Connecting two business systems can remove repeated data entry, shorten handoffs and give a team a clearer view of work. But an available application programming interface (API) does not automatically make an integration sensible or safe.
Before implementation, the business should be able to explain the outcome, the data flow, the owner, the failure path and the ongoing operating model. If those answers are unclear, the first task is not coding. It is scoping.
This article offers eight questions for evaluating an integration between systems such as a website, CRM, ecommerce platform, accounting tool, support desk or internal workflow application.
What does a system integration actually do?
An integration allows separate systems to exchange data or trigger actions. For example, a website enquiry might create a record in a CRM, an ecommerce order might pass fulfilment details to another platform, or a support request might open a task for the right team.
The connection may use an API, a webhook, a built-in connector, an automation platform or a more tailored component. The technical route matters, but the business design comes first: what should move, when should it move, which system is authoritative, and what should happen when the expected flow breaks?
The Agent Infinite Biz guide to building, buying or integrating makes the same distinction at a higher level: integration is useful when suitable systems already exist but their handoffs create unnecessary friction. The questions below help test whether a particular connection is ready to scope.
1. What business outcome should the integration improve?
Start with a specific handoff, not a broad instruction to “connect everything.”
A useful outcome might be:
- reduce duplicate entry between an enquiry form and a CRM;
- give operations a confirmed order without copying it from email;
- route a support request according to product or customer type;
- keep a customer status consistent across two approved systems; or
- make one agreed dataset available for reporting.
Describe the current process, the people involved, the delay or error it creates, and the desired result. This establishes a baseline for deciding whether an integration is proportionate.
If the workflow is rare, unstable or still changing every week, a simple manual step may remain the better temporary choice. Automating an unclear process can make its ambiguity faster and harder to see.
2. Which system is the source of truth?
When the same customer, order or task appears in more than one system, the team needs to know which record is authoritative for each field.
For example:
- the ecommerce platform may own order and payment status;
- the CRM may own sales-stage and relationship notes;
- the accounting system may own invoice and settlement records; and
- the identity system may own user access.
Without this decision, a two-way connection can create loops, conflicting edits and records that overwrite one another. Define the direction of travel for every important data element. Some fields may move one way only; others may need a carefully controlled update rule.
It is also worth defining how duplicate records will be identified. An email address, account number or order ID may be useful, but no single identifier is appropriate in every context. The matching rule should reflect the real data and the consequences of a wrong match.
3. What data genuinely needs to move?
An integration should not copy an entire record merely because the API makes it possible. List the minimum fields required for the stated outcome, who may access them, how long they are needed and whether any are sensitive.
Singapore’s Personal Data Protection Commission recommends tracing collected data to a valid purpose, minimising unnecessary collection, classifying datasets and assessing risks when external ICT parties are involved. Its Guide to Data Protection Practices for ICT Systems also emphasises clear accountability, access controls, retention and ongoing risk review.
That guidance is broader than API integration, but it supports a practical rule: move only what the workflow needs, protect it according to its sensitivity and do not leave copied data unmanaged.
This is not a substitute for legal or compliance advice. Requirements will vary by jurisdiction, industry, contract and the types of data involved.
4. How will access and credentials be controlled?
The integration may use an API key, access token, service account or another authentication method. The exact mechanism depends on the platforms, but the ownership questions are consistent:
- Who approves the connection?
- What is the minimum access it needs?
- Where are credentials stored?
- Who can rotate or revoke them?
- What happens when an employee or supplier leaves?
- How will access be reviewed?
The goal is to avoid a permanent, broadly privileged credential that nobody clearly owns. Recent NIST guidance on tokens and assertions highlights secure-by-design controls, key and token lifecycle management, verification and continuous monitoring. These are implementation concerns, but they should influence the scope before development begins.
5. What happens when one system is slow, unavailable or sends bad data?
The normal path is only part of an integration. The failure path determines whether the business can operate when something goes wrong.
Clarify:
- whether a failed action should retry automatically;
- how repeated submissions will avoid duplicate orders, contacts or tasks;
- where unsuccessful records will wait for review;
- who receives an alert and what information it contains;
- whether staff can complete the process manually; and
- how the two systems will be reconciled afterwards.
Some actions are safe to retry. Others, such as charging a payment method or sending a customer notification, need stronger protection against duplication. The answer depends on the workflow rather than a universal retry rule.
6. What platform limits and changes could affect the connection?
APIs and connectors have operating constraints. These may include request limits, usage charges, permission tiers, field restrictions, version changes, deprecation dates and differences between test and production environments.
Before committing to a design, check the current documentation and commercial plan for both systems. Record any assumptions that could change the scope or running cost.
The NIST Guidelines for API Protection for Cloud-Native Systems describe API protection across development and runtime, including an incremental, risk-based approach. OWASP’s API Security Project also calls attention to authentication, authorisation, resource consumption, inventory management and the unsafe consumption of third-party APIs. A small business integration may not need enterprise-scale architecture, but it still needs controls proportionate to the data and business impact.
7. Who will monitor and maintain the integration?
An integration is an operating component, not a one-time transfer. Assign a business owner and a technical owner, even if the same person initially holds both roles.
Agree what will be monitored, such as:
- successful and failed transactions;
- records waiting for manual action;
- unusual volume or cost;
- expired credentials;
- schema or field changes;
- vendor notices; and
- differences between the two systems.
The team should also know where documentation lives, who can contact each vendor and what response is expected when an alert appears. Monitoring without a named response owner simply creates another unattended inbox.
8. What is the exit or rollback plan?
A connection may need to be paused, replaced or removed because the process changes, a vendor is replaced or the integration no longer provides enough value.
Define how to:
- disable the connection without losing essential work;
- revoke credentials and access;
- identify records created or changed by the integration;
- export or retain required data;
- return temporarily to a manual process; and
- confirm that no obsolete endpoint or account remains active.
This does not require an elaborate disaster-recovery programme for every small workflow. It does require a realistic way to stop safely.
A practical way to stage the work
Once the eight questions have credible answers, a staged implementation can reduce risk:
- Map one narrow workflow. Document the trigger, systems, fields, decision rules and owners.
- Test with controlled data. Confirm field mappings, permissions and failure behaviour without relying on real customer activity where avoidable.
- Run a limited live scope. Start with one form, team, product line or transaction type.
- Observe and reconcile. Compare records, review failures and confirm that the outcome is actually improving.
- Expand deliberately. Add more cases only after the first scope is stable and owned.
This staged approach may reveal that a platform’s built-in connector is sufficient. It may also show that a process change is more valuable than custom development. That is a useful result: the goal is not to maximise technical complexity, but to create a reliable handoff.
What should go into an integration brief?
Before asking a developer, vendor or internal team to propose a solution, prepare a short brief covering:
- the current workflow and business problem;
- the desired trigger and outcome;
- the two systems and current subscription plans;
- the source of truth for each important field;
- example records and known exceptions;
- personal or sensitive data involved;
- access, approval and ownership requirements;
- expected transaction volume and timing;
- failure, alert and manual fallback needs; and
- success measures for the first stage.
This gives the technical team enough context to assess feasibility without assuming that every available endpoint should be used.
Integration starts with ownership, not the connector
A good integration should make a defined workflow easier to operate and easier to understand. The connection itself is only one part of that outcome.
If the purpose, data boundaries, access model, failure handling and ownership are clear, the business is in a stronger position to choose between a built-in connector, an automation platform or selected custom development. If they are not clear, resolving those questions first can prevent unnecessary cost and risk.
Agent Infinite Biz helps businesses assess workflow automation, system integration and selected custom development around real operating needs. Discuss a workflow automation, API integration or system requirement with Agent Infinite Biz to talk through an integration requirement and a sensible first scope.
Comments
comments powered by Disqus