An unreliable integration usually needs a traceable investigation before it needs a replacement. Start with an affected business record and follow its path from the source through transformation, submission and destination processing. A successful run status is useful evidence, but it does not establish that the business result is correct.
Treat source queries as integration dependencies
A saved search or source filter can be part of an integration contract. If someone changes the criteria, columns or permissions, the integration may receive a different population or stop receiving records without producing the error you expect.
Document the resources each flow depends on and give changes to those resources a review path. Use representative records to verify inclusion, exclusion and changed values. Distinguish a genuinely empty source from a query or permission problem.
Validate mappings against imperfect data
List the required source fields, allowed values and reference mappings. Decide how to handle a missing item, an unknown location or an incomplete customer record. Reject invalid work with an actionable reason rather than silently inserting a default that changes the business meaning.
Preserve identifiers and the company or subsidiary context used by the destination. Check units of measure, dates and the finance-approved treatment of tax and rounding. Review custom fields and workflows that run after an import, because downstream behaviour can change the result of a technically valid request.
Check the shared concurrency budget
Oracle documents account-level governance across web services and RESTlet requests. Other integrations can consume the same capacity. Inspect actual account settings and the available monitoring rather than assuming a fixed limit or increasing parallelism blindly.
Coordinate schedules, bound concurrency and handle temporary refusals with a controlled retry policy. Establish whether the bottleneck is request capacity, a slow operation, a source query or an external dependency before changing the architecture.
Prove a retry is safe
An interrupted response does not prove that a destination write failed. Check the destination using a durable business reference before replaying an uncertain transaction. Preserve the source-to-destination relationship so a repair can recognise work already completed.
Test repeated and out-of-order requests in an appropriate environment. A record that became invalid because an accounting period closed needs a business decision; repeatedly submitting it does not fix the underlying condition.
Make monitoring actionable
Check who receives alerts, which histories are available and how long useful retry data is retained. Review recurring failures, unresolved records and schedules that have not run. Use reconciliation to detect missing or duplicated business records that ordinary error monitoring can miss.
A useful incident record contains the symptom, affected references, suspected cause, confirmed evidence, recovery decision and observed result. Keep confidential payloads out of general-purpose reports and use synthetic examples in reusable documentation.
Give changes an owner
Record the relationship between configuration, scripts, source resources and business rules. Test a proposed change against both the failed record and unaffected examples. Agree promotion, rollback and reconciliation before releasing it.
After recovery, keep the operating instructions with the integration. The owner should know how to distinguish an ERP configuration issue, a source-data problem and an integration defect, and where to escalate each one.
Discuss NetSuite integration and ERP support
Use the Celigo account audit checklist