A reservation is a promise, not an occupied room. Where should a hotel focus its efforts to turn bookings into stays—and when would intervention actually pay?
The management question: prioritize preventable cancellations by volume and economic value, rather than targeting only the segment with the highest cancellation rate.
What this revision adds: estimated cancellation counts, a comparison holding booking-window mix constant, a correction for uneven calendar coverage, and an intervention break-even calculator.
Evidence basis: published dashboard summaries for 119,389 records. The original transaction file and SQL are unavailable, so calculations below are descriptive and approximate where they use rounded cancellation rates. These two historical properties do not represent the hotel industry as a whole.
Hotel filters update the charts and detailed tables. The property comparison always shows both hotels. TA = travel agent; TO = tour operator; lead time = days between reservation and arrival; ADR = average daily rate. CAD amounts retain the original page’s illustrative 1.61 conversion factor, not a verified historical exchange rate.
Cancellation rates among the displayed market segments. Groups: 61.1% versus 37.0% overall in the combined view—about 1.65 times, not more than double.
Published monthly ADR and booking counts. Counts cover a smaller population than the headline; July and August occur three times in the stated window, other months twice.
Published revenue measure and cancellation rate for the three displayed distribution channels. Revenue definition and commission treatment are unverified.
Descriptive cancellation rates across unequal lead-time bands. The pattern does not establish a linear relationship or a causal effect of booking earlier.
The source’s “Non Refund” group has a — cancellation rate. The dataset authors define deposit type from payments recorded before arrival or cancellation; it is not simply a randomized policy assignment at booking. This does not establish that deposits cause cancellations, nor that the result is a known reservation-status error. Verify collection timing, payment treatment and group composition before using it for policy or prediction. Dataset documentation ↗
In the combined view, Online TA has a lower cancellation rate than Groups, but almost three times as many bookings. Multiplying each group’s count by its reported rate gives approximately 20,727 cancellations for Online TA, versus 12,105 for Groups. The two segments represent about 74% of estimated cancellations among displayed segments.
| Market segment | Bookings | Cancellation rate | Estimated cancellations | Share of displayed cancellations |
|---|
Estimated counts = bookings × rounded cancellation rate. Rounding is not sampling uncertainty: a rate rounded to 0.1 percentage point implies a possible count difference of about ±0.0005 × bookings. For Online TA in the combined view, that is about ±28 records. The displayed segments omit two records; shares use only the displayed segments.
Market segment, distribution channel and booking window are overlapping descriptions of the same bookings. Their estimated cancellation counts must not be added together, and the summaries cannot identify the intersection “Online TA bookings over 100 days.”
City Hotel has more long-lead bookings, so part of the difference could reflect its booking mix. Assign both hotels the same proportions across the four lead-time bands while retaining each hotel’s rate within each band.
| Property | Observed rate, recomputed | Common lead-time mix |
|---|---|---|
| City Hotel | 41.73% | 40.70% |
| Resort Hotel | 27.76% | 29.46% |
Booking-window mix alone does not account for the difference. City Hotel’s reported cancellation rate is higher within every band. The common-mix comparison leaves an 11.24-percentage-point gap; customer composition, contract terms, prices and dates may still differ.
Use the combined dataset’s booking share in each lead-time band as a common weight. Multiply each property’s band-specific cancellation rate by these weights and sum. The observed rate uses the property’s own weights. This adjusts one dimension only, using rounded aggregate rates. It is not a causal effect of hotel type or a prediction for changing the booking mix. The source label “100+” is interpreted as 101+ because the preceding band includes day 100 and the source sort key is 101; the original SQL is needed to verify that boundary.
The page states a July 2015–August 2017 window. If that coverage is complete, July and August appear three times and every other month twice. Comparing their raw counts exaggerates their relative volume.
In the combined monthly series, August has 8,638 records versus May’s 7,114. Dividing by month occurrences gives approximately 2,879 per August and 3,557 per May. May becomes the highest-volume month on this limited normalization. It does not become the month with proven highest demand or occupancy.
| Arrival month | Published count | Month occurrences assumed | Count per occurrence | Published ADR (CA$) |
|---|
The population also needs reconciliation. Monthly counts sum to 75,166, not 119,389. That is close to the non-canceled population implied by the headline cancellation rate, suggesting—but not confirming—a non-canceled-booking filter. Until the SQL is recovered, call these published monthly counts, not total demand.
For a stronger seasonal analysis, compare the same calendar period across years, account for days per month and available rooms, separate cancellations from arrivals, and distinguish booking-weighted rates from room-night-weighted ADR. A booking count does not measure occupancy; room inventory and nights are missing here.
| Channel | Bookings | Cancellation rate | Published revenue (CA$) | Revenue per listed booking (CA$) |
|---|
The final column divides the source’s revenue measure by all listed bookings in that channel, including canceled bookings in the denominator. It is not room rate, customer lifetime value or profit. The revenue numerator’s treatment of nights, deposits and refunds cannot be checked without the original query.
Compare net contribution after agent commissions, direct-marketing costs, service costs and refunds before shifting budget. Direct bookings may be different customers choosing different stays. Lower historical cancellations do not show that moving an otherwise identical booking to Direct would change its behavior.
A canceled room may be resold, sometimes at a better rate. Retained deposits may offset part of the loss. Therefore, estimated cancellations multiplied by ADR is not a defensible revenue-loss estimate.
Use the selected hotel’s segment volume as a historical reference cohort. Specify an absolute reduction in cancellation probability, incremental value per prevented cancellation, and cost per booking contacted. These are assumptions—not estimated treatment effects or a forecast.
Expected net value = bookings × [(reduction in percentage points ÷ 100) × incremental value − contact cost]. The value assumption must reflect room resale, variable service costs and changes in deposit or cancellation-fee income; contact cost should include allocated setup costs. The scenario assumes every booking in the selected segment is contacted. Fractional expected counts are possible. A 2-point reduction means 36.7% becomes 34.7%, not a 2% relative reduction.
Risk is not responsiveness: the bookings most likely to cancel may be least affected by a reminder or flexible rebooking offer. Choose interventions based on measured incremental value, not cancellation probability alone.
| Priority | Evidence now | Next decision test |
|---|---|---|
| Investigate Online TA and Groups separately | Largest estimated cancellation counts among displayed market segments | Identify when cancellations occur, whether inventory resells and which offers retain profitable stays. |
| Investigate City Hotel within matched booking windows | 11.24-point gap remains under a common lead-time mix | Compare similar arrival dates, customer segments and rate terms before attributing the gap to property operations. |
| Rebuild the seasonal denominator | Unequal month coverage and a smaller monthly population | Separate arrivals, cancellations, room nights and inventory for a consistent year-over-year comparison. |
With future booking records, randomly assign eligible bookings to the current process or a reminder/reconfirmation offer, respecting group-booking clusters. Define eligibility, contact timing, a meaningful minimum effect and test duration in advance. Measure net contribution over completed stay outcomes, including resale, refunds and contact costs. Track cancellation timing, conversion to stays, guest complaints and repeat bookings as secondary outcomes. A reminder that brings cancellations forward may help resell rooms even if it does not lower the final cancellation rate.
Use only fields available at the moment a decision is made. Exclude final reservation status and final-status date, and verify timestamps for deposits, changes and room assignments. Train on earlier periods and test on later ones, with repeated guests or group reservations handled to avoid leakage. Assess calibration and economic value, rather than accuracy alone. The published summaries cannot support training or validating such a model.