Refundability is a product rule that determines what a customer may recover and what the provider owes after a qualifying cancellation. Cancellation hazard is a time-indexed behavioural probability conditional on a booking still being active. A retained-demand forecast needs both, but a refund rule is not itself a cancellation probability.
This specialist guide belongs to the Yield Management root. It also connects to The Gold Standard of Education | Learning That Transfers into Life and the How X Works hub. The worked cases are original teaching examples with stated inputs so their conclusions can be checked and transferred.
eduKateSG · Yield Management
Find your next learning step
Use this guide when a forecast gives every flexible booking one cancellation percentage, treats non-refundable bookings as certain arrivals, or subtracts expected refunds from physical demand. The worked cases keep product terms, risk sets, lead-time intervals, cancellation events, exchanges, retained bookings and cash liabilities in separate ledgers.
ROUTE 1 · CHAPTERS 1–3
Set up the distinction
Define the quantities, denominators and first worked case.
ROUTE 3 · CHAPTERS 6–8
Change the context
Change one premise, rebuild the result and track what remains comparable.
ROUTE 4 · CHAPTERS 9–10
Test limits and misconceptions
Separate the focused result from nearby claims requiring different evidence.
ROUTE 5 · CHAPTERS 11–12
Verify and explain
Use an evidence receipt, a transfer check and a concise final account.
Chapter contents
Set up the distinction · Chapters 1–3
1. Separate contract rules from observed behaviour
Work the calculation · Chapters 4–5
Change the context · Chapters 6–8
6. Calculate refund liability on a separate ledger
Test limits and misconceptions · Chapters 9–10
Verify and explain · Chapters 11–12
CHAPTER 1 OF 12 · Set up the distinction
1. Separate contract rules from observed behaviour
The short answer is that refundability states what happens if a qualifying cancellation occurs, while cancellation hazard estimates the conditional chance that an active booking cancels during a stated interval. This chapter treats separate contract rules from observed behaviour as a defined comparison. Keep booking identity, product terms, purchase and service times, interval risk set, cancellation timestamp, exchange or amendment links, refund eligibility, refund amount, retained status, no-show status, segment mix, censoring and forecast origin in separate fields. Two values can share a subject or unit and still answer different questions.
Worked example. A flexible booking may allow a full refund yet remain active, while a non-refundable booking may still cancel even though its refund is zero Write the unit and basis beside every number, keep the intermediate result visible, and retain enough raw information for another reader to reproduce the answer without guessing a software default.
Why the result follows. rights and probabilities occupy different columns and neither determines the other by definition The decisive operation is named before the conclusion. That makes a disagreement traceable to a denominator, measurement state, model or evidence threshold instead of leaving it as a vague conflict between labels.
Changed-context transfer check. change a product from full refund to a fixed fee while holding its observed hazard history unchanged and identify which ledger changes Rebuild the result from the altered premise. State the first intermediate value that changes, propagate it through the remaining steps, and identify any part of the earlier conclusion that still holds.
A common misconception is that nearby measurements are interchangeable. Here, The shortcut is to label flexible as likely to cancel and non-refundable as certain to show. Terms influence incentives but do not supply the probability Correct the shortcut by naming the operation and the information it retains or discards, then show one case in which the values move differently.
Limits and caveats. Behaviour can respond to terms, price, channel and traveller selection; causality requires more than a segment contrast The lesson explains measurement and retained-demand estimates; it does not choose an overbooking policy, set consumer terms or replace jurisdiction-specific obligations. Report any convention, instrument setting or model choice that can change the boundary of the claim.
Practical evidence check. Archive the exact terms version and observed event history separately before modelling. Preserve the raw inputs, exclusions, units, uncertainty basis and calculation trail. A reader should be able to change one premise and obtain the transfer result without guessing the sample, conditions, measurement state or evidence basis used.
CHAPTER 2 OF 12 · Set up the distinction
2. Choose one forecast origin and time scale
The short answer is that time-varying hazard must be indexed from a declared origin such as purchase time or time remaining before service. This chapter treats choose one forecast origin and time scale as a defined comparison. Keep booking identity, product terms, purchase and service times, interval risk set, cancellation timestamp, exchange or amendment links, refund eligibility, refund amount, retained status, no-show status, segment mix, censoring and forecast origin in separate fields. Two values can share a subject or unit and still answer different questions.
