If you apply one cancellation assumption across your whole book, you are wrong by close to three times on part of it.
That is not an estimate. In more than 34,500 decided bookings from a single Gulf desk, the cancellation rate ran from under 14% to close to 40% by destination alone. A desk pricing Bodrum risk and Makkah risk identically is mispricing one of them by roughly that factor, on every booking, quietly.
Here is how to fix it without hiring anybody.
Step 1 — Get the denominator right first
This is where most attempts fail before they start.
A cancellation rate is only meaningful if you say what it is divided by. There are at least three defensible denominators and they do not produce the same number:
- Cancelled ÷ decided — cancelled over confirmed plus cancelled. Pending bookings excluded. This is the one I publish, because a pending booking has not decided anything yet and including it measures your response time, not your cancellation risk.
- Cancelled ÷ all rows — everything in the file. On the same data this reads about a point lower.
- Cancelled ÷ confirmed — a ratio, not a rate, and it will flatter you.
Pick one. Write it down. Use it everywhere. Half the arguments about cancellation numbers inside an agency are two people using two denominators and neither saying so — a problem that also explains why published benchmarks disagree.
Step 2 — Build the table
You need four columns out of your back office: destination city, destination country, booking status, and booking value. One or two years of history is plenty.
For each destination: count cancelled, count confirmed, divide cancelled by the sum. Then drop every destination with too few bookings to mean anything. In my own index a city needs a minimum observation count before it is allowed into a ranking, and a higher one before it appears in a destination cut. Pick a floor and hold it. A destination with eleven bookings and four cancellations is not a 36% risk, it is an anecdote.
Step 3 — Tier it, do not model it
Three tiers is enough. More than three and nobody will use it.
- High risk. Long-haul, visa-bearing, booked weeks ahead. On this data: France, Switzerland, Germany, the United Kingdom.
- Standard. Everything in the middle.
- Low risk. Short-haul, visa-light, repeat and pilgrimage traffic, booked days ahead. On this data: the Gulf itself, Makkah and Madinah, Doha.
You will notice these tiers line up almost exactly with how far ahead each destination is booked. That is not a coincidence — it is the mechanism. Elapsed time between decision and travel is where cancellations come from. If you have no cancellation history at all, tier by median lead time instead. You will land in close to the right place.
Step 4 — Apply it in three places, not one
Margin. A high-risk destination carries more operational cost per confirmed booking, because you are handling work that will not convert. That belongs in the price, not in the year-end surprise.
Payment and deposit terms. Where you have room to set them, tier them. A refundable rate on a high-cancellation destination is a decision, and it should be a deliberate one.
Attention. A high-risk booking benefits from a confirmation touch that a low-risk one does not need. Spend the follow-up where it changes an outcome.
What not to do
Do not tell your team that some customers are unreliable. The data does not say that. The same customers who cancel Geneva at close to 40% cancel Makkah at under 14%. The risk is in the trip structure, not the person, and framing it as a customer-quality problem will produce exactly the wrong behaviour at the desk.
Do not rebuild this every quarter. Destination risk moves slowly. Once a year is enough.
And do not automate the pricing decision that comes out of it. Build the table, tier it, put it in front of whoever owns margin, and let them decide. The table is the work. The judgement stays where it is.
Source: the Gulf Outbound Hotel Index, version 1.0 — more than 36,100 B2B wholesale hotel bookings from one Gulf desk. Findings and full tables · Methodology · CC BY 4.0 · 10.5281/zenodo.21796038