Manual vs automated testing: which tests still need a human?

Automated testing saves time, especially for predictable and repetitive flows. But full automation is neither possible nor cost-effective. Human judgement is still needed for tests that involve user experience, business logic and acceptance testing.

Automation has made testing in many system projects both faster and more predictable, particularly when it comes to regression testing during system maintenance. The goal of automated testing is often described as greater speed, but speed is only part of the benefit. The real value lies in confidence: knowing that the whole business keeps running as usual after every change to the code.

A well-targeted test strategy is therefore not measured by the share of test cases that are automated, but by the ability to match the right kind of test to the right situation. In large system implementations, getting this balance right is essential. If you automate in the wrong places, you risk more than a red figure in a test report. You risk real disruption to your day-to-day business.

So the question is not whether you should automate, but what you should automate, and how to avoid building up costly maintenance debt.

Verification vs. validation

The difference between what can and cannot be automated often comes down to two concepts:

• Verification (“Did we build the system right?”): Covers technical checks and whether the code does what it should. Automated regression tests excel here.
• Validation (“Did we build the right system?”): Covers whether the system actually supports day-to-day work. Only the business can answer this.

Six types of tests that require a human

Business-critical processes

A script can confirm that each individual step produces the correct technical result, but not that the whole process works as the business requires from start to finish. By all means automate the recurring sub-steps, but let people with domain knowledge confirm that the whole process holds together. An error in a business-critical process is felt immediately across the business.

Flows with complex knock-on effects

When a change affects data, calculations and reports across several integrated systems, automated scripts become extremely difficult to build and even harder to keep up to date. In these cases, it is safer for someone who knows the process to test the flow and judge whether the overall result is correct.

User interfaces (UI/UX)

Interface tests are difficult to automate. Traditional test scripts break easily whenever the layout changes even slightly. Newer AI-based tools can adapt to some visual changes, but neither conventional scripts nor AI can judge whether an interface feels intuitive, looks visually appealing or is easy to use.

Tone, language and overall impression

Some tests have no answer that can be written as pass or fail. This applies, for example, to whether a text has the right tone, whether a translation sounds natural or whether a layout looks polished. People spot these flaws in seconds, whereas a script attempting the same judgement takes longer to build than it saves.

Features under active development

A feature that is still being developed changes often, sometimes daily. This means the automated test constantly has to be rewritten, so the work becomes more maintenance than actual testing. Test new features manually and automate only once the feature has settled.

User acceptance testing (UAT)

The final check before go-live takes place during user acceptance testing (UAT), when end users confirm that the system supports their day-to-day work. Even if a flow is technically flawless, it can be unusable in practice if it does not match how the business actually works. Only the business can make the decision to approve the system.

How to decide which tests to automate

A test is suitable for automation if it runs frequently, uses predictable data and covers a stable feature.

The best candidates for automation:

• Regression tests: Check that existing features still work after a change.
• Smoke tests: Quickly confirm that the most important features work after a new release.
• Integration tests: Check that data is transferred correctly between systems.
• Data-driven tests: Run the same steps with large volumes of varied data.
• Cross-environment tests: Ensure that the same flow works across different browsers and devices.
• Performance and load tests: Simulate high loads that cannot be tested manually.

Checklist: what to consider before you automate

• Is the feature stable? Never automate functionality that still changes daily. Wait until the flow has settled.
• What is the real cost? Factor in the time it takes to maintain the scripts over time, not just the time it takes to build them in the first place.
• How often will the test run? A test that only runs once a year is rarely worth automating. Focus on tests that run often, for example with every new release.
• Are the inputs and results predictable? Automated tests require exact, binary answers (yes/no, right/wrong). If the test requires judgement, it should stay manual.
• Do you have the right skills in the team? Building and maintaining automated tests requires development skills. Make sure there are resources to maintain the scripts so they do not turn into “test debt”.
• Do you have a complete picture? Make sure both manual and automated test results are gathered in one place so you keep an overall view of quality.

Read more

Which requirements and test management tool suits your organisation, and how do Reqtest, Jira, Azure DevOps, TestRail and Excel differ?

Read our white paper on the key factors for a successful ERP system and what determines whether an ERP implementation succeeds.

• How to calculate whether a requirements and test management tool pays off in your system implementation.

FAQ

Can all testing be automated?

No, only tests that can be written as checks with an expected, binary answer. Tests that require judgement, creativity or business knowledge require a human.

Is manual testing on its way out?

No, manual testing remains in almost every organisation, but the tasks are shifting. Automation takes over the repetitive checks so that testers can spend more time on what requires human judgement.

How does AI affect which tests should be automated?

AI shifts the line on what is worth automating by making scripts easier to create and reducing maintenance when interfaces change. But AI still cannot judge whether a flow actually works in the business’s day-to-day operations or whether the system is ready to go live.

How do you keep track of both manual and automated tests in one place?

The key is to bring all requirements, manual test cases and automated test results together in a dedicated test management tool. When you link every test to the same requirements, you get full traceability and a clear overview of quality, so you can make confident go-live decisions.