When an export, payment platform, delivery system, or buyer portal reports conflicting invoice statuses, an AI-assisted AR workflow should not choose a winner silently. Preserve each source claim, define the decision that matters, and route the conflict to a named human reviewer with a reversible record.

Model the conflict explicitly

Statuses often answer different questions. An accounting export may show an open balance, a payment platform may show a pending attempt, and a delivery source may show the invoice was viewed. Those can all be true at the same time.

Create one conflict record with:

  • invoice ID and source-system identifiers;
  • source name, status, amount, and timestamp;
  • extraction or observation time;
  • the business question being decided;
  • conflict type;
  • reviewer, decision, rationale, and review time; and
  • any follow-up or rollback action.

The AI AR source-lineage checklist provides the foundation for linking each normalized field back to its origin.

Classify before resolving

Use a compact taxonomy:

| Conflict | Example | Review focus | | --- | --- | --- | | Timing | Open export; later paid event | Preserve snapshot cutoff and append the later event. | | Scope | Customer subtotal; invoice-level balance | Reconcile included records and avoid double counting. | | Definition | Viewed; pending; paid; deposited | Identify which operational question each status answers. | | Identity | Two customer records appear to be one payer | Require supported entity mapping. | | Amount | Invoice, credit, and residual disagree | Trace adjustments and payment application. |

Do not resolve a definition conflict by applying a generic source priority. The authoritative source depends on the question. An accounting export supports the recorded open balance at its cutoff; a processor supports its own attempt status; bank evidence supports observed cash availability.

Require bounded human review

NIST's AI risk-management material emphasizes documented governance, measurement, and management practices. GAO's AI accountability framework likewise organizes oversight around governance, data, performance, and monitoring. Applied here, the practical control is a reviewer who can accept, change, reject, or escalate the proposed status with source evidence.

Use the finance AI review record to preserve that decision. The workflow should not contact customers, edit books, merge records, or change cash assumptions without the authorized owner.

Keep the output honest

If the conflict remains unresolved, show the invoice as unresolved, preserve its open amount for reconciliation, and avoid a confident collection week. Give it an owner and next review trigger. Do not convert missing resolution into a zero amount, a failed payment, or guaranteed delay.

If the input schema changed, first apply the source-schema drift checklist. A renamed or missing field can create a false status conflict that is really an ingestion problem.

Decision rule

Keep every source claim, choose authority by business question, and record the human resolution. AR Pulse is designed around invoice-level source traceability; it does not require accounting credentials, change books, contact customers, or guarantee collections. Finance teams can apply for AR Pulse for an authorized-export analysis with a ranked action list and four-week range.