The operational problem
An order, a provider payment and a bank settlement describe different events. A cash-on-delivery parcel can be delivered before the merchant receives a settlement. A provider can report a payment while an order still shows an outstanding balance. The operational problem is to explain that difference without counting the same money twice. AI can help summarize exceptions; the rules authorizing a financial record change should remain explicit and testable.
Evidence to collect before acting
- Order snapshot
- Keep the order identifier, current financial state, amount still due, currency and the time the snapshot was read. Recheck them immediately before any update.
- Payment evidence
- Use an authoritative provider record with a transaction ID, payment state, gross amount and currency. Keep fees and settlement amounts separate from the customer payment.
- Matching record
- Retain the reference linking the payment to the order and any prior processing result. A similar customer name or equal amount is not a sufficient match.
A practical workflow
Step 1
Read and match
Read the order and payment records without changing either system. Require a unique match; route ambiguous references to review.
Step 2
Check payment meaning
Shopify distinguishes a transaction kind from its status. A successful authorization is not the same event as a captured payment. Pending, failed or unknown evidence must not be treated as receipt of money.
Step 3
Prepare a bounded update
Show the exact order, evidence reference and proposed amount. Check that the amount is positive, in the correct currency and within the current outstanding balance. Use the merchant’s explicit policy to decide whether review is required.
Step 4
Write once, then read back
Record a stable processing key before attempting a change. After the authorized update, read Shopify again. A timeout leaves an uncertain result until the destination is checked; it is not a reason to repeat the write blindly.
Worked example · Illustrative
Illustrative example, not a customer result: order DEMO-104 has an outstanding balance of 120 PLN. One provider record confirms 120 PLN, but a second imported row contains the same transaction ID. The workflow recognizes one payment, ignores the repeated record and prepares one order update. If only 80 PLN had been received, it would keep the remaining 40 PLN visible rather than marking the whole order paid.
Exceptions that need a separate decision
| Situation | Handling approach |
|---|---|
| Partial or excess payment | Keep the discrepancy in a review queue. Do not round a payment up, discard an excess amount or invent an explanation. |
| Already-paid order | Check whether the same transaction was processed earlier. Preserve the evidence and avoid creating another payment record. |
| Missing provider data | Report the check as unavailable and retain its last observation time. No new records is different from a failed connection. |
How to measure the result
Track uniquely matched payments, independently verified order updates, ambiguous matches, duplicate records and unresolved write outcomes. Measure human handling time on comparable orders before estimating savings, and include exception handling and maintenance. Successful API responses alone do not measure the business result.
Common questions
Can a delivered parcel prove that the order is paid?
No. Delivery, collection of payment and settlement to the merchant are separate facts. Select the authoritative payment event your reconciliation policy requires.
Should AI decide whether an uncertain payment is real?
It can organize the evidence and explain the discrepancy. Missing or contradictory financial evidence should remain unresolved until an authoritative source or an authorized reviewer settles it.
Primary sources
Platform documentation supports the distinctions described here. Workflow controls and measurement recommendations are design guidance from LINKIO.ONE, not evidence of a completed LINKIO deployment.