API integration testing is easy to buy badly. A vendor can show a large test suite, a dashboard of green checks and a competitive day rate while the risks that matter remain untested: a partner changes a field, a retry creates a duplicate payment, a token expires halfway through a batch.
This buying guide is for a business that depends on integrations but does not have a large quality-assurance team. It covers the requirements to settle before conversations, the provider models to compare, a weighted scorecard, red flags, pilot terms and the handover you need at the end.
Define the risk before the scope
Testing is a risk-reduction purchase, so name the risks first. List every external system the business depends on: payment gateway, courier, ERP, CRM, marketplace, identity provider and bank. For each one, write the failure that would hurt most and the business consequence. A failed payment creates a customer complaint; a silently duplicated payment creates a reconciliation problem and a trust problem.
Then set acceptance criteria in business language. "Orders created in the storefront appear once in the ERP within five minutes, with the correct tax and delivery charge, and a failed ERP call retries without creating a duplicate" is testable. "Robust integration" is not.
Requirements to settle before vendor conversations
- Integration inventory: systems, owners, protocols, API versions and documented contracts for each connection.
- Environment access: sandbox or test credentials, test data and a named contact on each side. A vendor cannot test a partner API it cannot reach.
- Test data: representative products, customers, orders, refunds and edge cases, with personal data masked or synthetic.
- Acceptance criteria: measurable outcomes for correctness, latency, retries and error handling.
- Compliance constraints: data residency, retention, access control and any sector rules that affect log content.
- Current defects: known issues, workarounds and incidents that must not be repeated.
If a partner cannot provide a test environment, decide in advance whether the vendor will use a mock, a recorded contract or a production read-only check. Each option has a limit; write it down rather than discovering it during the pilot.
Provider categories
| Model | Best for | Strengths | Limits |
|---|---|---|---|
| Independent QA firm | Established product with several integrations | Test discipline, reporting, independent judgement | May lack deep knowledge of a niche partner API |
| Integration specialist | Complex partner or ERP connections | Protocol and domain depth, faster diagnosis | Testing may be tied to their implementation choices |
| Staff augmentation | A team that already has a test strategy | Flexible, works inside your process | You manage priorities, quality and knowledge transfer |
| Tooling vendor plus internal team | Mature engineering with CI discipline | Cheapest at scale, reusable assets | Tooling is not a strategy; gaps can hide in green dashboards |
| Managed API platform | Many partners and a need for ongoing monitoring | Contract testing, observability and alerts | Less bespoke investigation; platform lock-in risk |
A weighted scorecard
Score each vendor from one to five on the criteria below. Keep the evidence behind each score, because the discussion is more valuable than the total.
| Criterion | Weight | What good looks like |
|---|---|---|
| Integration and domain depth | 20 | Has tested the same protocol, partner type and failure modes |
| Test strategy | 20 | Risk-based plan with negative, contract and recovery tests |
| Environment and data handling | 15 | Clear plan for test data, masking, access and retention |
| Automation and CI fit | 15 | Tests run in your pipeline, not only on a vendor laptop |
| Defect triage and reporting | 10 | Severity definitions, reproducible evidence and daily summaries |
| Handover and documentation | 10 | Test assets, runbooks and a working session for your team |
| Commercial model | 10 | Fixed pilot, transparent change rates, no surprise environment fees |
What a credible test plan contains
A plan that only checks happy-path responses is a demo, not integration testing. Ask for the following, and expect the vendor to explain how each applies to your systems rather than reciting a template.
- Contract and schema checks: required fields, types, nullability and version changes.
- Authentication and authorization: expired tokens, wrong scopes and a user reaching another customer's object.
- Idempotency and retries: the same request sent twice must not create two payments or two orders.
- Error mapping: partner errors translated into useful messages and correct retry decisions.
- Rate limits and timeouts: behaviour at the limit, during a slow response and after a dropped connection.
- Data consistency: the same order, stock or payment state across every system after a batch.
- Peak load: the expected busy-hour volume, not only a single request.
- Recovery: replay, reconciliation and rollback when a batch fails midway.
Red flags
- A promise of full coverage without reading your integration contracts.
- No negative tests, no contract tests and no plan for duplicate requests.
- Testing only in production because a sandbox "does not exist".
- Test evidence that cannot be reproduced: screenshots without request IDs or logs.
- Unclear data retention or a plan to copy production personal data onto laptops.
- No handover, so the test suite leaves with the vendor.
- A price that excludes environment setup, test data or re-testing after fixes.
Pilot and acceptance
Buy a short pilot on one integration path, not the whole estate. A useful pilot runs for two to three weeks, covers one critical flow and includes a fixed set of acceptance tests agreed in advance. The vendor should provide a test plan, the executed results, a defect log and a short handover session.
Acceptance should be binary. A test either passed with evidence or it did not. Track defects by severity and require a re-test after each fix. If the pilot cannot reach the partner's sandbox, record that limitation and decide whether the residual risk is acceptable before signing a larger contract.
Contract and handover terms
Own the test assets. The contract should state that test cases, scripts, fixtures and reports belong to you, and that the vendor will transfer them in a usable repository at the end. Include a warranty period for defect fixes, a defined response time and an exit clause that covers knowledge transfer. Avoid a renewal that locks you into the same vendor to keep the tests running.
Decision framework
- Choose an independent QA firm when you need a defensible test opinion across several systems.
- Choose an integration specialist when one partner or ERP connection is the main risk.
- Choose staff augmentation when you have a test strategy and need extra hands inside the team.
- Choose tooling plus internal ownership when the engineering team can maintain automation after the pilot.
- Choose a managed platform when ongoing contract monitoring matters more than bespoke investigation.
Integration testing is part of Custom Software & Integrations, alongside API design and delivery. The pricing page explains how a fixed pilot differs from ongoing test support. For the failure patterns this work is meant to catch, read API integration testing problems and fixes.
Sources
- OWASP API Security Top 10, 2023 edition, accessed 17 November 2026.
- OWASP Application Security Verification Standard, accessed 17 November 2026.
- RFC 9110, HTTP Semantics, accessed 17 November 2026.
- RFC 9457, Problem Details for HTTP APIs, accessed 17 November 2026.
- NIST SP 800-115, Technical Guide to Information Security Testing and Assessment, accessed 17 November 2026.
Responses (0)
No comments yet. Add the first one below.
Leave a response
Your email address will not be published. Required fields are marked *.