Skip to content

AI uses for records

Hospitality service recovery dataset: count what each offer displaces

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

A hospitality service recovery dataset should record what each recovery offer spent and who else it touched, not only whether the first guest was satisfied. Label the decision (offer, condition, substitute, defer, escalate or decline), the primary outcome and the displacement outcome separately. Records often lose the trail: in open-source QloApps, none of three room-change paths keeps the previous room in its own field.

Agent run

Hospitality service recovery

This page covers hospitality service recovery dataset: count what each offer displaces.Illustrative

Key takeaways

  • Every recovery offer spends a room, a rate, a legal hold, a turnover window, a package's content or an approver's limit that someone else may need.
  • Label the decision, the first guest's outcome and the displacement outcome as three fields; one success flag lets a happy guest hide a broken commitment.
  • In QloApps' code, a room swap changes two bookings but writes its note only on the initiating order, or no note at all when made from the Book Now screen, and a cross-type move deletes the original booking row.
  • Hard negatives matter most: moving a guest into an accessible room while other rooms of that type are unsold, or granting a late checkout over a promised early check-in.
  • Close the observation window only after every affected booking has checked out, and label overwritten trails unknown rather than clean.

What decision does a hospitality service recovery dataset need to capture?

It needs the choice a front desk agent, guest services lead or duty manager makes after something goes wrong: which remedy to offer, on what terms and with whose approval, given the rooms, rules and promises in force at that moment. The right answer is often conditional, partial or deferred, because every remedy uses something that another guest, a rate rule or the next shift may need.

A recovery that fixes the reported problem can still be the wrong decision. A move into the last free room of a type can strand tonight's arrival assigned to it; a late checkout can push back a promised early check-in; a swap can hand one guest's honored request to another. The published guide to service recovery labels scores whether the first guest's stay was put right. This page covers the steps around it: the choice among remedies, and the second problem a remedy can create.

Useful evidence has three parts: a snapshot of inventory, arrivals, rate rules and open promises at the time of the offer; the offer as made, with its conditions and approver; and the records over the following hours and days that show whether anyone else absorbed a cost. Without the third part, a dataset can only teach generosity.

Every recovery offer spends something someone else may need

Each remedy consumes a resource with other claims on it: a room for tonight, a room type's revenue, an accessible room held back by law, the turnover time before the next arrival, a package's agreed content or an approver's limit. The tradeoff runs across availability, policy and guest commitments, and the dataset needs fields for all three.

Two public rules show how policy can turn a kind offer into a breach. In the United States, the ADA regulation on reservations at places of lodging, 28 CFR 36.302(e)(1)(iii), requires accessible guest rooms to be held for guests with disabilities until every other room of that type has been rented and the accessible room is the only one left. In the EU, Article 13 of the Package Travel Directive, (EU) 2015/2302, requires suitable alternative arrangements when a significant part of a package cannot be delivered, with an appropriate price reduction when the alternative is of lower quality.

Commitments are the third claim and the hardest to see. A promise made to another guest, such as a high floor, connecting rooms or an early check-in, often lives in a reservation comment or a message thread rather than in the room assignment, so an agent looking only at availability will miss it. A dataset meant to teach this tradeoff has to capture those promises in the snapshot, with the time each was made.

The map lists common offers, what each spends and the record that would reveal its side effect:

Every recovery offer spends something someone else may need
OfferWhat it spendsWho else can be affectedRecord that reveals the side effect
Move to a free room of the same typeThat room for the remaining nightsAn arrival already assigned to the roomRoom assignment history and the arrivals list
Upgrade to a higher room typeHigher-category inventory and its revenueLater guests who booked or were promised that typeRoom type charged against room type occupied
Swap rooms with another bookingAnother guest's assigned roomThat guest, and any request honored in their assignmentChange history on both bookings
Move into an accessible roomInventory held for guests with disabilitiesA later guest who needs that roomAccessible-room flag and unsold rooms of the type at the time of the move
Late checkoutThe room's turnover windowThe next arrival and the housekeeping scheduleDeparture time change and the next arrival's check-in time
Credit, voucher or complimentary itemRevenue and an approver's limitRate rules and the approver's budgetFolio adjustment, voucher record and approval log
Substitute inside a packageThe package's agreed contentThe traveler's claim to a price reduction (EU)Package components and any reduction posted

Label set: the decision, the fix and the side effect

Label three things separately: what was decided, whether the first guest's problem was fixed, and whether the remedy pushed a cost onto someone or something else. Folding them into one success flag lets a delighted guest hide a broken commitment, and a broken commitment hide a sound recovery.

