A Reconciliation Discrepancy Workflow: Classification, Attribution and Closure

Discrepancies Are the Norm, Not the Exception

At 100k orders a month, hundreds to thousands of variances per reconciliation cycle is a normal water level. The question is not whether discrepancies exist, but whether they are classified, attributed and closed. A mature workflow should auto-attribute over 80% of variances and route the rest to humans within explicit deadlines.

A Five-Category Classification

Timing differences — revenue recognized at month end while cash arrives next month, or invoices crossing periods. No one is at fault; park them and let the next cycle clear them automatically. Amount differences — fees actually charged deviate from the rate card, or discount allocation calibers disagree. Verify against the rate card, then appeal to the platform or adjust the internal caliber. One-sided records — a statement row with no order (missed sync) or an order with no settlement row (not yet settled). Fix the data pipeline or park until the next cycle. Rounding differences — cent-level rounding and currency conversion; auto-attribute within tolerance and log. Disputed differences — wrong charges, duplicate deductions, missing reimbursements; generate an appeal ticket and track to recovery or write-off.

Attribution: From Symptom to Cause

Attribution answers three questions: which side is short, which fee item carries the gap, and which stage produced it. Work in order. First split sides: variance = receivable − received, establishing whether income is understated or fees overstated. Then split items: break the gap down by accounting item — merchandise payment, service fee, commission, shipping, reimbursement — where most variances reveal themselves. Finally locate: use bridge tables to trace back to the original statement row, confirming whether the source data really says so or the parsing stage introduced the error. Attribution that cannot land on a specific raw row remains guesswork.

Example auto-attribution rules: a variance equal to rate × base suggests a rate-caliber mismatch; an existing order missing from the statement before its settlement date is parked as unsettled; a variance under one cent is rounding; two charge rows for one order number triggers a duplicate-charge appeal.

Closure Loop and Deadlines

Every variance lives through registration (automatic), attribution (automatic or manual), handling (appeal, adjustment, or parking), review, and archiving. Suggested SLAs: rounding and timing differences close automatically on the same day; one-sided records get a pipeline investigation within one day; amount and disputed differences are verified within three days, with appeals tracked as a separate status and auto-cleared when the platform pays out.

Review Mechanisms

Sample 5% of auto-attributed variances for manual review and feed the findings back into the rules; force human review before closure for any single variance above a threshold (say 1,000 CNY); and publish a monthly variance distribution report by category, platform and store — stores with abnormal variance-rate swings deserve a business-side investigation. The maturity of the discrepancy workflow directly determines month-end closing speed. Treat variances as data assets to manage, not nuisances to eliminate.