Worked example. A cancellation two days after purchase can also be twenty days before departure; the same event occupies different bins on the two clocks Write the unit and basis beside every number, keep the intermediate result visible, and retain enough raw information for another reader to reproduce the answer without guessing a software default.
Why the result follows. hazard interpretation changes with the risk-set clock even when the event timestamp is unchanged The decisive operation is named before the conclusion. That makes a disagreement traceable to a denominator, measurement state, model or evidence threshold instead of leaving it as a vague conflict between labels.
Changed-context transfer check. rebuild a model indexed by days-to-service after first tabulating it by days-since-purchase Rebuild the result from the altered premise. State the first intermediate value that changes, propagate it through the remaining steps, and identify any part of the earlier conclusion that still holds.
A common misconception is that nearby measurements are interchangeable. Here, The shortcut is to mix both clocks in one rate without defining exposure. Bookings purchased at different lead times then receive incomparable intervals Correct the shortcut by naming the operation and the information it retains or discards, then show one case in which the values move differently.
Limits and caveats. Either clock can be useful, and richer models may use both as covariates The lesson explains measurement and retained-demand estimates; it does not choose an overbooking policy, set consumer terms or replace jurisdiction-specific obligations. Report any convention, instrument setting or model choice that can change the boundary of the claim.
Practical evidence check. Store purchase time, service time, forecast time, interval boundaries and the active-booking count at every boundary. Preserve the raw inputs, exclusions, units, uncertainty basis and calculation trail. A reader should be able to change one premise and obtain the transfer result without guessing the sample, conditions, measurement state or evidence basis used.
CHAPTER 3 OF 12 · Set up the distinction
3. Compute interval hazards on shrinking risk sets
The short answer is that each interval cancellation hazard divides cancellations in that interval by bookings active at its start under the stated censoring rule. This chapter treats compute interval hazards on shrinking risk sets as a defined comparison. Keep booking identity, product terms, purchase and service times, interval risk set, cancellation timestamp, exchange or amendment links, refund eligibility, refund amount, retained status, no-show status, segment mix, censoring and forecast origin in separate fields. Two values can share a subject or unit and still answer different questions.
Worked example. A flexible segment starts with 100 active bookings; 10 cancel in interval one, so interval-one hazard is 0.10 and only 90 enter interval two Write the unit and basis beside every number, keep the intermediate result visible, and retain enough raw information for another reader to reproduce the answer without guessing a software default.
Why the result follows. later hazards condition on survival through earlier intervals rather than the original cohort The decisive operation is named before the conclusion. That makes a disagreement traceable to a denominator, measurement state, model or evidence threshold instead of leaving it as a vague conflict between labels.
Changed-context transfer check. add five administrative withdrawals before interval two and state whether they are censored or cancellation events before recomputing the denominator Rebuild the result from the altered premise. State the first intermediate value that changes, propagate it through the remaining steps, and identify any part of the earlier conclusion that still holds.
A common misconception is that nearby measurements are interchangeable. Here, The shortcut is to divide every interval event count by the original 100. It understates conditional risk after the risk set shrinks Correct the shortcut by naming the operation and the information it retains or discards, then show one case in which the values move differently.
Limits and caveats. Discrete intervals approximate event time and require consistent same-day ordering The lesson explains measurement and retained-demand estimates; it does not choose an overbooking policy, set consumer terms or replace jurisdiction-specific obligations. Report any convention, instrument setting or model choice that can change the boundary of the claim.
Practical evidence check. Preserve risk-set entries, events, censoring, interval boundaries and denominator reconciliation. Preserve the raw inputs, exclusions, units, uncertainty basis and calculation trail. A reader should be able to change one premise and obtain the transfer result without guessing the sample, conditions, measurement state or evidence basis used.
