Philippines staffing research · Updated September 18, 2026

What causes candidate interview rescheduling loops?

A cohort study of interview requests, confirmations, changes, and completed events for recruitment coordination.

Decision in scope. This study is designed for a buyer considering recruitment coordination support. It answers whether interview coordination has stable inputs and escalation rules, which scheduling defects deserve repair, and where recruiters must retain judgment. It does not estimate a universal industry benchmark, promise savings, or rate individual workers. The useful output is a reproducible description of one buyer's current queue and the controls needed before responsibility expands.

Research question. After an interview request is issued, which observable events precede one reschedule, repeated rescheduling, completion, withdrawal, or an unresolved appointment? The question is fixed before records are reviewed. A clear question prevents analysts from selecting a favorable metric after seeing the data and keeps the work tied to a real staffing decision rather than a generic outsourcing claim.

Evidence base and checked sources. The protocol was prepared from current primary guidance checked on September 18, 2026. GAO's internal-control standards inform authorization, segregation, accurate recording, service-organization communication, and monitoring. NIST SP 800-53 informs attributable, time-ordered records and least-privilege boundaries. FTC guidance supports data minimization and need-based access. The Philippine Data Privacy Act provides the local privacy context. AAPOR transparency elements inform disclosure of methods, exclusions, and limitations. These sources support the research design; none supplies results for the buyer's queue.

Population and study window. Include all first interview requests created for two defined job families during ten consecutive weeks, followed for fourteen days or until a terminal event. Freeze the start and end dates, eligibility rules, systems, working timezone, and terminal events in a short protocol before export. Keep a screening log with counts for included records, excluded tests, missing identifiers, out-of-window events, and records withheld because access was not permitted.

Unit of analysis. Count one candidate-stage interview request, retained under the same identifier through time changes and channel changes. Replies, reminders, exports, edits, and status changes are events attached to that unit, not additional units. This distinction is essential: counting activity instead of cases can make a noisy process appear larger or more successful without showing whether the underlying work reached a usable outcome.

Field dictionary. Capture only the minimum fields necessary for the registered question: request time, proposed windows, timezone shown, interviewer roster, required attendees, confirmation event, calendar delivery, candidate response, change initiator, reason code, final event, and source completeness. For every field, record its allowed values, source, timestamp meaning, timezone, missing-value treatment, and whether it can change. Store stable pseudonymous identifiers in the analysis file and keep direct identifiers in the approved operating system.

Controlling and supporting records. Treat the applicant-tracking-system event history paired with calendar delivery and response records permitted for the study; free-text notes are minimized and coded only when needed. When two sources disagree, retain both values, their effective times, and the conflict class. Do not silently select the value that makes the case look complete. Source precedence must come from the buyer's approved operating rule, not from analyst preference.

Outcome taxonomy. Classify each unit as completed as first scheduled, completed after one change, repeated change, candidate withdrew, company cancelled, interviewer unavailable, timezone conflict, delivery failure, duplicate request, and unresolved at cutoff. The classes must be mutually understandable even when they are not perfectly mutually exclusive. If multiple conditions can coexist, preserve a primary disposition and separate flags. Keep an unresolved class; forcing incomplete histories into a successful or failed state creates false precision.

Measures. Report raw numerators and denominators before percentages. The primary measure is the distribution of registered outcome classes. Secondary measures are completion proportion, reschedules per request, elapsed time to confirmation, change initiator, delivery failures, missing timezone, interviewer conflicts, withdrawal timing, and loss to follow-up. Show medians and selected percentiles for elapsed times, because a mean can hide a long tail. Report missingness and censoring beside every affected measure.

Chronology. Reconstruct events in the site's configured timezone while retaining original offsets. Record creation, first action, each material transition, owner request, response, correction, and terminal event. Establish a deterministic rule for events sharing a timestamp. Never infer active work from elapsed time: distinguish preparation, waiting, scheduled delay, system delay, and unknown time whenever the record permits.

Quality checks. Two reviewers independently reconstruct a stratified sample containing completed, repeatedly changed, withdrawn, and unresolved requests, then reconcile against a frozen event dictionary. Reviewers use the same frozen codebook and record disagreements before discussion. Publish agreement counts for the sample and list rules changed after reconciliation. If a rule changes materially, recode the affected population or label the analysis exploratory rather than mixing definitions.

Operational example. An interviewer changes availability after a candidate confirms, the calendar invitation updates, and the candidate replies in a separate channel. The chronology must preserve all three events rather than blaming the last person who touched the record. This example demonstrates classification, not an observed company result. The real study must preserve the actual chronology and may reach a different distribution. Examples should be used to test whether the protocol gives two reviewers the same answer, not to manufacture evidence.

Role design boundary. A Philippines-based specialist can collect permitted records, link identifiers, apply the codebook, flag conflicts, prepare exception packets, and maintain the study log. Candidate assessment, accommodation decisions, employment eligibility, selection, rejection, compensation, and exceptions to hiring policy stay with authorized recruiting or people owners. Location does not change authority. Approaching deadlines and absent owners do not convert silence into permission.

Privacy and security. Use named accounts, least privilege, approved exports, encrypted transfer and storage, and a written retention period. Do not copy whole inboxes or unrestricted personnel, customer, or financial histories when a bounded event table will answer the question. Remove analysis copies when the retention purpose ends and log any access or data-quality incident through the buyer's existing route.

Bias and alternative explanations. Separate description from inference. Differences can reflect case complexity, workload, system design, source quality, shift coverage, policy changes, or missing records rather than coordinator performance or an outsourcing model. Stratify only on variables registered in advance and show small groups as counts; suppress or combine cells where privacy or instability requires it.

Known limitations. Private messages may be missing, calendars can rewrite event histories, reason codes may reflect convenience rather than cause, and job family or interview stage may confound comparisons. A single buyer, short window, and operational sample do not establish causality or external validity. The study cannot prove what a person knew, why an event occurred, or how another company would perform. Record changes to systems, policy, staffing, and demand that overlap the window.

Decision rule. Before collection, state what evidence would lead to each action: proceed with a narrow pilot, repair records and repeat measurement, retain the process internally, or request specialist review. The rule should consider evidence coverage and unresolved risk, not only speed. A faster queue with unclear authority or missing records is not a successful operating design.

Pilot design. If the evidence is usable, test one bounded queue with a named internal owner, backup, permitted actions, stop conditions, sample review, and rollback path. Compare the pilot with the registered baseline using the same units and classifications. Do not add tasks mid-pilot without versioning the scope and explaining how the change affects comparability.

Niche conclusion. Use event-linked evidence to repair scheduling inputs and delivery paths; do not turn reschedule counts into candidate or coordinator performance rankings. For an OutsourcedCompany.com buyer, the staffing question follows the process question: document what can be observed, keep consequential decisions with accountable owners, and expand only when the evidence path remains intact under ordinary exceptions.