cross-training: staff able to work across several departments, so that an hour worked can be directed wherever the day’s load requires it. It is a lever that seasonal properties, with larger and more specialised teams, use less.

In both cases the comparison between models leads to the same conclusion: the structural headcount is not decided by copying last year, but on the load curve forecast for the year ahead, built department by department.

On the short horizon, forecasting changes in nature. Two weeks out, a good part of the bookings, arrivals, departures and room composition is already known. As the date approaches, the estimate becomes more precise thanks to new bookings and cancellations. This is where forecasting the load of the coming days produces its most concrete value: saying in advance how many extra hours will be needed, in which department and in which band.

The critical point is how that information is used. The temptation is to treat flexible staff as a reserve that is always available, to be switched on or off at the last minute according to the pick-up. The data say that that reserve does not exist in the quantity imagined. According to the Excelsior information system run by Unioncamere and the Ministry of Labour, for the April-June 2026 quarter the accommodation sector expected 22,500 hires, with front-desk and room-cleaning roles among the most sought after, and a difficulty of recruitment of around 41%. Nationally, in May 2026, 42.9% of open positions were considered hard to fill. Whoever looks for staff at the last minute is looking when everyone else is looking, and often does not find them, or finds them less prepared.

There is a second effect, less visible and more expensive. Research by the Shift Project, a group of sociologists at Berkeley and Harvard who have spent years studying hourly workers in retail and food service in the United States, has shown on a sample of over 1,800 workers followed over time that schedule instability is a strong predictor of quitting. Short notice, shifts cancelled or changed at the last minute, unpaid on-call availability: practices that reduce labour cost in the short run but feed staff turnover. And turnover costs. The study by Tracey and Hinkin for Cornell’s Center for Hospitality Research, conducted across 33 hotels, found that the heaviest component of turnover cost is neither recruitment nor training, but lost productivity during the period when the new arrival is not yet working at full capacity.

From which follows the central claim of this section: forecasting load is there to give notice, not to cut at the last minute. Used well, it allows extra shifts to be confirmed days or weeks ahead, gives flexible staff a calendar they can believe in, and keeps the best people from one season to the next. Used badly, it becomes a cost-compression instrument that turns back on the property itself, in the form of turnover and service quality.

The contractual instruments for flexibility exist. Intermittent employment remains fully usable in tourism and hospitality under the seasonality criteria, as clarified by the Ministry of Labour in August 2025. Occasional work is permitted within strict economic and headcount limits. But none of these instruments solves the upstream problem: knowing when and how much will be needed. Without a forecast, even the best contractual form ends up used reactively.

The fourth step: controlling the variance

The cycle closes with control, the fourth step in Thompson’s sequence, and in independent properties usually the most neglected. Every week, forecast hours should be compared with hours actually worked, department by department, and payroll cost per occupied room with what was expected. A systematic variance means the forecasting model, or the productivity standards, need correcting. An isolated variance means there is an event to understand.

Control by department is also the only way to see where margin is being lost. An overall payroll ratio in line with expectations can hide inefficient housekeeping offset by an understaffed front desk, or the reverse. The aggregate figure reassures; the departmental figure informs.

This step requires a technical condition that is often underestimated: that booking, scheduling and cost data talk to each other. A load forecast built on a booking system, schedules kept on a spreadsheet and costs read only at year-end in the accounts allows no timely control at all. Data integration is the prerequisite, not an implementation detail.

A note on method

The sources used have different levels of reliability, and it is worth stating that once and for all. The data on seasonality and on the labour market come from Italian institutional sources (ISTAT, the Excelsior system). The methodological framework on scheduling and on the cost of turnover comes from academic research published by Cornell. The research on schedule instability concerns US retail and food-service workers, not Italian hospitality specifically: the direction of the effect is solid, its magnitude in our context is not measured. The cleaning times for departures and stayovers are indicative ranges circulated by industry operators and should be replaced with internal measurements. The 30% payroll-to-revenue reference is an industry heuristic, not a statistical benchmark.

Conclusions

Planning staff on occupancy alone means using a number that describes how full the property is, not how much work it will produce. The work depends on departures and stayovers for housekeeping, on guests present and time bands for breakfast, on the arrivals and departures curve for the front desk, and on the availability of light days for maintenance.

On that basis the two main decisions become sounder. The structural headcount is sized on the shape of the forecast load curve, with staggered starts in seasonal properties and cross-training in year-round ones. The flexible share is decided in advance, because staff available at the last minute are scarce and because schedule instability is paid for in turnover and lost productivity.

The value of forecasting, in the end, does not lie in cutting hours at the last minute, but in knowing sooner. Whoever knows sooner can hire better, confirm shifts in advance, keep the best people and place movable work on the right days.

Frequently asked questions

Why is occupancy not enough to plan hotel staffing? Occupancy measures how many rooms are sold, but workload depends on other factors. At the same occupancy, a day with many departures requires appreciably more housekeeping hours than one with many stayovers, and a day with many arrivals weighs on the front desk far more than one where guests stay put. Each department has its own factor that determines its load.

How much permanent staff should a seasonal hotel have? There is no percentage valid for everyone. The criterion is to size the fixed headcount to a load level that is exceeded on most operating days, covering the peaks with flexible staff and with staggered starts through the season. The correct level depends on the shape of the load curve, which has to be estimated department by department.

How is housekeeping staffing requirement calculated? Start from the forecast number of departures and stayovers for each day, multiply each by the standard time measured in the property and divide the total by the effective hours in a shift. You then have to consider the available window — when rooms are released and when new guests arrive — which determines how many people are needed simultaneously.

Is on-call staff the answer to demand variability? It is a useful instrument but not an answer in itself. Excelsior data indicate that in 2026 more than 40% of the roles sought in tourism are hard to fill, and academic research associates schedule instability with higher quit rates. Flexible staff work best when shifts are decided in advance on the basis of a load forecast.

Which indicators should be used to control hotel payroll cost? The most useful are the comparison between forecast and worked hours by department, payroll cost per occupied room, and the payroll-to-revenue ratio read department by department. The aggregate figure can hide inefficiency in one department offset by shortfalls in another.

What is needed to forecast the workload of future days? Up-to-date booking data, with arrivals, departures, stayovers and guest composition, the productivity standards of each department, and the history of hours actually worked. The essential condition is that these data are connected to each other, so that the forecast updates with every new booking or cancellation.

Sources

ISTAT, Movement of customers in accommodation establishments, 2023 (Statistica Today, November 2024); ISPRA, Tourism intensity indicator, 2024 data.
Unioncamere – Ministry of Labour, Excelsior Information System, April-June and May 2026 bulletins.
G. M. Thompson, Labor Scheduling, parts 1-3, Cornell Hotel and Restaurant Administration Quarterly, 1998-1999.
Cornell Hospitality Quarterly, 2014, on the fixed and variable decomposition of housekeeping costs.
J. B. Tracey and T. R. Hinkin, The Costs of Employee Turnover: When the Devil Is in the Details, Cornell Center for Hospitality Research, 2006.
J. Choper, D. Schneider, K. Harknett, Uncertain Time: Precarious Schedules and Job Turnover in the US Service Sector, ILR Review, 2022.
Ministry of Labour, circular no. 15 of 27 August 2025 on intermittent employment.