Ask an integration partner who is alerted when a sync fails, how failed messages are retried or quarantined, which logs your team can read, and who is named as owner of each component after go-live. Written answers to these four areas show whether the partner has operated integrations in production, not just built them.
Key takeaways
- Ask how failures are detected before you ask how they are fixed.
- Failed messages need a defined retry, quarantine and replay path that someone actually runs.
- Logs must be readable by your own team, and personal data in them should be kept to the minimum needed.
- Every component, including credentials, hosting and vendor-side dependencies, needs a named owner in writing.
Why does monitoring matter before you sign?
Monitoring decides how quickly you learn that an integration has stopped working. Without it, a broken connection between your order system and your warehouse may surface only when a customer complains or a month-end report is wrong.
Most integration failures are silent. An API returns a success code but the payload is incomplete, a field mapping changes after a vendor update, or a queue fills up slowly over a weekend. A partner who only tests the happy path will not notice these cases, so the questions you ask should focus on how they would.
For background on how connected systems are designed, see our guide to maximising business efficiency through ERP and CRM integration, which covers the data flows that monitoring has to watch.
What should monitoring cover?
Monitoring should cover four layers: whether the connection is up, whether messages are flowing, whether the data is correct, and whether the business outcome happened. Each layer catches a different kind of failure, and a partner who only offers the first one leaves gaps.
Connection and throughput
Ask which endpoints are checked, how often, and what counts as a failure. A timeout after thirty seconds may be a failure for one system and normal for another. Thresholds should be written down and agreed, not left to defaults.
Data correctness
Reconciliation checks are often the most valuable and the least common. A daily comparison of record counts or totals between two systems can catch silent data loss that no connection check will ever detect.
Alert routing
An alert that goes to a shared mailbox nobody reads is the same as no alert. Ask who receives each alert, what the escalation path is at night and at weekends, and how alerts are silenced when planned maintenance happens.
How should a partner handle errors?
A good partner separates errors into those that can be retried automatically and those that need a person to decide. Transient network faults usually retry safely, while a rejected invoice with a missing tax field should stop and wait for review.
The mechanics matter. Ask whether retries use backoff, whether there is a maximum attempt count, and whether a message that is retried can be processed twice. Duplicate orders and duplicate payments are the usual cost of a retry without idempotency keys.
- Failed messages are stored in a dead-letter queue or equivalent, not discarded.
- Each failed message records the source, the error, the time and the attempt count.
- There is a documented replay procedure, and someone is allowed to run it.
- Manual fixes are logged so the same error does not return unnoticed.
Logs are part of error handling, and they often contain personal data such as names, email addresses or order details. Regulation (EU) 2016/679 (the GDPR, published on EUR-Lex) sets principles such as data minimisation and storage limitation that apply to what is written to logs and how long it is kept. This article is general information and not legal or tax advice. Rui Codex can provide technical implementation support for logging design, but not legal compliance advice, audits or certification, so confirm retention rules with your own counsel.
Who owns the integration after go-live?
The party that builds an integration is not always the party that runs it. Ownership should be written per component, so that a failure at 2 a.m. has a clear first responder and a clear decision-maker.
Consider the components separately: the integration code, the hosting environment, the credentials and certificates, the vendor APIs your system depends on, and the field mappings that encode your business rules. Each has a different owner in most real projects.
Hypothetical examples of an ownership gap
Here are two hypothetical scenarios:
In the first, a company's partner hosts the integration on its own infrastructure, but the API key belongs to the company's former IT contractor. When that person leaves, nobody can rotate the key, and the sync stops until the contractor is reached. The fix is a credential register with named holders, reviewed at each staff change.
In the second, a partner maintains the mapping between an ERP and an e-commerce platform, but a product team changes a tax category on the ERP side without telling anyone. Orders import with the wrong rate and no error is raised. The gap is not technical; it is that no owner was responsible for mapping changes. A change process with a named approver would have caught it.
Both examples are constructed to show failure modes and contain no results from real organisations.
Questions to put to a partner in writing
Written answers are more useful than verbal reassurance, because they can be compared between bidders and referred back to later. Use the list below as a starting point for the proposal stage.
- Which events trigger an alert, and who receives it at each hour of the day?
- How are duplicate messages prevented when a retry happens?
- Where are failed messages stored, and how long are they kept?
- What reconciliation checks run, and how often?
- Who owns credentials, hosting, mappings and vendor dependencies after handover?
- What is the process for planned maintenance, and how are stakeholders told?
If you are comparing providers for a project that connects an ERP and a web shop, the ERP and CRM integration service page describes the scope Rui Codex covers, and the integrations service page covers the wider range. Published case studies show the projects that have been documented publicly.
Frequently Asked Questions
What is the first question to ask an integration partner about monitoring?
Ask how the partner would know that the integration had failed without a user reporting it. The answer should name specific checks, alert recipients and the time it would take to respond, and it should not rely on the partner simply watching a dashboard.
Should a partner handle retries automatically?
Automatic retries are appropriate for temporary faults such as network timeouts, but only with a maximum attempt count and protection against duplicate processing. Errors that depend on data, such as a missing mandatory field, should stop and go to a person for review rather than being retried repeatedly.
Who should own credentials and certificates after go-live?
A named person at your company should own them, with the partner documented as having operational access. Record the expiry dates and the holder of each secret in the handover pack so that a staff change does not stop the integration.
Is a reconciliation check necessary if the connection is working?
Yes, because a working connection does not prove that the right data arrived. Reconciliation compares totals or record counts between systems and detects silent data loss that connection monitoring cannot see.
Conclusion
Monitoring, error handling and ownership are the parts of an integration that are easiest to skip during a project and hardest to fix after go-live. A partner who answers specific questions about alerts, retries, logs and named owners is showing how the work will be run, not only how it will be built. Compare proposals on these written answers as carefully as on the technical design. To discuss your own integration requirements, request a project consultation at Rui Codex or email info@ruicodex.com.
Sources
- Regulation (EU) 2016/679 (General Data Protection Regulation) - EUR-Lex, Publications Office of the European Union