Philippines staffing research
When does redacted audio make a quality item unscoreable?
A reliability study for exclusion rules, secondary evidence, and reviewer disagreement.
Published September 9, 2026. This note examines which scorecard criteria become unobservable when required call audio is muted or redacted. It describes a study design for a defined outsourced-calling workflow. It does not report a measured effect or promise a business result.
Methodology: define the observation unit as one scorecard criterion whose evidence window overlaps a documented redaction. Set the eligibility rule, start point, end point, and exclusions before looking at outcomes. Keep unsuccessful and unresolved records in the denominator.
Data collection should cover call pseudonym, criterion, timestamp, redaction span, secondary record, reviewer decision, adjudication, and exclusion reason. Use pseudonymous identifiers when the analysis does not need a person's name. Restrict access, set a retention period, and keep consent and suppression records in their authoritative systems.
Analysis: report counts and denominators across criterion, redaction reason, reviewer, evidence source, and final scoreability decision. Show missing fields and unresolved cases. Use medians or interval bands for skewed elapsed times. Publish uncertainty estimates only when the sampling design supports them.
Scope and inference limits: Results apply to the sampled recordings and scorecard. They do not measure the caller behavior hidden by the redaction. Assignment may also differ by customer mix, source age, call lane, season, owner availability, and workflow changes. Treat those as competing explanations.
Quality control: pilot the codebook on a small set of eligible records. A second reviewer should classify ambiguous cases without seeing the first decision. Report disagreements, exclusions, and any rule changed after the pilot.
Limitations: missing audio, late notes, inconsistent timestamps, small groups, and incomplete owner outcomes can distort the result. Show how many records each limitation affects. Do not fill a missing outcome with the most likely disposition.
Use the finding narrowly. It may justify a clearer field, a revised queue rule, or a prospective test. It cannot justify a claim about an individual caller or contact without evidence from that case.
Sources: NIST Privacy Framework (https://www.nist.gov/privacy-framework); FTC Telemarketing Sales Rule (https://www.ftc.gov/legal-library/browse/rules/telemarketing-sales-rule); AAPOR Best Practices (https://aapor.org/standards-and-ethics/best-practices/); Philippines National Privacy Commission, Data Privacy Act guidance (https://privacy.gov.ph/data-privacy-act/).