Quality engineering / 7 min read
QA automation should optimize confidence, not test count
A risk-led approach to deciding what to automate, where tests belong, and how to keep suites useful as products change.
Published 24 August 2026
A large automated suite can still provide weak release confidence. Tests may repeat the same low-risk checks, miss integrations, depend on unstable data, or fail so often that teams stop trusting the result.
The goal of automation is dependable information at the moment a decision must be made. Coverage should follow product risk and tests should live at the lowest level that can prove the behaviour effectively.
Build coverage from risk and journeys
Identify business-critical workflows, irreversible actions, security boundaries, complex rules, integrations, and historically fragile areas. Use these to define what the release process must prove.
Use the right level for each question
Fast component and API checks should cover most deterministic behaviour. Reserve browser and end-to-end tests for complete journeys, integration boundaries, and platform behaviour that cannot be proven more cheaply.
Engineer the test system
Test data, environments, service dependencies, observability, ownership, and failure diagnosis determine whether automation remains useful. A flaky test is an operational defect in the delivery system.
- Deterministic data creation and cleanup
- Clear failure evidence and logs
- Named ownership for unstable checks
- Regular removal of duplicated or obsolete scenarios
Connect tests to release decisions
Quality gates should express risk, not ceremony. Define which failures block release, which trigger investigation, and what evidence is required for an exception. In MTD’s pharmaceutical SaaS work, automated delivery and end-to-end journeys created a repeatable path for frequent change.
Practical takeaways
Prioritise automation by business and technical risk.
Prove behaviour at the lowest effective test level.
Treat test data and environments as product infrastructure.
Make quality signals directly useful to release decisions.
Continue exploring