Hotel forecast: estimating the next 90 days from what is on the books

A hotel forecast is the revenue team's current estimate of where each future stay date will finish, rebuilt every week from rooms on the books, pickup still expected, cancellations and allotment releases, by segment. It differs from the budget, which is fixed for the year, and from the target, which is an ambition.

A Friday 90 days out has 40 rooms on the books and the general manager wants to know whether the hotel will fill. No report answers that question. The answer is an estimate a person builds from four inputs, writes down, and rebuilds next week. This guide shows how to do hotel forecasting by hand, segment by segment and stay date by stay date, how moving bayram dates change the picture, and how to keep a log that makes the method improve.

What is a hotel forecast, and how does it differ from a budget?#

A forecast is the revenue team's honest estimate of where each future stay date will finish, and it is rebuilt every week. A budget is a fixed annual plan and a target is an ambition; neither is revised because Tuesday was quiet.

Three numbers, three jobs#

The budget allocates cost and revenue across the year and is approved once. The target states what owners or management want to reach. The forecast states what the team currently believes will happen, date by date. Each has its owner, its rhythm and its use, and confusing them is the most common failure in hotel forecasting.

The forecast explains variance; it does not hide it. When the month closes below budget, the forecast history shows whether the miss came from pickup, cancellations, releases or rate, which is the only information that changes next month. Never let the forecast be edited to match the budget: a forecast bent toward the plan stops being a decision tool and becomes a report nobody acts on.

One owner, one version. Name the person who publishes the weekly number and the day it is published; competing spreadsheets from sales, reservations and finance destroy trust in all three. If you are also building next year's budget, see our guide to the hotel budget: how to build next year's plan month by month.

Why the forecast is built by stay date#

A forecast is built by stay date, not by booking date and not by month. A booking taken in January for 12 March belongs to 12 March, and a monthly total hides the fact that one weekend is full and the rest of the month is empty.

The horizon is rolling. Ninety days is long enough to change rate, restrictions, rosters and purchasing, and short enough that the estimate stays close to reality. Group and contract decisions need a longer, looser view on top of it, kept separate from the 90-day number so the two do not contaminate each other.

What goes into a hotel occupancy forecast?#

Four inputs per stay date and per segment: rooms already on the books, pickup still expected, cancellations still expected, and committed rooms likely to be released. Events and holidays sit on top as adjustments, not as a fifth column.

On the books, pickup, cancellations, releases#

Rooms on the books means confirmed reservations for that stay date, whatever day they were taken. Pull them by stay date and by segment code, not by arrival date of the reservation, or long stays and split stays land in the wrong column.

Pickup is the new business that will arrive between today and the stay date. Estimate it from your own history for the same lead time and the same segment, never from a published industry curve; a resort in July and a city hotel in February have nothing in common. To read the pace of what has already arrived, see our guide to the hotel pickup report: how to read booking pace day by day.

Cancellations are not a surprise, they are a rate. Track your own cancellation share by segment and by lead time, and subtract it from the transient on-the-books figure rather than from the total. Releases are the third subtraction: allotment rooms an agency holds but will hand back by a deadline. Forecast them as a separate line so the number can be checked later, not folded silently into the total.

Known events, bayram dates and school holidays adjust the pickup estimate for specific dates. Write the adjustment and its reason in the grid; an unexplained uplift cannot be reviewed.

Why the denominator must be written down#

Cap the result at sellable rooms, meaning physical rooms minus out-of-order and maintenance, and state that denominator next to the occupancy percentage. An occupancy figure without its denominator cannot be compared with anything.

How do you forecast hotel occupancy by hand, segment by segment?#

Build one grid: stay dates down the side, segments across the top, and four columns per cell for on the books, expected pickup, expected cancellations and expected releases. Add the first two, subtract the last two, and cap the result at sellable rooms.

Segment codes come first. Direct, OTA, agency allotment, group and corporate each behave differently on pickup and cancellation, so a single blended ratio produces a number that is wrong everywhere. If your segment codes need work first, see our guide to hotel market segmentation: how to set up segment codes.

Work cell by cell, not month by month. For each stay date and segment, start from rooms on the books, add the pickup you expect for the remaining lead time, subtract the cancellation share, subtract the rooms you expect to be released. Use your own history as the ratio and write the ratio down. If direct pickup in the last four weeks of a 30-day lead time averaged a certain number of rooms, that is the number to use this week; next month it may be different, and the log will show it.

Handle groups separately: a contracted block is not on the books until it is picked up, so forecast the block at its expected pickup, not at its contracted size, and note the cut-off date.

A worked example, for example#

For example, take a hypothetical 120-room hotel with 6 rooms out of order, so 114 sellable rooms, forecasting Friday 12 March 2027:

That is 99 rooms against 114 sellable rooms, 86.8 percent. All figures are invented for the example.

The two checks before publishing#

Check the arithmetic twice: the segment totals must equal the hotel total, and no date may exceed sellable rooms. A forecast above capacity is not optimism, it is a data error. Publish the result as rooms and as an occupancy percentage with the sellable-room denominator, and keep last week's number beside it so the change is visible.

What do allotment release dates do to a forecast?#

A release date hands unsold committed rooms back to the hotel, so those rooms are not sold and must not sit in the forecast as if they were. Forecast them on their own line, with the date they return.

