Quality assurance is easy to buy badly. A vendor can supply testers, a wall of automated checks and a monthly report, and still leave the business unable to say whether a release is safe. The useful purchase is not capacity but evidence: a documented risk model, acceptance criteria that can be tested, a feedback loop the team trusts and a named person who can stop a release.
This guide is for a business buying QA support for its own product, or for an agency that needs a partner to cover testing. It covers the decision, the brief, provider types, a scorecard, pricing questions, a pilot and the exit terms that prevent dependency.
Decide what you are buying
Four decisions look similar and require different proposals. You may be buying release confidence for an existing product, stability for a legacy system, QA built into a new product, or evidence for an audit or enterprise customer. Write the decision in one sentence and name the consequence of getting it wrong. Name the systems in scope, the release cadence and the people who will make the acceptance decision. If the business cannot describe what a failed release costs, the first task is to measure that, not to hire more testers.
Requirements for the brief
- Risk-based scope: which journeys, data and integrations carry the most business risk, and which are explicitly out of scope.
- Acceptance criteria: the format you expect, with examples of normal, failure and permission cases.
- Environments and test data: who owns them, how data is refreshed and how personal data is protected.
- Automation: the layers you want, who maintains the suite and how flaky tests are handled.
- Non-functional testing: performance, security and accessibility boundaries, and who performs them.
- Defect management: severity definitions, triage cadence and the evidence attached to a defect.
- Release authority: the named role that can block a release, and the escalation path when deadlines conflict.
- Reporting: the measures that matter, such as escaped defects and feedback time, not test counts.
- Handover: documentation, training, repository access and the transition period.
Provider categories
Staff augmentation adds testers under your direction. It is flexible and fast, but you still own the strategy, the prioritisation and the release decision.
A managed QA service supplies a team and a process. Coverage improves, but distance from product decisions can produce test plans that miss what customers actually do.
A test automation specialist builds and maintains suites. This is valuable when automation is the bottleneck, but the specialist can automate the wrong things very efficiently.
A QA consultancy diagnoses the problem and writes a roadmap. It is useful when the strategy is unclear, but the implementation gap must be planned before the engagement ends.
Embedded QA places quality inside the product team. It produces the closest feedback loop and depends most on the team's maturity.
A scorecard
| Criterion | Evidence to request |
|---|---|
| Risk model | A sample showing business impact, not only technical severity |
| Acceptance criteria | Three real examples, including failure and permission paths |
| Automation | Repository structure, CI integration and flaky-test policy |
| Environments | Ownership, refresh process and data privacy controls |
| Release authority | A named role and a worked escalation example |
| Reporting | Escaped defects and feedback time, with definitions |
| Handover | Documentation, training and access transfer |
| Commercials | Billing unit, change control and exit assistance |
Pricing questions that change the total
Ask what the billing unit is and what happens when scope changes. Confirm whether re-tests are included, who pays for environments and tool licences, and whether there is a minimum term. Ask for a worked example using your release calendar rather than a generic package. A low day rate with excluded environments, reporting and handover is often more expensive than a higher rate that includes them.
Pilot design
Run one release train for four to six weeks. Baseline escaped defects, test cycle time, feedback time and the number of releases blocked for a good reason. Set acceptance criteria before the pilot starts. A pilot that produces only a report is a demonstration; a pilot that changes how the team releases is evidence. Compare the baseline with the pilot on the same measures. If feedback time and escaped defects do not improve, the model is not working, even if the reports look professional.
Red flags
A promise of complete automation, test counts as the main metric, no access to your repositories or CI, no named release authority, testers who never speak to developers, and no plan for test data. Also treat a contract that owns the test assets as a warning: you should be able to continue without the vendor.
Decision framework
Choose staff augmentation when you have a strong QA lead and need capacity. Choose a managed service when you need coverage and a process. Choose a specialist when automation is the bottleneck. Choose a consultancy when the problem is unclear. Choose embedded QA when quality belongs to the product team and the team is ready to own it.
Contract and exit
Own the test code, test data and documentation. Require CI access and a transition period with overlap. Define what happens to environments, tools and credentials at exit. A QA partner that cannot hand over is a dependency, not a service.
We build and review delivery teams under Custom Software & Integrations. The pricing page explains how discovery, delivery and support are separated. For the failure patterns this purchase is meant to correct, read software QA strategy problems and fixes.
Sources
- International Organization for Standardization, ISO/IEC 25010:2023, Systems and software quality models, accessed 17 November 2026.
- ISTQB, International Software Testing Qualifications Board, accessed 17 November 2026.
- Martin Fowler, The Practical Test Pyramid, accessed 17 November 2026.
- OWASP, Application Security Verification Standard, 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 *.