Skip to content

Systems and records

Travel dataset sample acceptance: how buyers judge a workflow sample

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

A buyer should accept a travel workflow sample only when it covers the changes, cancellations and disruptions the model will handle, traces every case to a named system and export, attaches the fare, property or regulatory rule in force on each event date with time zones, and keeps traveler, payment and contract-restricted data out. Samples showing only final booking states, or showcase cases chosen by staff, should be queried or rejected.

Checklist

Travel dataset sample

This page covers travel dataset sample acceptance: how buyers judge a workflow sample.Illustrative

Key takeaways

  • Judge a travel sample on four reviews: task coverage, provenance, dated policy evidence and permissions.
  • Final booking states hide the task; keep ticketing deadlines, automatic cancellation rules and hotel penalty deadlines with their time zones.
  • Attach the rule as it stood on the event date, plus the measured facts a regulation depends on.
  • Draw from servicing events, stratified by task type, and report population counts next to sample counts.
  • Card data, passport numbers and contract-restricted content stop a sample until they are removed.
  • A sample that passes shows the records can support the tasks; model gains need a separate pilot.

How should a buyer evaluate a travel workflow dataset sample?

A buyer should accept a travel workflow sample only when it passes four reviews: task coverage (the sample holds the changes, cancellations and disruptions the model will face, in useful proportions), provenance (every case traces to a named reservation, property or servicing system and export), policy evidence (each case carries the fare, property or regulatory rule in force at the time, with dates), and permissions (traveler, payment and distribution data are handled or excluded). A sample that shows only final booking states fails, however clean it looks.

This page is written for travel and hospitality companies that expect to offer servicing records and want to know how the sample will be judged before they build it. The same review works for an AI team on the buying side. It is general information; privacy and contract questions about specific records go to counsel.

Why does a final booking record hide most of the task?

Travel servicing work is about dependencies and deadlines, and the final reservation shows neither. A change to one flight segment can break a connection, a linked hotel night and a car pickup. Whether a change was allowed depended on the fare purchased, the time remaining before a deadline, and sometimes a regulation that applied only because of how much the schedule moved.

Distribution systems record these conditions as separate, dated fields, which is what a sample has to preserve. Amadeus's published Flight Offers Search specification includes a last ticketing date, described as the date until which an offer booked on the day of the search remains valid for ticketing. Its Flight Create Orders specification includes a ticketing agreement with an option such as delay to cancel, a delay before an automatic process applies if no ticket is issued, and the affected segment identifiers. Its Hotel Search specification describes a cancellation policy deadline after which a penalty applies, expressed in the hotel's local time with the time zone offset.

Those three fields are a good test of any sample. If the cases do not show the ticketing deadline, the automatic cancellation rule and the hotel penalty deadline in a form that can be compared with event timestamps, nobody can check whether the recorded action was right.

Which policy evidence has to sit inside each case?

Every case should carry the rule that governed it, identified and dated, because travel rules change and the current rule is often not the one that applied. Carrier and property policies are the bulk of this. Regulation adds a layer that a buyer can check against primary text.

For U.S. air travel, the Department of Transportation's April 2024 refunds rule defines a significant change to include departures or arrivals moved by three hours or more for domestic itineraries and six hours or more for international ones, along with airport changes, added connections and downgrades, and it requires automatic refunds when the passenger does not accept the change or alternatives. DOT also requires carriers to hold a reservation at the quoted fare, or allow cancellation without penalty, for 24 hours when the booking is made at least a week before departure. In the EU, Regulation 261/2004 sets passenger rights for denied boarding, cancellation and long delay, including care, a choice of reimbursement or re-routing, and compensation that a carrier can avoid in some cases when it shows extraordinary circumstances.

Which policy evidence has to sit inside each case?
Rule layerWhat the case must recordCheck against
Fare rules at purchaseFare basis or brand, change and refund conditions, purchase timestampFare rule text captured when the ticket was issued
Ticketing deadlinesLast ticketing date, automatic cancel option and delay, segment linksOrder event log with timestamps
Property cancellation policyRate type, penalty deadline in hotel local time with offset, penalty basisRate plan as sold, not the current plan
U.S. refund ruleSchedule change size, airports, connections, cabin, passenger responseDOT refunds rule thresholds and effective dates
U.S. 24-hour ruleBooking time, departure time, hold or cancel option usedCarrier customer service plan and DOT rule
EU passenger rightsDeparture point, carrier, delay or cancellation timing, notice given, care providedRegulation 261/2004 and the carrier's stated reason
Waivers and goodwillWaiver code or name, issue date, eligible travel dates, approverWaiver notice in force on the event date

What should the task coverage, provenance and policy evidence review include?

Use the review below as the acceptance sheet for a sample. Each line has an inspection a reviewer can run on the sample itself, an acceptance condition and a reason to query or reject. A supplier that runs it first will usually find the same problems a buyer would.

