Before an AI-assisted accounts receivable review begins, compare the current export schema with the approved schema. Missing, renamed, duplicated, or type-shifted columns should trigger a human review instead of being silently mapped into invoice priorities or a collections forecast.
Why schema drift matters
An AR analysis depends on field meaning, not only file shape. A column called balance may mean current open balance in one export and original amount in another. A date may switch from due date to invoice date. Text values may replace structured statuses. Even when every row loads successfully, those changes can alter the evidence used for prioritization and timing.
NIST's AI risk-management materials emphasize monitoring, documentation, and response to changes in data and system behavior. GAO's AI accountability framework also organizes governance around evidence, performance, monitoring, and human oversight. For AR work, a practical response is a deterministic schema check followed by named finance review.
The QuickBooks AR export-columns checklist defines the business meaning that should accompany a column list.
Run four checks before analysis
Presence
Confirm every required field exists: stable invoice identifier, customer reference, invoice date, due date, original amount, open balance, status, and snapshot context. A smaller approved schema may be appropriate, but it must still support the stated analysis.
Identity
Detect renamed, duplicated, or reordered fields. Do not map a new column solely because its label looks similar. Require an approved mapping with source and target definitions.
Type and format
Check dates, currencies, decimal separators, negative values, blank values, and status enumerations. Record whether a change is a format conversion or a change in business meaning.
Distribution and reconciliation
Compare row count, distinct invoice IDs, total open balance, date range, and status counts with the prior authorized snapshot. These are control totals, not proof that every row is correct. Material differences need explanation.
Human review record
When a check fails, preserve the source file identifier, export time, expected schema version, observed difference, affected fields, reconciliation totals, proposed mapping, reviewer, disposition, and rollback target. Possible dispositions are accept, correct source, update approved mapping, exclude affected field, or stop the run.
The finance AI review record provides the accept, change, reject, and escalate pattern. The forecast-ready AR data checklist helps evaluate whether the corrected source supports decision use.
Do not send credentials, unnecessary personal data, or unrelated customer documents merely to resolve a mapping problem. AR Pulse works from an authorized customer-provided export and does not require QuickBooks credentials.
Decision rule
Schema change is a review event, not a hidden preprocessing detail. Stop affected calculations, reconcile meaning and totals, approve the mapping, and retain a reversible record. Finance teams can view the AR Pulse example for the intended source-row traceability boundary.