The decision label has six values, matching the outcomes found in front-office logs: offer as asked; offer with a condition, such as a time limit, an availability check at a set hour or a price difference; substitute a different remedy; defer with a stated time to confirm; escalate to an approver; or decline with the reason. The schema:

Label set: the decision, the fix and the side effect
FieldValuesKnown whenSource record
DecisionOffer as asked, offer with condition, substitute, defer with a time, escalate, decline with reasonAt the offerMessage thread, note or approval log
ConditionTime-bound, availability-bound, approval-bound or payment-bound, with its bound stored as a valueAt the offerThe offer text, normalized by the reviewer
Primary outcomeFixed, partly fixed, not fixed or unknownBefore the guest's next night, or at checkoutWork order, room move, folio and follow-up contacts
Displacement outcomeNone found, absorbed by the property, transferred to another guest, rule breached or unknownAfter the window closes for every affected bookingLinked bookings, later moves, complaints and adjustments
Displacement lagHours or days from the offer to the first record of the side effectWhen the side effect is recordedTimestamps on the linked records
AuthorityApprover role and the policy version in forceAt the offerRole settings and policy history

What one open-source hotel system keeps after a room move

Less than a labeler needs. We read the source code of QloApps, an open-source hotel booking and reservation system, at its commit of October 27, 2026, to see what the database keeps when staff move a guest. Of its three room-change paths, none stores the previous room in a field of its own. Moved from the order page, the previous room number survives only in an automatically written private note on that order, followed by an optional remark of up to 250 characters; moved from the Book Now screen, which posts the same form, nothing is noted and the remark is discarded.

Two paths lose more than that. Reallocating a booking to a different room type deletes the original booking row, creates a new one and deletes any refund request detail recorded against the old row, along with the request itself when no other booking remains in it. Swapping exchanges rooms between two bookings of the same type and dates, so it always changes a second booking, usually another guest's, yet the note is written only on the order the staff member was working from, or on neither order when the swap is made from the Book Now screen; when the other booking sits on a different order, that order records nothing.

The hotel module does keep a structured history, but for status changes only: its booking status history table records the old and new status, the employee and a remark, and none of the three room-change paths writes to it. Both room-change functions fire a hook when they finish, and no file in the repository listens for either one.

Commercial systems vary. Oracle's OPERA Cloud documentation describes room move reasons kept as a configured code list, which a parameter can make mandatory, and a Room Type to Charge setting that shows when a move was a complimentary upgrade. The same recovery can leave a coded reason in one system and a free-text note in another. What QloApps records on each path:

What one open-source hotel system keeps after a room move
Room-change pathWhat changes in the databasePrevious room in its own fieldNote writtenSecond guest's record
Reallocate to a free room of the same typeRoom ID and number overwritten on the booking rowNoOrder page: private note on the order, optional remark; Book Now screen: noneNot affected
Reallocate to a different room typeBooking row deleted and recreated; any refund request on it deletedNoOrder page: private note on the order, optional remark; Book Now screen: noneNot affected
Swap with another booking of the same type and datesRoom ID and number overwritten on both booking rowsNoOrder page: private note on the initiating order only; Book Now screen: noneChanged; when it sits on another order, that order gets no note
Booking status change, for comparisonRow added to the status history tableYes: old and new status, employee and remarkNot neededNot affected
Swimlane diagram of a room swap made from the QloApps order page: guest A's booking moves from room 412 to 508 and guest B's from 508 to 412, an order note is written for guest A only and guest B's order gets none, a line adds that the same swap made from the Book Now screen notes neither booking, and a panel states that none of three room-change paths keeps the previous room in its own field.
Room numbers are illustrative; the behavior shown is read from QloApps source code at commit 4352b49 (October 27, 2026).

Casebook: six recovery offers and their hard negatives

Each case pairs the decision a reviewer would accept with the tempting answer that fixes the first guest and creates a second problem. The cases are illustrative composites written for this guide and describe no real property or guest.

Sample hard negatives in proportion to how often each conflict occurs at the supplying property, and keep near misses: nights when the obvious remedy was right because no one else had a claim on the resource. A set made only of conflicts teaches refusal, and a set made only of clean upgrades teaches that every room is free.