The short answer is that retention through several intervals is the product of one minus each interval hazard for the relevant segment. This chapter treats convert hazards into retained demand as a defined comparison. Keep booking identity, product terms, purchase and service times, interval risk set, cancellation timestamp, exchange or amendment links, refund eligibility, refund amount, retained status, no-show status, segment mix, censoring and forecast origin in separate fields. Two values can share a subject or unit and still answer different questions.
Worked example. Flexible-booking hazards of 0.10, 0.08 and 0.06 give retention 0.90 times 0.92 times 0.94 equals 0.77832 Write the unit and basis beside every number, keep the intermediate result visible, and retain enough raw information for another reader to reproduce the answer without guessing a software default.
Why the result follows. multiplying conditional survival factors follows the bookings that remain exposed to the next interval The decisive operation is named before the conclusion. That makes a disagreement traceable to a denominator, measurement state, model or evidence threshold instead of leaving it as a vague conflict between labels.
Changed-context transfer check. change only the last hazard to 0.12 and recompute the complete product Rebuild the result from the altered premise. State the first intermediate value that changes, propagate it through the remaining steps, and identify any part of the earlier conclusion that still holds.
A common misconception is that nearby measurements are interchangeable. Here, The shortcut is to subtract the sum 0.24 from one. The hazards apply to successively smaller risk sets Correct the shortcut by naming the operation and the information it retains or discards, then show one case in which the values move differently.
Limits and caveats. Independence between people is not required for the arithmetic expectation, but correlation matters for capacity risk The lesson explains measurement and retained-demand estimates; it does not choose an overbooking policy, set consumer terms or replace jurisdiction-specific obligations. Report any convention, instrument setting or model choice that can change the boundary of the claim.
Practical evidence check. Save every hazard, survival factor, product, segment and forecast horizon. Preserve the raw inputs, exclusions, units, uncertainty basis and calculation trail. A reader should be able to change one premise and obtain the transfer result without guessing the sample, conditions, measurement state or evidence basis used.
CHAPTER 5 OF 12 · Work the calculation
5. Work a two-product retained-demand estimate
The short answer is that segment-specific retained counts should be calculated before aggregation when products have different time paths. This chapter treats work a two-product retained-demand estimate as a defined comparison. Keep booking identity, product terms, purchase and service times, interval risk set, cancellation timestamp, exchange or amendment links, refund eligibility, refund amount, retained status, no-show status, segment mix, censoring and forecast origin in separate fields. Two values can share a subject or unit and still answer different questions.
Worked example. For 100 flexible bookings with retention 0.77832 and 100 restricted bookings with hazards 0.03, 0.02 and 0.01, retained demand is 77.832 plus 94.1094 equals 171.9414 bookings Write the unit and basis beside every number, keep the intermediate result visible, and retain enough raw information for another reader to reproduce the answer without guessing a software default.
Why the result follows. each segment follows its own conditional process and aggregation occurs only after compatible cohort-wide expectations exist The decisive operation is named before the conclusion. That makes a disagreement traceable to a denominator, measurement state, model or evidence threshold instead of leaving it as a vague conflict between labels.
Changed-context transfer check. change the mix to 160 flexible and 40 restricted while keeping segment hazards and recompute the portfolio retention Rebuild the result from the altered premise. State the first intermediate value that changes, propagate it through the remaining steps, and identify any part of the earlier conclusion that still holds.
A common misconception is that nearby measurements are interchangeable. Here, The shortcut is to average the six hazards and apply one rate to 200 bookings. Timing and segment weights are lost Correct the shortcut by naming the operation and the information it retains or discards, then show one case in which the values move differently.
Limits and caveats. The numbers are original teaching inputs and do not describe a specific airline or hotel The lesson explains measurement and retained-demand estimates; it does not choose an overbooking policy, set consumer terms or replace jurisdiction-specific obligations. Report any convention, instrument setting or model choice that can change the boundary of the claim.
Practical evidence check. Retain segment starting counts, interval hazards, products, unrounded expected counts and the sum check. Preserve the raw inputs, exclusions, units, uncertainty basis and calculation trail. A reader should be able to change one premise and obtain the transfer result without guessing the sample, conditions, measurement state or evidence basis used.