Committed, taken and returned#

An allotment is a promise of availability, not a sale: the agency holds rooms until a deadline and returns what it has not sold. Until pickup happens, the rooms are neither revenue nor free inventory. Read the contract for the release period, the minimum-stay or stop-sale terms and the cut-off date, and put every release date that falls inside the 90-day horizon into the grid.

Forecast two numbers for each allotment: rooms the agency will probably take by the deadline, and rooms likely to come back. The second number is inventory you can resell, and it changes both the forecast and the rate decision. Do not treat a returned room as sold because the agency is a long-standing partner; the forecast follows pickup, not relationships. If releases are large and repeated, the contract is mis-sized: the fix belongs in the next negotiation, not in the forecast. For how much to commit in the first place, see allotment vs free sale: how much inventory to commit to agencies.

The resale gap after a release#

Expect a resale gap. Rooms returned late in the lead time sell at whatever the market will pay then, which is usually less than the rate you would have set three months earlier. The forecast should show the returned rooms at a realistic rate, not at the contract rate.

How do moving holidays change a 90-day forecast?#

Bayram dates move against the calendar every year, so the same date last year tells you little. Forecast by the position of the holiday in the week, not by the date number, and mark the eve separately.

The eve is a half day#

Under Law No. 2429 on national and general holidays, Ramazan Bayramı lasts 3.5 days and Kurban Bayramı 4.5 days, each starting at 13:00 on the eve, called arefe, which is why the eve behaves like a half travel day rather than a normal afternoon (Mevzuat Bilgi Sistemi, consolidated text of Law No. 2429 (opens in a new tab), July 2024).

According to Diyanet's 2027 religious days calendar (opens in a new tab), Ramazan Bayramı runs from Tuesday 9 March to Thursday 11 March 2027 with the eve on Monday 8 March, and Kurban Bayramı runs from Sunday 16 May to Wednesday 19 May 2027 with the eve on Saturday 15 May.

Source: Diyanet 2027 calendar and Law No. 2429.

When two holidays meet#

The last day of Kurban Bayramı falls on 19 May, which is already a general holiday under the same law, so the return pattern differs from a year when the two do not meet. Check the overlap before copying last year's shape.

Map each holiday onto the weekday pattern: a bayram that starts on a Tuesday produces a different arrival and departure curve from one that starts on a Saturday, and the eve half-day shifts the first peak. School holiday dates change by year and are set separately; confirm the current calendar with the Ministry of National Education before adjusting pickup, and do not assume a bridge day or administrative leave around any holiday. Write the adjustment and its reason on the affected dates in the grid, so that next year's team can see what was assumed and whether it held.

How do you measure forecast error and keep a forecast log?#

Write every forecast down before the date arrives, compare it with the final number after the date closes, and read the error by segment. Without that record, the method never improves and nobody can tell whether the forecast is worth reading.

Log the forecast at the moment it is published, not afterwards. A number reconstructed from memory always looks better than it was.

Bias versus noise#

Record both signed and absolute error. Signed error shows bias, meaning the team is consistently optimistic or pessimistic, while absolute error shows size, meaning how far off the estimate typically lands. Attribute the error to an input: if pickup was close but cancellations were double the assumption, the fix is in the cancellation share, not in the whole method.

Review monthly, not weekly. Weekly reviews chase noise; a month of records shows whether one segment or one lead-time band is consistently wrong.

One change at a time#

Change one input at a time and keep the old ratio beside the new one for a few weeks. Changing three assumptions at once teaches nothing. Share the error openly with sales, reservations and finance: a forecast whose misses are discussed is trusted; one whose misses are hidden is ignored.

What should the forecast change?#

Rates, restrictions, staffing and purchasing, in that order of reversibility. A forecast that changes no decision is a report, and reports do not need rebuilding every week.

Rate and restrictions first#

Rate and restrictions are the fastest levers: if a date is forecast well below the pace needed, act on rate, minimum stay, closed-to-arrival or closed-to-departure while there is still lead time to sell. The mechanics of these levers are covered in our guide to hotel rate restrictions explained: minimum stay, CTA and CTD.

Protect the strong dates before discounting the weak ones. A date forecast near capacity should carry restrictions that stop cheap business from displacing better business, and the forecast is what tells you which dates those are. Group decisions use the forecast too: a block that fills dates already forecast near capacity displaces transient business, and the displacement is visible only if the forecast exists first.

Staffing and purchasing from the same grid#

Staffing follows the forecast by date: housekeeping hours, front desk shifts and restaurant covers are set from rooms and guests expected, not from last month's average. Purchasing follows the same grid: food, beverage, laundry and amenities are ordered against forecast occupancy and the expected mix of board types, which is why segment detail matters beyond revenue.

Set a review rhythm: publish on the same day each week, review the month once, and change the method only when the log shows which input is wrong.

What to do next#

Start this week: build one grid of stay dates for the next 90 days, by segment, with on the books, expected pickup, expected cancellations and expected releases, and write the number down before the week ends. Next week, build it again and compare. The forecast is not a prediction to be right about; it is a weekly estimate whose error, once measured, tells the revenue team which input to fix and which decision to take while there is still time to take it.