At a glance

What should the task coverage, provenance and policy evidence review include

What should the task coverage, provenance and policy evidence review include?
Review lineWhat to inspect in the sampleAccept whenQuery or reject when
Task mixCount of cases by task type: voluntary change, schedule change, cancellation, refund, name correction, hotel modification, waiver useEvery task type the buyer named appears, with the rare ones flaggedOne or two task types make up nearly the whole sample
Disruption shareCases with involuntary changes, misconnects or property overbookingDisruptions appear at a rate the supplier can explain from operationsDisruptions are missing or look hand-picked
Selection methodThe written rule used to draw the sample and the population it came fromRandom or stratified draw with the strata statedShowcase cases chosen by staff
System provenanceSource system per field, export date and record countsEach field traces to a named system and extractFields appear with no stated source
Event timelineOrdered events per case with time zonesEvery event has a timestamp and zone; local and UTC can be reconciledTimes are local with no zone, or dates only
Linked bookingsShared identifiers across air, hotel, car and traveler profileDependent components join reliablyLinked components cannot be matched
Rule evidenceFare rule, rate plan, waiver or regulation attached to each decisionRule text and version effective on the event dateOnly the current policy is attached
AuthorizationWho approved exceptions and on what basisOverrides are labelled with role and reasonOverrides are indistinguishable from policy
OutcomeFinal reservation state, ticket or folio status and refund recordOutcome verified in the source systemOutcome inferred from the last agent note
Permissions and privacyTraveler identifiers, payment data, contract restrictionsIdentifiers removed or tokenized; no card data; distribution terms checkedPassport numbers, card data or restricted content present

What provenance should a travel sample prove before anyone looks at quality?

Provenance comes before quality because a well-formed case from an unknown source cannot be licensed. A travel company's servicing history is usually spread across several systems: a booking engine or reservation system for air, a property management system such as Oracle OPERA or Mews for hotels, a contact center platform such as Salesforce Service Cloud or Zendesk for requests and notes, and a mid-office or accounting system for refunds and folios. Each contributes different fields to one case.

For the sample, record which system supplied each field, the extract method and date, the filters used and the record counts per system. Note where records were received from another party, such as a consolidator, a global distribution system or a property under a management agreement, because those records may arrive under that party's terms rather than yours. Where two systems disagree, for example a refund amount in the contact center note and a different amount in accounting, keep both and mark which one is the system of record.

A buyer will also ask whether the sample came from the same pipeline as the full dataset. If the sample was assembled by hand and the full delivery will be automated, say so, and plan a check that the automated output matches the sample on the same cases.

How much coverage is enough for a sample to be representative?

Coverage is judged against the buyer's task list, not against booking volume. Most reservations are never changed, so a random draw of bookings yields few servicing cases. Draw from servicing events instead, stratified by task type, and say so.

A workable approach is to agree the strata first, then set a minimum count per stratum and draw randomly within each. As an illustrative example, a sample might target an equal minimum for each of seven task types and then add cases in proportion to their observed frequency, so rare tasks are visible without distorting the overall picture. Report the population counts per stratum alongside the sample counts; a buyer can then judge how the full dataset will look.

Coverage also has a depth dimension. A cancellation case with only the request and the final status covers the task type but not the work. Count how many cases in each stratum carry the full event timeline, the rule evidence and the verified outcome, and report that number as well; it is often the figure that decides acceptance.

Seasonality matters more in travel than in most domains. A sample drawn from one month can miss winter storm disruption, summer peak overbooking or a waiver period. Spread the draw across the period offered, or state the months it covers and why.

Which dependent booking, purchase-time condition or approval limits the proposed change?

A reviewer should be able to answer this question from the case alone. If the answer needs a phone call to the operations team, the case is incomplete. These rules help spot the gaps.

  • A segment change must show its effect on the connection and on any linked hotel or car booking; a case that shows only the flight is a partial task.
  • Change and refund eligibility follows the fare or rate as purchased, so the case needs the purchase timestamp and the rule text from that moment.
  • Deadlines need time zones. A hotel penalty deadline stated in hotel local time must be comparable to a cancellation request logged in UTC or in the agent's time zone.
  • Regulatory triggers depend on measured facts: the size of a schedule change, the itinerary type, the booking lead time. Keep the measurements, not just the conclusion.
  • Corporate travel adds approval policy. A traveler's request may be valid under the fare and still need a travel manager's approval under the company's policy; keep that approval or note its absence.
  • Waivers override normal rules only within their stated dates and routes, so the case must name the waiver and its dates.

Worked review (illustrative): a schedule change on a two-segment trip with a linked hotel

This case is illustrative and uses counts and durations only. A traveler holds a two-segment domestic itinerary with one connection and a two-night hotel stay at the destination. The carrier retimes the first segment, which breaks the connection; the carrier rebooks the traveler on a later connection that arrives four hours after the original arrival time. The traveler declines the rebooking and asks for a refund, then cancels the hotel 30 hours before arrival.