CHAPTER 6 OF 12 · Change the context
6. Calculate refund liability on a separate ledger
The short answer is that expected refund liability combines cancellation probability with the product-specific eligible refund amount and event timing rather than reducing physical retained demand twice. This chapter treats calculate refund liability on a separate ledger as a defined comparison. Keep booking identity, product terms, purchase and service times, interval risk set, cancellation timestamp, exchange or amendment links, refund eligibility, refund amount, retained status, no-show status, segment mix, censoring and forecast origin in separate fields. Two values can share a subject or unit and still answer different questions.
Worked example. If each flexible cancellation returns 80 dollars and cumulative cancellation probability is 0.22168, 100 such bookings create an expected refund of 1,773.44 dollars under the simplified rule Write the unit and basis beside every number, keep the intermediate result visible, and retain enough raw information for another reader to reproduce the answer without guessing a software default.
Why the result follows. the cash expectation is event probability times eligible amount, while the retained-booking expectation counts active demand The decisive operation is named before the conclusion. That makes a disagreement traceable to a denominator, measurement state, model or evidence threshold instead of leaving it as a vague conflict between labels.
Changed-context transfer check. introduce a 20-dollar late-cancellation fee in the last interval and calculate that interval’s refund using its own event probability Rebuild the result from the altered premise. State the first intermediate value that changes, propagate it through the remaining steps, and identify any part of the earlier conclusion that still holds.
A common misconception is that nearby measurements are interchangeable. Here, The shortcut is to subtract refund dollars from expected arrivals. Money and service units use different dimensions Correct the shortcut by naming the operation and the information it retains or discards, then show one case in which the values move differently.
Limits and caveats. Taxes, exchanges, credits, partial use and timing can require a fuller accounting model The lesson explains measurement and retained-demand estimates; it does not choose an overbooking policy, set consumer terms or replace jurisdiction-specific obligations. Report any convention, instrument setting or model choice that can change the boundary of the claim.
Practical evidence check. Preserve fare paid, eligible amount by interval, fee, currency, event probability, settlement type and accounting date. Preserve the raw inputs, exclusions, units, uncertainty basis and calculation trail. A reader should be able to change one premise and obtain the transfer result without guessing the sample, conditions, measurement state or evidence basis used.
CHAPTER 7 OF 12 · Change the context
7. Do not treat non-refundable as non-cancellable
The short answer is that a booking can leave the active demand set even when its refundable amount is zero. This chapter treats do not treat non-refundable as non-cancellable as a defined comparison. Keep booking identity, product terms, purchase and service times, interval risk set, cancellation timestamp, exchange or amendment links, refund eligibility, refund amount, retained status, no-show status, segment mix, censoring and forecast origin in separate fields. Two values can share a subject or unit and still answer different questions.
Worked example. A restricted ticket cancelled by the customer releases capacity despite producing no cash refund under the stipulated terms Write the unit and basis beside every number, keep the intermediate result visible, and retain enough raw information for another reader to reproduce the answer without guessing a software default.
Why the result follows. operational inventory and refund liability respond differently to the same event The decisive operation is named before the conclusion. That makes a disagreement traceable to a denominator, measurement state, model or evidence threshold instead of leaving it as a vague conflict between labels.
Changed-context transfer check. change the workflow so an unreported abandonment remains active until no-show and explain how retained-demand timing changes Rebuild the result from the altered premise. State the first intermediate value that changes, propagate it through the remaining steps, and identify any part of the earlier conclusion that still holds.
A common misconception is that nearby measurements are interchangeable. Here, The shortcut is to keep every non-refundable booking in forecast demand until service. It can overstate arrivals and hide released capacity Correct the shortcut by naming the operation and the information it retains or discards, then show one case in which the values move differently.
Limits and caveats. Whether an unused booking is explicitly cancelled, exchanged or left to no-show depends on data capture and customer behaviour The lesson explains measurement and retained-demand estimates; it does not choose an overbooking policy, set consumer terms or replace jurisdiction-specific obligations. Report any convention, instrument setting or model choice that can change the boundary of the claim.
Practical evidence check. Link status transition, inventory release time, refund amount and terminal outcome without inferring one from another. Preserve the raw inputs, exclusions, units, uncertainty basis and calculation trail. A reader should be able to change one premise and obtain the transfer result without guessing the sample, conditions, measurement state or evidence basis used.
The short answer is that a pooled cancellation rate changes when the flexible-restricted mix changes even if each segment hazard stays constant. This chapter treats control aggregation and mix shift as a defined comparison. Keep booking identity, product terms, purchase and service times, interval risk set, cancellation timestamp, exchange or amendment links, refund eligibility, refund amount, retained status, no-show status, segment mix, censoring and forecast origin in separate fields. Two values can share a subject or unit and still answer different questions.
Worked example. A historical half-flexible portfolio and a current eighty-percent-flexible portfolio cannot share one pooled rate when flexible hazards are higher Write the unit and basis beside every number, keep the intermediate result visible, and retain enough raw information for another reader to reproduce the answer without guessing a software default.
Why the result follows. composition affects the weighted expectation before any behavioural drift occurs The decisive operation is named before the conclusion. That makes a disagreement traceable to a denominator, measurement state, model or evidence threshold instead of leaving it as a vague conflict between labels.
Changed-context transfer check. hold segment hazards fixed, change only the mix and explain the resulting forecast movement Rebuild the result from the altered premise. State the first intermediate value that changes, propagate it through the remaining steps, and identify any part of the earlier conclusion that still holds.
A common misconception is that nearby measurements are interchangeable. Here, The shortcut is to interpret a changing portfolio rate as customer behaviour change. It may be a product-mix change Correct the shortcut by naming the operation and the information it retains or discards, then show one case in which the values move differently.
Limits and caveats. Too many narrow segments can be sparse and unstable, so pooling choices need validation The lesson explains measurement and retained-demand estimates; it does not choose an overbooking policy, set consumer terms or replace jurisdiction-specific obligations. Report any convention, instrument setting or model choice that can change the boundary of the claim.
Practical evidence check. Record product mix at training and forecast dates, segment counts, pooling rules and calibration by segment. Preserve the raw inputs, exclusions, units, uncertainty basis and calculation trail. A reader should be able to change one premise and obtain the transfer result without guessing the sample, conditions, measurement state or evidence basis used.
CHAPTER 9 OF 12 · Test limits and misconceptions
9. Handle exchanges, amendments and competing outcomes
The short answer is that refund, exchange, date change, provider cancellation and customer cancellation are distinct transitions that require linked booking history. This chapter treats handle exchanges, amendments and competing outcomes as a defined comparison. Keep booking identity, product terms, purchase and service times, interval risk set, cancellation timestamp, exchange or amendment links, refund eligibility, refund amount, retained status, no-show status, segment mix, censoring and forecast origin in separate fields. Two values can share a subject or unit and still answer different questions.
Worked example. A Friday booking exchanged into Saturday should not appear as one cancellation plus an unrelated new customer Write the unit and basis beside every number, keep the intermediate result visible, and retain enough raw information for another reader to reproduce the answer without guessing a software default.
Why the result follows. linkage preserves the underlying demand intention and prevents pickup and cancellation reports from both inflating The decisive operation is named before the conclusion. That makes a disagreement traceable to a denominator, measurement state, model or evidence threshold instead of leaving it as a vague conflict between labels.
Changed-context transfer check. break the exchange link and quantify how Friday retention falls while Saturday acquisition rises artificially Rebuild the result from the altered premise. State the first intermediate value that changes, propagate it through the remaining steps, and identify any part of the earlier conclusion that still holds.
A common misconception is that nearby measurements are interchangeable. Here, The shortcut is to treat each status row as an independent person. Replacement records then distort hazards Correct the shortcut by naming the operation and the information it retains or discards, then show one case in which the values move differently.
Limits and caveats. Identity linkage may be uncertain and must respect privacy and retention controls The lesson explains measurement and retained-demand estimates; it does not choose an overbooking policy, set consumer terms or replace jurisdiction-specific obligations. Report any convention, instrument setting or model choice that can change the boundary of the claim.