Case A is the control, a generous offer that is right, so the set does not reward caution for its own sake. The others each turn on one fact in the snapshot:

Casebook: six recovery offers and their hard negatives
Case (illustrative)Decisive fact at the offerAccepted decisionHard negativeWhat settles the label
A. Noisy room, suite empty all weekNo suite arrivals or holds for the stayOffer the upgrade as askedRefuse or delay out of habitNo suite conflict through checkout
B. Broken shower, night 1 of 3The only free room of the type is assigned to an arrival from tomorrowOffer the move with a condition: tonight only, or a different free type for the stayMove the guest for all three nightsTomorrow's arrival keeps the assigned room
C. Heating fault at a U.S. propertyThe free king is accessible; standard kings remain unsold but are not yet cleanedDefer until a standard king is ready, or move to another typeUse the accessible room nowInventory by type at the time of the move
D. Guest asks to swap for a quieter roomThe candidate room belongs to a guest whose high-floor request was honoredDecline the swap; offer another free room or a set time to confirmSwap the two bookingsNo complaint or move on the second booking
E. Late checkout to make up for a bad nightThe room is assigned to an arrival with a confirmed early check-inSubstitute luggage hold and day-use space, or a shorter extensionGrant the full late checkoutThe arrival's check-in time against the promise
F. Spa slot cancelled in an EU packageThe substitute treatment is of lower valueOffer the substitute with a price reductionOffer the substitute alonePackage lines and the reduction posted

How the side effect becomes known, and how long to wait

A displacement shows up in someone else's record, often after the first guest has left, so the observation window has to cover every booking the offer touched. A swap's cost appears when the second guest arrives; a late checkout's cost appears at the next arrival's check-in; an accessible-room move's cost may appear only if a guest who needs that room arrives later the same night.

Set the window by offer type before labeling: through the checkout of every affected booking for moves and swaps, through the next arrival's check-in for late checkouts, and through settlement for credits. Link affected bookings by room and date rather than by guest name, and store the key that made each link. Where the system overwrote the trail, as QloApps does for the swapped guest, the honest label is unknown, not none found.

Satisfaction scores cannot stand in for this field. A 2007 meta-analysis of the service recovery paradox by de Matos, Henrique and Rossi found a significant positive effect of recovery on satisfaction but no significant effect on repurchase intentions, word of mouth or corporate image. And a first guest's rating cannot see the second guest's evening at all.

Domain review: who adjudicates and what they sign

A front office or revenue manager adjudicates each case, with accessibility and package cases also checked by someone who knows the rules that apply, because condition types, displacement and authority are judgment calls. The reviewer signs this checklist, and a failed item narrows or excludes the case:

At a glance

Domain review: who adjudicates and what they sign

  • Snapshot cut: inventory, arrivals, holds and open promises are taken as of the offer time, with nothing recorded later.
  • Decision class: one of the six values, assigned from the snapshot alone; a second reviewer repeats a sample, and agreement is reported.
  • Condition: typed and bounded, such as until 1 p.m. or if the suite is still free at 3 p.m.
  • Linked bookings: every booking the remedy touched is found by room and date and joined with a stored key.
  • Window: closed for every linked booking, or the case is marked open.
  • Rule check: accessibility holds, package terms, rate rules and the approver's limit as they stood that day.
  • Unknown over guessing: overwritten trails are labeled unknown rather than none found.
  • Privacy: names, contact details, room numbers tied to dates, and loyalty and card numbers are removed, and health or accessibility details are reviewed or excluded.

What a hotel's records can and cannot show

A hotel's records can show what was offered, what moved and what it touched, where the system kept the trail. They cannot show the remedies staff weighed and rejected, the rooms visible to them at that moment, or a condition agreed on the phone unless someone wrote it down. A hospitality label on a dataset establishes neither the right to license it nor its fitness for decisions with legal stakes, and four limits follow:

  • Systems differ: one property keeps coded move reasons and a room-type-to-charge field, another keeps free text and overwrites the room, so pooled data needs field-by-field mapping.
  • Rules differ by country and brand: the ADA holding rule and the EU package rules apply in their own jurisdictions, and brand standards and franchise terms add more.
  • Rights sit with several parties: owners, management companies, franchisors and booking platforms can each have a say in who may license reservation records.
  • Bounded use: the set supports drafting offers, flagging conflicts for staff and evaluating agents; it does not support automated decisions on accessibility requests.

Writing the recovery data request

Ask for linked cases, not tickets: the offer, its approver, the snapshot and every booking the remedy touched, with the window closed. State the acceptance test, for example: on a sample of swaps and moves, both affected bookings are present and joined, and no snapshot contains a record dated after the offer.

SourceX gathers datasets like this on request from the businesses that hold them rather than from stock. Its request form wants the scope, volume, language and format of the records along with timing and licensing needs, and the intended use and the test belong beside them. The hotel group or operator that holds the records approves the scope and terms and signs before anything moves, and SourceX removes identifying details itself or through its processing partner. Nothing here implies that such records are available now or that a given set will pass the test.

