A flight-refund routing study should answer a specific operational question: when a traveler asks for a refund, does the case reach a party that can decide it, with enough ticket evidence to act? The study does not estimate a universal refund rate and does not decide whether every traveler is legally entitled to a refund. It examines the path taken by observed cases, the information available at each transfer, the elapsed time, and the recorded outcome.
Refund contacts are difficult to compare because the word "refund" covers different situations. A significant schedule change is not the same event as a voluntary cancellation. A ticket issued by an agency may remain under agency control even when the operating carrier handled the disruption. One itinerary can include multiple coupons with different states. A traveler may also ask about a duplicate charge, unused ancillary service, or a payment reversal already in progress. The study must preserve those distinctions before it calculates a routing measure.
Define the case and the decision under study
The unit of analysis is one refund request tied to a ticket and a first recorded contact. A case may include later contacts about the same request, but those contacts remain attached to the original case rather than becoming fresh successes or failures. The study window and observation cutoff are fixed before extraction. Cases still open at the cutoff remain open; analysts do not force them into an outcome category.
Eligibility rules should name the request types included, the channels observed, and the ticket records required for review. Test bookings, spam, contacts without a recoverable ticket reference, and matters completed before the observed contact can be listed separately. Exclusion is not a way to clean up inconvenient results. Every exclusion needs a reason code and count.
The primary routing decision is whether the first receiving team retained the case, transferred it to the correct carrier or agency path, or sent it somewhere unable to act. The definition of "correct" comes from the documented ticket control and request condition available at that time. Later facts may explain an error, but they should not rewrite what the agent could see.
Reconstruct the ticket before judging the route
Reviewers need ticket stock, validating carrier, coupon status, operating carrier, disruption record, form of payment, booking source, prior exchange or refund activity, and the contact chronology. The extraction should retain event times and provenance. A status copied into a service ticket without its source is weaker evidence than a ticketing event joined by a stable identifier.
Coupon-level detail matters. One coupon may have been used while another remains open. An exchanged ticket may point to an earlier document. A refund request that appears stalled can actually be attached to the wrong document number. The reviewer records the chain used to reach the active coupon rather than treating the newest number as automatically authoritative.
Free-text customer messages should be minimized. The study needs enough text to classify the request and understand the promised action, not a reusable copy of personal travel details. Identifiers used for joining records should be protected in the working file and omitted from published examples.
Separate disruption cases from voluntary requests
A disruption-related request should carry the event that changed the itinerary and the time that event became available to support staff. A voluntary request should carry the fare or ticket restriction consulted by the agent. If the record does not support either classification, the case belongs in an unknown group.
This separation prevents an apparent routing problem from absorbing several different questions. A team may correctly send agency-issued voluntary cancellations back to the agency while mishandling carrier-controlled disruption refunds. Combining those groups would hide the operational difference. Results should therefore be reported by booking source, control, and request condition, with denominators shown for every row.
The study may describe how observed handling corresponds with published procedures, but it should not turn a customer-service review into a legal conclusion. The U.S. Department of Transportation refund guidance explains the Department's public refund position, while 14 CFR Part 260 supplies the governing federal rule text. Those primary sources define context for coding disruption-related requests; they do not decide an individual case inside this proposed study. Application to a particular case still depends on the facts and the responsible qualified party.
Observe handoffs, not just destinations
A transfer can reach the right team and still fail operationally. Reviewers should record whether the receiving party accepted ownership, whether the ticket evidence traveled with the case, and whether the traveler received a usable next step. A contact marked "transferred" without acceptance evidence is not equivalent to an owned case.
Agency and carrier handoffs need their own fields. The record should state why control rested with that party, what the traveler was told, and whether a later contact repeated the same transfer. When a traveler is bounced between organizations, each handoff remains visible. The study should not credit the first team merely because its routing label sounded plausible.
Elapsed time is measured between defined events: first contact, first routing action, acceptance by a deciding party, decision, and recorded settlement or closure. Missing timestamps remain missing. Analysts should not infer acceptance from a later note unless the protocol explicitly defines that evidence.
Code outcomes without inventing certainty
The outcome set should distinguish approved refund, declined refund, alternate resolution, duplicate request, withdrawn request, unresolved at cutoff, and unknown. A payment event can support settlement, but a service note saying "refund submitted" supports only the submission step. The codebook states which evidence is sufficient for each outcome.
Duplicate refund risk also needs a separate flag. Two contacts about one request are not automatically duplicates if one concerns an unused coupon and another concerns a fee. Conversely, two differently worded requests can seek the same payment action. The reviewer follows the ticket and payment evidence rather than matching phrases.
The primary measure can report the share of eligible cases routed to an acting owner, using the eligible cases as the denominator. Supporting measures can show time to acceptance, time to decision, repeat-contact frequency, unresolved share, and missing-evidence share. Every measure must publish its denominator, exclusions, and cutoff.
Test whether reviewers apply the rules consistently
A second reviewer should independently code a declared subset. The comparison covers eligibility, request condition, ticket control, first route, ownership acceptance, and outcome. Disagreement is informative. It may show that the codebook confuses ticket stock with operating carrier, that an agency handoff lacks a clear acceptance rule, or that settlement evidence is too loosely defined.
The team resolves disagreements without silently replacing the first labels. It records the original codes, final adjudication, reason, and date of any codebook change. If a definition changes materially, previously coded cases affected by that definition should be reviewed again. The study reports the size of the double-coded subset and the fields where disagreement occurred.
Analyze timing without claiming causation
Routing delay may vary with contact channel, interval staffing, disruption peaks, agency hours, ticket complexity, or missing coupon evidence. Those variables help describe the observed cases. They do not prove that staffing caused a refund result. A simple comparison between fast and slow cases is especially vulnerable because easier cases may both route faster and settle sooner.
The analysis should first show counts and distributions. Median or percentile timing can be useful when the underlying timestamps are reliable, but a single average can hide a long unresolved tail. The report should display unresolved cases at the cutoff rather than deleting them from timing calculations.
If the operation changes its routing instructions, the study can preserve the original cohort and run a later sample under the revised rule. That before-and-after comparison still requires caution because disruption mix, carrier behavior, and demand may change. The purpose is to detect whether the operating path improved, not to manufacture a causal claim.
Preserve a reproducible study package
The study package includes the observation window, extraction time, join logic, field definitions, inclusion and exclusion rules, codebook, adjudication log, denominators, missing-data counts, and analysis output. A second analyst should be able to rebuild every published table from the retained records without relying on a private explanation from the first analyst.
Publication should include the boundaries. The findings apply to the observed channels, ticket sources, and period. They do not establish a general airline benchmark. They do not replace case-specific review, and they do not show that routing alone produced settlement.
The proposed analysis should compare observed carrier commitments with the Department's Airline Customer Service Dashboard and treat the Department's consumer complaint guidance as a description of the public escalation channel. These sources support interpretation of recorded routes; they are not empirical findings from the study. The protocol records when each source was checked and keeps source interpretation separate from observed service data.
The final conclusion should state where cases were lost: missing ticket evidence, incorrect control assumptions, unaccepted transfers, incomplete traveler instructions, or unresolved decisions at cutoff. That conclusion gives operators a repair target while staying within what the records can support.