To pass review, the case needs the original and changed schedules with time zones, the measured change in arrival time, the rebooking offer and the traveler's response, the refund record, the hotel rate plan as sold with its penalty deadline, and the hotel cancellation timestamp. With those, a reviewer can check two decisions: whether the air refund was owed under the significant change test and whether the hotel cancellation fell before or after the penalty deadline once both times are in the same zone.

Compare the version most samples actually contain: the final refunded ticket, a cancelled hotel folio and an agent note saying the customer was unhappy with the new times. Every decision in that version is unverifiable. It may still be worth keeping as a negative example of documentation, but it should not count toward the coverage target for schedule changes.

The exception: if the carrier issued a waiver for the route that week, the waiver may have offered free changes that the traveler could have used instead. The case needs the waiver notice, or a statement that none applied, before the refund decision can be graded. A sample that contains many such cases without waiver evidence should be queried rather than accepted.

When should a travel sample be narrowed, rejected or sent for specialist review?

Reject or narrow first on data that should never move. The PCI Security Standards Council's storage guidance says full track data, card verification codes and PINs must not be stored after authorization, so any sign of them in notes, call transcripts or attachments stops the sample and goes to security. Passport and other government identifiers, date of birth and loyalty numbers should be removed or tokenized before any sample is shared.

Send for specialist review when cases involve minors traveling alone, travelers' medical or accessibility requests, corporate traveler records governed by a client contract, or content received under distribution, consolidator or property agreements that limit reuse. Those questions turn on contracts and privacy law rather than data quality.

Recheck an accepted sample when the ground moves. A new fare family, a revised property cancellation policy, a change to the refund rules, a system migration or a change in what permissions travelers gave can each make earlier cases unrepresentative of the records that will be delivered. Keep the acceptance sheet with a version date, and rerun the lines the change touches rather than assuming the earlier result still holds.

Limit the claims made about a sample, too. A sample that passes this review shows the records can support the tasks; it does not prove model gains. If the buyer needs that, run a separate pilot with a held-out set and agree the test before the full dataset is scoped.

Preparing a travel sample before a SourceX assessment

Travel and hospitality companies that decide to license servicing records can do so through SourceX, which covers sourcing and qualification, rights review, contracting, data preparation, delivery and payment. The first check looks at the company, not at uploaded records.

Running the review above on your own records before that conversation shows where coverage, provenance or rule evidence is thin. Records stay with the company until it has approved the scope and terms and signed, and de-identification is then carried out by SourceX through approved processing routes.

Frequently asked questions

What is travel dataset sample acceptance?

It is the buyer's decision that a sample of travel servicing records is good enough to scope a full dataset. The decision rests on task coverage, traceable provenance, rule evidence dated to each event and safe handling of traveler, payment and contract-restricted data.

Why is a random sample of bookings a poor travel workflow sample?

Most bookings are never changed, so a random draw of reservations yields very few servicing tasks. Drawing from servicing events and stratifying by task type gives the buyer the changes, cancellations and disruptions the model will actually handle, with population counts reported alongside.

Do time zones really matter for travel workflow data?

Yes. Hotel penalty deadlines are commonly stated in the property's local time, while requests and system events may be logged in UTC or another zone. Without zones on every timestamp, a reviewer cannot tell whether a cancellation came before or after a deadline.

Should regulatory rules like the DOT refund rule be attached to each case?

Attach the facts the rule depends on and a reference to the rule version in force on the event date. Rules change, and thresholds such as the size of a schedule change only matter if the measured change is in the record.

Can a travel sample include payment card details if they are masked?

Keep card data out of samples entirely. PCI guidance prohibits storing card verification codes, PINs and full track data after authorization, and masked card numbers add nothing to a servicing task that a token cannot provide.

Sources

  • Amadeus's Flight Offers Search specification defines a last ticketing date until which an offer booked on the day of the search is valid for ticketing. Source
  • Amadeus's Flight Create Orders specification defines a ticketing agreement with an option such as delay to cancel, a delay before an automatic process applies if no ticket is issued, and affected segment identifiers. Source
  • Amadeus's Hotel Search specification defines a cancellation policy deadline after which a penalty applies, expressed in hotel local time with the time zone offset. Source
  • DOT's April 2024 refunds rule defines significant changes, including departure or arrival moves of three hours or more domestic and six hours or more international, and requires automatic refunds when passengers do not accept the change. Source
  • DOT requires carriers to hold a reservation at the quoted fare or allow penalty-free cancellation for 24 hours on bookings made at least one week before departure. Source
  • EU Regulation 261/2004 sets passenger rights to care, reimbursement or re-routing and compensation for denied boarding, cancellation and long delay, with an extraordinary circumstances defense to compensation. Source
  • PCI SSC storage guidance says full track data, card verification codes and PINs must not be stored after authorization. Source

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies