Guest mix by channel: comparing stay length, spend and cancellations from your own data
A channel report counts bookings; your PMS shows what guests did. Build a four-field extract to compare stay length, spend and cancellations by source. Make one decision you can check next quarter.
Contents
Key takeaways
- A channel report counts bookings; a guest-level extract from your own system shows what those guests spent, how long they stayed and whether they cancelled.
- Compare spend per room-night and the length-of-stay distribution, not revenue per booking, or a channel that sends long stays will look better than it is.
- A cancellation is a timing event: the same rate three months out and 48 hours out are different problems for the same room.
- Print the count next to every rate; a percentage built on a handful of bookings describes those bookings, not the channel.
- One written decision per review, with an owner and a check date, is worth more than five impressions nobody can test.
A hotel guest mix by channel analysis compares what guests from each booking source actually did: how many nights they stayed, what they spent on the folio, and whether the booking cancelled or the guest never arrived. It is built from a four-field extract of your own PMS records, not from channel booking reports.
Your channel report tells you which booking sites sent the most reservations last quarter. It does not tell you whether those guests stayed longer, ate in the restaurant, or cancelled two days before arrival. Those questions belong to the guests, not the bookings, and the only system that holds the answer is your own property management system. After reading this you will be able to build the extract, read it without overclaiming, and turn it into one decision you can actually check.
What does a guest mix by channel analysis actually measure?
It measures outcomes per booking source, not volume per booking source. The channel is the origin recorded on the reservation; everything else is what happened to that reservation afterwards.
Define the channel as the source recorded at the moment of booking: the hotel's own website, a booking site, a travel agency, a tour operator, a corporate account, a walk-in. Freeze that value at booking and never overwrite it later. If a reservation is later reclassified as direct, the original source is lost and the whole history of the channel quietly degrades.
Treat direct as several rows, not one. Walk-in, telephone, email, the website and returning guests behave differently in length of stay, spend and cancellation, and lumping them together hides exactly the part of direct business the hotel can influence through its own marketing and front-desk work.
Volume answers where bookings came from. The guest mix answers what those bookings were worth and how risky they were, which is a different question and needs different numbers. Three outcomes carry the analysis: nights (length of stay), spend (the room plus everything posted to the guest's folio), and status (stayed, cancelled, no-show).
Revenue per booking misleads, because a long stay at a low rate and a short stay at a high rate can produce the same total. Spend per room-night is the comparable figure. Choose one unit and keep it: the reservation, the guest, or the room-night. Mixing units inside one table is the most common way this analysis goes wrong.
| Field | What it holds | The mistake to avoid |
|---|---|---|
| Channel | Source recorded at booking, frozen | Reclassifying a booking as direct later and losing the original |
| Nights | Departure date minus arrival date | Counting the nights of a stay that was cancelled |
| Spend | Room plus restaurant, spa, minibar and other folio postings, net of voids and refunds | Counting only the room rate |
| Status | Stayed, cancelled with its date, no-show | Filing a no-show as a cancellation |
Finally, state what the extract cannot tell you: why the guest chose that channel, and what would have happened without it. It shows what occurred, not the counterfactual.
How do you build the extract?
One row per reservation, four fields, exported from your own system for a fixed period, with the definitions written down before the first export.
Export from the property management system, not from each channel's extranet. The extranet knows the bookings it sent; your own system knows what happened to them. This extract is different work from the morning extranet routine on your booking sites, which checks yesterday's production rather than last quarter's outcomes.
One row per reservation. Split group blocks into the individual reservations that actually arrived, and keep the block itself out of the channel comparison. Rooms taken from a block are pickup, a separate measure with its own meaning.
Fix the period twice: by booking date and by arrival date. The first answers how far ahead people book, the second answers what the month looked like. Net the spend: voids, reversals and refunds reduce it, and a deposit is not spend until it is applied to the folio. Record the cancellation date, not only the cancellation flag, because the date is what makes the timing analysis possible.
Reconcile the row count against your system's own reservation count for the period before you analyze anything. If the two do not match, the definitions are wrong somewhere, and every figure downstream inherits the error. Keep the raw file: every number in the review must be reproducible from it six months later.
ChecklistYour extract checklist
- Channel frozen at booking
- One row per reservation
- Nights taken from arrival and departure dates
- Spend net of voids and refunds
- Cancellation date recorded, not just the flag
- No-show flagged separately from cancellation
- Period stated by booking date and by arrival date
- Row count reconciled to the system total
- Definitions written in a one-page note and dated
How do stay length and spend differ by channel?
They differ mostly because different channels attract different trip types, and the differences only become readable when you compare medians and per-night figures instead of totals.
Median before mean. One very long stay or one very large folio moves an average enough to change the ranking of channels. The median length of stay and the median spend per room-night are more stable summaries, and the two together tell you more than either alone.
Report the length-of-stay distribution, not just the average: the share of one-night stays, of two to three nights, and of longer stays tells a revenue manager more than a single number. Spend per room-night is comparable across channels; spend per booking is not, because it folds length of stay into the figure.
Separate room spend from everything else posted to the folio: restaurant, spa, minibar, extras. A channel that fills rooms and never fills the restaurant is a different customer, not a better one, and the difference matters for pricing, packages and what you show each channel.
Compare like with like: the same months, a similar room-type mix, the same board types. A channel that happens to sell suites will always look richer. Put lead time in the same table, because bookings made far ahead tend to be longer and cheaper per night, which is a property of the trip, not of the channel. Lead time belongs to the family of demand signals such as pace and lead time that revenue managers already track.
Write down what you expected before you open the file. Otherwise the story gets fitted to the numbers after the fact, and the review confirms whatever the reviewer already believed.
What do cancellations and no-shows look like by channel?
A cancellation rate on its own says very little. The useful measures are when the cancellation arrived and whether the room was sold again.
Define the denominator and print it every time: cancellations divided by cancellations plus stays, for the same arrival period. A rate without its denominator is not a finding.
Sort cancellations into timing bands: more than 30 days before arrival, 7 to 30 days, under 7 days, under 48 hours. The same rate in different bands is a different problem. A room released 60 days out usually sells again; a room released at 48 hours often does not. The cost of a cancellation lives in that difference, so track resale alongside the rate.
A no-show is not a cancellation. There was no notice, and whether it is chargeable depends on the terms of the rate the guest booked, which the hotel sets and writes into that rate. Keep the two statuses separate in the extract and in the review.
The cancellation policy is part of what a channel offers. Two channels at a similar rate with different policies are not comparable, and the policy should sit in the table as a column. Link a cancelled booking that comes back at a different rate, or note it; otherwise the same guest is counted twice and the channel looks worse than it is. Keep group blocks out of this comparison entirely.
How small is too small to compare channels?
When a channel produced only a handful of bookings in the period, its figures describe those bookings and nothing more.
Count first. Print the number of bookings beside every rate, in every table, without exception. A percentage on a small base swings on a single booking, and the same channel can look best one quarter and worst the next while nothing about it has changed.
Pool periods instead of comparing single months: a year of a thin channel is more informative than one busy month of it. To separate a shift from noise, ask whether the movement is within what two or three bookings could produce; if it is, treat the channel as unchanged.
Set the hotel's own rule in writing and apply it every quarter. For example, a hotel might decide to act only on channels with at least 30 bookings in the period, and to say so in the review when a channel falls below that. The threshold is the hotel's choice; what matters is that it is fixed before the data arrives.
Watch outA rate without its count is not a finding
A channel with a handful of bookings can show any percentage at all, and a decision taken on that number is a decision taken on noise. Print counts, pool periods, and fix the hotel's own minimum before the review.
The same caution applies to review scores, NPS and complaint rates as small-sample measures, which move for the same reasons. Write what your own file shows, with the count, not what the channel is: "in our data for this period" rather than "this channel sends short-stay guests".
What decision can one review support, and how often should you repeat it?
One decision per review, written down with an owner and a check date. The review itself runs quarterly, on the same definitions each time.
Pick the decision the data can actually carry: the cancellation terms of a rate plan, a minimum-stay rule, where inventory is allocated, or what a channel's content says about the property. One decision, not five; five changes made at once make it impossible to know which one moved anything.
Write the expected effect and the evidence you will accept before you act, not after. Set the check date and re-run the same extract for the next period. If the definitions change, the comparison is lost.
TipOne-page decision log template
Period; channel; what the extract showed, with counts; the decision taken; the owner; the expected effect; the check date; the outcome at the check date. One page, one decision, dated and signed.
Quarterly for the mix, monthly only for cancellations close to arrival, annually for contract terms. Different questions, different rhythms. Between reviews, keep watching the extranets and which review sites a hotel in Türkiye should watch, so the quarterly file is not the only place the hotel learns something about a channel.
If you are evaluating systems that can produce this kind of extract, it helps to know what to look for: a hotel operating system that keeps reservations, folio postings and status in one place, and a distribution setup where each channel's bookings arrive already labelled with their source. The extract itself is your own work either way; the system decides how much of it is manual.
What to do next
Take one quarter of reservations out of your own system, keep the four fields, and write down one decision with an owner and a check date. Length of stay, spend and cancellations only become useful when they are read together, on the same extract, with the counts in front of you, and repeated on a fixed rhythm so this quarter can be compared with the last. Start with the definitions note, not the spreadsheet: the note is what makes next quarter comparable.
Frequently asked questions
Should I compare channels by revenue per booking or per room-night?
Per room-night for pricing and mix, because it removes length of stay from the figure and makes channels comparable. Per booking is still useful for cost-to-serve questions, such as how much handling each reservation takes. Report both, always with the number of bookings behind them.
How do I handle a booking that was cancelled and then rebooked at a different rate?
If your system records the link, keep it and count the guest once. If it does not, count the original as cancelled and the new one as a new booking, and write a note in the review saying so. The point is that the reader of the table knows which convention was used.
Does a channel with a high cancellation rate cost me money?
It depends on timing and resale, not on the rate alone. Measure how far ahead the cancellations arrived and whether the rooms were sold again. A high rate of cancellations made months out may cost almost nothing; a low rate close to arrival may cost several rooms.
How many bookings do I need before a channel comparison means anything?
There is no universal threshold. The hotel should fix its own minimum in writing before the review and print counts beside every rate. Below that minimum, pool several periods or treat the figures as an anecdote rather than a finding.
Where does direct fit in a channel comparison?
Split it. Walk-in, telephone, email, the hotel's own website and returning guests behave differently in length of stay, spend and cancellation, and lumping them together hides the part of direct business the hotel can actually influence.