Practical evidence check. Store original and replacement IDs, transition reason, initiator, timestamps, financial treatment and unresolved matches. Preserve the raw inputs, exclusions, units, uncertainty basis and calculation trail. A reader should be able to change one premise and obtain the transfer result without guessing the sample, conditions, measurement state or evidence basis used.
CHAPTER 10 OF 12 · Test limits and misconceptions
10. Distinguish policy effect from customer selection
The short answer is that a lower observed hazard among restricted products does not by itself prove that removing refunds caused the difference. This chapter treats distinguish policy effect from customer selection as a defined comparison. Keep booking identity, product terms, purchase and service times, interval risk set, cancellation timestamp, exchange or amendment links, refund eligibility, refund amount, retained status, no-show status, segment mix, censoring and forecast origin in separate fields. Two values can share a subject or unit and still answer different questions.
Worked example. Travellers who value certainty may choose one term while uncertain travellers select flexibility, producing different hazards before the rule acts Write the unit and basis beside every number, keep the intermediate result visible, and retain enough raw information for another reader to reproduce the answer without guessing a software default.
Why the result follows. observed product groups combine incentive effects with self-selection and channel differences The decisive operation is named before the conclusion. That makes a disagreement traceable to a denominator, measurement state, model or evidence threshold instead of leaving it as a vague conflict between labels.
Changed-context transfer check. compare a term change within a stable product and also check whether purchaser composition changed Rebuild the result from the altered premise. State the first intermediate value that changes, propagate it through the remaining steps, and identify any part of the earlier conclusion that still holds.
A common misconception is that nearby measurements are interchangeable. Here, The shortcut is to read a segment coefficient as a causal price-rule effect. Confounding can remain Correct the shortcut by naming the operation and the information it retains or discards, then show one case in which the values move differently.
Limits and caveats. Causal analysis may require experimental or quasi-experimental design and careful fairness review The lesson explains measurement and retained-demand estimates; it does not choose an overbooking policy, set consumer terms or replace jurisdiction-specific obligations. Report any convention, instrument setting or model choice that can change the boundary of the claim.
Practical evidence check. Preserve eligibility, offer set, prices, channel, customer covariates, policy date and identification assumptions. Preserve the raw inputs, exclusions, units, uncertainty basis and calculation trail. A reader should be able to change one premise and obtain the transfer result without guessing the sample, conditions, measurement state or evidence basis used.
The short answer is that hazard and retention estimates should be calibrated for the exact forecast origin, lead-time window and active risk set in which they are used. This chapter treats validate at the decision horizon as a defined comparison. Keep booking identity, product terms, purchase and service times, interval risk set, cancellation timestamp, exchange or amendment links, refund eligibility, refund amount, retained status, no-show status, segment mix, censoring and forecast origin in separate fields. Two values can share a subject or unit and still answer different questions.
Worked example. A model may calibrate lifetime cancellations yet overpredict final-week attrition among bookings still active seven days before service Write the unit and basis beside every number, keep the intermediate result visible, and retain enough raw information for another reader to reproduce the answer without guessing a software default.
Why the result follows. conditioning changes the eligible cohort and available information The decisive operation is named before the conclusion. That makes a disagreement traceable to a denominator, measurement state, model or evidence threshold instead of leaving it as a vague conflict between labels.
Changed-context transfer check. move the forecast origin from purchase to seven days before service and rebuild the test set from bookings active then Rebuild the result from the altered premise. State the first intermediate value that changes, propagate it through the remaining steps, and identify any part of the earlier conclusion that still holds.
A common misconception is that nearby measurements are interchangeable. Here, The shortcut is to report one accuracy score across all horizons. Class balance and exposure change over time Correct the shortcut by naming the operation and the information it retains or discards, then show one case in which the values move differently.
Limits and caveats. Historical policy interventions can alter labels and create feedback The lesson explains measurement and retained-demand estimates; it does not choose an overbooking policy, set consumer terms or replace jurisdiction-specific obligations. Report any convention, instrument setting or model choice that can change the boundary of the claim.