How the QloApps audit was done

Source: the Qloapps/QloApps repository at commit 4352b49, committed on October 27, 2026, sparse-cloned on October 28, 2026. A script read the files as text and ran none of them. Files: the hotel module's database installer (HotelReservationSystemDb.php), its booking class (HotelBookingDetail.php), its Book Now screen's controller (AdminHotelRoomsBookingController.php), the admin orders controller (AdminOrdersController.php) and the room reallocation modal template.

Checks as applied: the column lists of the booking detail and booking status history tables, with any column name containing prev, orig, old, reason, move or realloc in any case flagged (none was); inside the reallocateBooking and swapBooking functions, found by matching braces, every assignment to the room ID or room number, every delete call, every refund request deletion, any use of the status history class, any new message and every hook name; inside the order page controller's branches for reallocation and swap, each new message with the order it is attached to and its private flag; inside the same two branches of the Book Now controller, which renders the same form and posts it to itself, each call to the two functions (lines 244 and 286), any new message (none) and any read of the remark field (none); every remark text area in the template with its required attribute and maximum length; and, across every PHP file in the checkout, any method named after the two room-change hooks, matched in any case (none was found). Not covered: add-on modules sold separately, the database contents of any live hotel, front-office screens and any SourceX data. The script, whose second version added the Book Now controller after a recount, and its outputs are kept in the writer's scratch folder W09-main-004/bin.

Frequently asked questions

What is a service recovery offer in a hotel dataset?

It is the remedy staff propose after a failure, such as a room move, an upgrade, a late checkout, a credit or a substitute service, recorded with its conditions, its approver and the rooms, rules and promises in force at that moment. A dataset needs the offer and that context, not only the apology in the message thread.

Why isn't a satisfied guest enough to label the offer correct?

Because a remedy's costs often land on someone else: the next arrival, a swapped guest or a legal hold on accessible inventory. Research on the service recovery paradox also finds satisfaction gains that do not carry through to repurchase intentions or word of mouth, so ratings belong in a separate field, not in the label.

Can a hotel move a guest into an accessible room as a recovery?

Under the U.S. ADA lodging rule, accessible rooms must be held for guests with disabilities until all other rooms of that type are rented and the accessible room is the last one left. Whether a specific move complied depends on the inventory at that moment, so a dataset should store the room-type snapshot. This is general information, not legal advice.

What records show that a room swap affected another guest?

Change histories on both bookings, joined by room and date, plus any later move, complaint or adjustment on the second booking. Some systems note a swap only on the initiating order, as QloApps does from its order page (its Book Now screen notes neither), so the second guest's side may have to be rebuilt or labeled unknown.

How is this different from labeling whether a recovery succeeded?

A success rubric scores whether the first guest's stay was put right. This label set adds the decision itself and its side effects, so a remedy that fixed one stay and broke another does not count as a clean success.

Sources

  • QloApps, an open-source hotel booking and reservation system (OSL 3.0), at commit 4352b49 of October 27, 2026; no file in the repository defines a listener for the room reallocation or swap hooks. Source
  • The admin orders controller saves one private order message per reallocation or swap, attached to the order being viewed, with an optional staff remark. Source
  • The hotel module's Book Now controller accepts the same reallocation and swap form, calls the same two room-change functions (lines 244 and 286) and saves no order message, so the staff remark is not stored. Source
  • reallocateBooking overwrites the room on same-type moves and, for a different room type, deletes the original booking row and its refund request details; swapBooking overwrites the room on both bookings; neither function writes status history or a message. Source
  • The hotel module's booking detail table has no column for a previous room or a move reason; its booking status history table records old and new status, employee and remark. Source
  • The reallocation and swap forms each offer an optional remark text area limited to 250 characters. Source
  • 28 CFR 36.302(e)(1)(iii): accessible guest rooms are held for individuals with disabilities until all other rooms of that type have been rented and the accessible room is the only remaining room of that type. Source
  • Directive (EU) 2015/2302, Article 13: when a significant proportion of package services cannot be provided, the organiser offers suitable alternatives; an alternative of lower quality brings an appropriate price reduction. Source
  • De Matos, Henrique and Rossi (Journal of Service Research, 2007), a meta-analysis of the service recovery paradox: significant positive effect on satisfaction, nonsignificant for repurchase intentions, word of mouth and corporate image. Source
  • OPERA Cloud: moving an in-house reservation records a room move reason chosen from a list, and leaving Room Type to Charge at the original type makes the move a complimentary upgrade. Source
  • OPERA Cloud: room move reasons are configured as codes with descriptions, and a parameter can make a reason mandatory. Source

Tell us what data you need

Share scope, volume, format, timing and licensing needs. Datasets are sourced on request.

See if your company qualifies