ABA Scheduling KPI Dictionary

Start with the decision

State who uses the metric and what decision it informs. A measure for daily staffing needs a different cohort and refresh cycle from a quarterly access review. Remove metrics that have no owner, action, or clear interpretation.

Use a standard dictionary record

Each KPI entry should be complete enough for another analyst to reproduce it:

Include an example with raw records and expected output. A label alone is not a definition.

Define events and clocks

Name the start and end event for every duration. For a proportion, define each numerator condition and the complete eligible denominator. State the reporting window, maturity period, time zone, late-arriving data treatment, exclusions, and source version.

Keep several units from collapsing together

Visits, service hours, unique clients, staff shifts, authorizations, claims, and dollars can show different patterns. State the unit in the metric name and report raw counts. One canceled four-hour visit and four canceled one-hour visits are equal by visit count but different by lost hours.

When a person or visit appears in several categories, define a mutually exclusive primary state for cohort totals and separate contributing factors for improvement analysis. Do not sum overlapping segments into a company total.

Keep different states separate

Useful scheduling states include requested, reviewed, released, delivered, canceled, held, unstaffed, and reconciled. HealthCare.gov cautions that preauthorization does not promise cost coverage. Authorization, claim acceptance, adjudication, payment, and clinical response require their own measures.

Define missing, unknown, and not applicable

A blank value can mean data was not collected, source unavailable, field irrelevant, or integration failed. Give each meaning an explicit state. Excluding blanks without explanation can improve a rate while hiding the exact records that need work.

Report source completeness and unmapped values with the KPI. Use not applicable only when the rule genuinely excludes the record. Preserve the accountable owner and correction plan for unknown or missing data.

Assign data and interpretation owners

Operations may own schedule-state definitions and source reconciliation. Qualified clinicians own clinical interpretation. Payer, payroll, access, privacy, and finance roles own their evidence. The CASP Organizational Guidelines public overview supplies high-level business, clinical-operations, and risk context; this dictionary is an editorial design.

Build a controlled change process

A proposed change should identify the problem, old and new definition, affected reports, historical comparability, owner, approval, effective date, and validation cases. Test ordinary, boundary, missing, late, duplicate, changed, and corrected records before release.

Display a definition version beside the dashboard result. If the organization restates history, retain the original publication and reason. Avoid changing a denominator during a performance period without making the break visible.

Validate the metric through a decision rehearsal

Choose one ordinary record, one boundary case, one missing-data case, and one corrected record. Ask an analyst to calculate the metric from the dictionary alone, then ask an operator to explain what action the result permits. Compare their cohort, event times, exclusions, maturity rule, and interpretation. If two careful people produce different answers, revise the definition or identify the unresolved source-of-truth decision.

Test sensitivity before setting a target. Show how the result changes when the maturity window extends, late-arriving records post, one site uses a different status, or unique clients replace visits as the unit. A stable decision should not depend on a hidden denominator change. If the measure is too sensitive for daily action, use it for trend review and choose a nearer operating signal with a clearer clock.

A fictional metric

Harbor Pine ABA locks 100 released visits for June. Eighty-six are delivered, six client-canceled, four practice-canceled, three unstaffed, and one inaccessible. Delivered yield is 86 of 100, or 86%. The dictionary also requires raw counts, service hours, reason owner, and linked recovery status.

Reconcile the fictional cohort

The five outcome groups sum to 100. Delivered yield counts only visits delivered under the defined event. A separate service-opportunity measure could include approved same-period changes if its definition says so. The six client-canceled visits remain in the released cohort and should not be removed after review.

Report planned and delivered hours, unique people affected, notice timing, access or system contributors, and recovery. The 86% result describes operations. It does not establish clinical quality, payer acceptance, or payment.

Govern changes

Version every definition and show the effective date. Recalculate historical results only under a declared restatement policy. Track source gaps, unmapped states, manual overrides, late data, and metric changes. A dashboard comparison is meaningful only when the underlying definition and population match.

Audit the metric from records to decision

Periodically sample source records, recompute the result, and compare it with the dashboard. Verify segments, exclusions, late updates, and manual corrections. Then ask whether leaders used the measure for its stated decision and whether that action had the intended operating result.

Retire metrics that duplicate another measure, cannot be reproduced, or create harmful incentives. Keep the historical definition available for earlier reports and regulatory or contractual review.

Owner KPI questions

• Does the dictionary name the decision, unit, cohort, numerator, denominator, clock, maturity, exclusions, and source?
• Are scheduled, authorized, delivered, billed, adjudicated, paid, and clinically meaningful states kept separate?
• Can a second person reproduce the value from locked records and explain its limits?
• Are missing, unknown, not applicable, corrected, and late-arriving data visible?
• Does a version change preserve historical comparability or label the break clearly?
• Is every target paired with a guardrail against distorted behavior or hidden service loss?

The metric is ready when its value, limits, and permitted action are all understandable. A precise number without a governed decision can still mislead.

A denominator-change example

River Stone ABA reports a 92% schedule confirmation rate using visits due tomorrow as the denominator. A new leader changes the report to all visits in the next seven days without updating the dictionary, and the rate falls to 71%. Neither value can be compared because the cohort and time available for confirmation differ. The team restores the original metric for trend continuity, creates a separately named seven-day readiness measure, and documents the decision each supports.

The example also shows why late data needs a rule. Confirmations received after the daily snapshot stay in the original cohort as late-arriving events and appear in a later final view, rather than silently rewriting the number used that morning. Leaders can compare preliminary and mature results, investigate the difference, and avoid treating data latency as staff performance.

Related resources

• ABA Appointment Status Taxonomy
• ABA Scheduling Ownership and RACI Matrix
• ABA Late Arrival Response Workflow
• ABA Schedule Exception Approval Log

Sources

• Council of Autism Service Providers, Organizational Guidelines public overview
• Behavior Analyst Certification Board, Ethics Code for Behavior Analysts
• HealthCare.gov, Preauthorization glossary