Practical evidence check. Retain training cutoff, forecast origin, calibration curve, scoring rule, segment results, drift checks and confidence intervals. Preserve the raw inputs, exclusions, units, uncertainty basis and calculation trail. A reader should be able to change one premise and obtain the transfer result without guessing the sample, conditions, measurement state or evidence basis used.
CHAPTER 12 OF 12 · Verify and explain
12. Build a retained-demand and refund receipt
The short answer is that the final report should connect terms, hazards, risk sets, retained counts and refund liability without collapsing their units. This chapter treats build a retained-demand and refund receipt as a defined comparison. Keep booking identity, product terms, purchase and service times, interval risk set, cancellation timestamp, exchange or amendment links, refund eligibility, refund amount, retained status, no-show status, segment mix, censoring and forecast origin in separate fields. Two values can share a subject or unit and still answer different questions.
Worked example. A receipt states flexible retention 0.77832 from three hazards, restricted retention 0.941094, portfolio retained expectation 171.9414 and expected flexible refund 1,773.44 dollars under the declared rule Write the unit and basis beside every number, keep the intermediate result visible, and retain enough raw information for another reader to reproduce the answer without guessing a software default.
Why the result follows. the chain makes each number reproducible and shows which quantity is operational and which is financial The decisive operation is named before the conclusion. That makes a disagreement traceable to a denominator, measurement state, model or evidence threshold instead of leaving it as a vague conflict between labels.
Changed-context transfer check. change the product mix and late-cancellation fee, then issue a new receipt rather than editing only the final totals Rebuild the result from the altered premise. State the first intermediate value that changes, propagate it through the remaining steps, and identify any part of the earlier conclusion that still holds.
A common misconception is that nearby measurements are interchangeable. Here, The shortcut is to publish one cancellation percentage and one net revenue number. Readers cannot recover timing, terms or denominators Correct the shortcut by naming the operation and the information it retains or discards, then show one case in which the values move differently.
Limits and caveats. Expected values do not guarantee a single service date and do not set policy The lesson explains measurement and retained-demand estimates; it does not choose an overbooking policy, set consumer terms or replace jurisdiction-specific obligations. Report any convention, instrument setting or model choice that can change the boundary of the claim.
Practical evidence check. Include cohort, terms version, clock, intervals, risk sets, hazards, survival, retained units, refund formula, uncertainty, limits and approval wording. Preserve the raw inputs, exclusions, units, uncertainty basis and calculation trail. A reader should be able to change one premise and obtain the transfer result without guessing the sample, conditions, measurement state or evidence basis used.
Contents · Previous chapter · Return to the Yield Management subject guide
Frequently asked questions
Does refundable mean likely to cancel?
No. Refundability is a rule. Cancellation likelihood must be estimated from an appropriate active-booking risk set, and product choice can reflect customer selection.
Can a non-refundable booking cancel?
Yes. It can leave active demand even if the eligible refund is zero. Operational inventory and financial settlement must be recorded separately.
Why multiply survival factors?
Each interval hazard applies only to bookings that survived earlier intervals. The product follows the shrinking risk set.
Can retained demand set an overbooking limit?
No. A decision also needs the arrival distribution, capacity, correlated disruption, consequences, customer protections and governance.
Continue through the verified reading routes
Start with the wider system
The Gold Standard of Yield Management · The Gold Standard of Education | Learning That Transfers into Life · How X Works hub
Compare the closest mechanisms
Pre-arrival cancellation hazard and day-of-service no-show probability · Booking pace and booking pickup between snapshots · Expected empty capacity and probability of a service shortfall
Choose the next lesson by reader job
Use the cancellation-versus-no-show lesson to keep terminal events separate, the pace-versus-pickup lesson to preserve snapshot logic, and the shortfall lesson before turning an expected retained count into a capacity decision.
Authoritative sources and evidence checks
Transportation Research Part B — hazard model of passenger refund and exchange behaviour · Transportation Research Board TRID — hazard model record · Yugoslav Journal of Operations Research — overbooking under cancellation, refund and no-show
The cited research supports time-varying conditional cancellation modelling and the separation of cancellation, refund and no-show processes. All numerical cohorts and terms in this guide are original teaching examples.