Skip to content
App development

Testing a Web Application: Why Automated Tests Pay Off

A small form change breaks ordering, but nobody repeats that journey before release. Automated tests help catch these regressions. They are especially useful when an application changes frequently and the same important rules need repeated checking. The value comes from choosing meaningful scenarios and trustworthy results, rather than accumulating a large test count.

Automated tests: important risks, verified outcomes and ownership of failed runs.

A useful test suite gives the team specific evidence about what still works after a change. It does not prove that a product is defect-free. A business owner should understand the failure each test addresses, when it runs and who investigates a failure. This guide explains how to commission that protection and extend it sensibly.

01Start with the failure that matters

List the tasks whose failure would interrupt work or create an incorrect transaction: signing in, saving a quote, calculating a price, submitting an order and accessing documents. Record the impact, usage frequency and frequency of changes to each area. A critical journey that changes regularly is a different candidate from a one-off information page.

SituationWhat the test must establishUseful first check
Price calculationCorrect result at a discount boundaryRule test with specified inputs
Access to another client’s jobThe server refuses the requestAn authorisation integration test
Order submissionExactly one correct order is createdAPI checks and the main user journey
Partner service failureWork already entered is preservedA controlled failing response

Do not prioritise only what is easiest to automate. A homepage check may be simple to write while offering little protection for ordering. Start with a limited suite tied to business outcomes, then expand it using evidence from changes and actual incidents.

02Specify the outcome, not just the clicks

A scenario needs starting data, a user role, an action and an expected final state. “Click Submit” is not a sufficient order test. Check the stored items, price, ownership and confirmation. Consider repeated submission and unsuccessful responses separately. A success message alone does not establish that the database contains the correct order.

Automation cannot establish whether customers understand the offer. Observation and prototype validation address that question. Technical tests protect agreed behaviour; user research helps decide what behaviour the product should offer.

03Combine test layers around the risk

LayerWhat it checksWhat it cannot establish alone
Unit testsIndividual rules and calculationsDatabase integration or the real interface
Integration testsComponents working together, data and permissionsThe complete task in a browser
End-to-end testsSelected journeys through the normal interfaceEvery combination or user need
Manual and exploratory testsUnfamiliar situations, clarity and unexpected problemsCheap repetition of a large fixed checklist

There is little value in exercising every calculation through a complete browser journey. Check discount rules directly with varied inputs, database interactions through integration tests, and a few important journeys through the real interface. Avoid a universal percentage for each layer. The useful balance depends on architecture, defects and maintenance effort.

Hiding a menu link does not enforce authorisation. Test whether the server rejects unauthorised requests. See our guides to API security and authentication and user management. Being signed in and being allowed to read a particular document are separate rules.

04Control the environment and test data

Each test should begin from a known state. Use separate test accounts and data, remove records created by tests and minimise dependence on execution order. Parallel checks must not compete over the same order. Replace real payments and emails with safe test modes unless they are explicitly part of a controlled production check.

Playwright’s testing guidance emphasises isolation, observable behaviour and resilient ways to identify controls. For example, locate a button by its role and name instead of its incidental position in the layout. A styling change should not unnecessarily break a test about submitting an order.

A simulated partner response makes errors and delays reproducible. It does not prove compatibility with the actual partner. Check that separately in the provider’s supported test environment and monitor interface changes. A test report should make clear which part of the journey was simulated.

05Make tests part of the release decision

Testing workflow: select a risk, define the expected outcome, run the check and investigate the finding.
A workflow for repeatable checks. A passing run supports only the scenarios and environment actually tested.

Run quick checks when code changes and a broader suite before release or on an agreed schedule. Associate results with the precise application version. If checks ran against a different build, or configuration subsequently changed, show that limitation. Decide how critical failures affect a release before the team is under pressure to ignore them.

Tests complement deployment and rollback planning. After release, perform a safe check of an important task in the target environment too. Our blue-green and canary deployment guide covers traffic management; the key here is knowing what evidence supports the next step.

06Treat an unreliable result as work to investigate

A test fails once and passes on the next run. The cause could be an application defect, shared data, an unready environment or unsuitable waiting logic. Do not automatically blame either the test or the product. Preserve the version, inputs, error and appropriate execution evidence. Limit sensitive data in logs and screenshots, including who can access them and how long they remain.

Playwright distinguishes passed, flaky and failed tests: a test that succeeds only on retry is classified as flaky. Retrying can aid investigation, but should not erase the initial failure from reporting. A temporarily quarantined test needs an owner, a repair deadline and alternative checking appropriate to the risk.

07Compare effort including setup and maintenance

Consider an illustrative model of repeating the same existing set of checks. Manual execution takes three hours per run. Automation requires twenty additional setup hours and an average of one hour of human work per run, including both review and maintenance. The model does not assume that running automated tests is free.

Model inputManual checkingAutomated approach
Additional setup0 h20 h
Human work per run3 h1 h
Total after 5 runs15 h25 h
Total after 10 runs30 h30 h
Total after 20 runs60 h40 h
Illustrative model for the same checking scope, not client results. Manual scenarios already exist. Each automated run includes half an hour of review and half an hour of average maintenance.
Model: cumulative human effort (hours)
  • After 1 runManual approach: 3Automated approach: 21
  • After 5 runsManual approach: 15Automated approach: 25
  • After 10 runsManual approach: 30Automated approach: 30
  • After 20 runsManual approach: 60Automated approach: 40

Model: manual effort = 3 × runs; automated effort = 20 + runs. Excludes licences, computing resources and differences in hourly rates. This is not a financial return calculation.

In this model, effort breaks even at the tenth run: 3 × 10 = 20 + 10. After twenty runs the difference is twenty hours. A financial comparison must also price the different roles, licences and infrastructure. Freed capacity does not automatically reduce a bill. If the feature is soon retired or the test needs frequent rewriting, the result changes.

08Keep human judgement in the process

Automation is good at repeating questions already formulated. Exploratory testing also looks for questions the team has not considered. Investigate unusual combinations, ambiguous wording, interrupted work and actual use on target devices. High code coverage shows that tests executed particular code; it does not establish that their expectations are correct or their scenarios complete.

W3C explains that tools cannot automatically assess every aspect of accessibility. Combine automated checks with expert and user evaluation. Similarly, a routine regression suite does not replace an application security assessment.

09Commission a pilot with a useful handover

Choose a few important scenarios that genuinely recur. Ask for a scope description, test data, execution in the agreed environment, a demonstration of a detected fault and instructions for interpreting results. Record the areas the suite does not check. Handover should establish how the tests will be used, rather than end with a green report screenshot.

Tell us the application’s main task, how often you release changes and an example of a recurring defect. Listify can propose an initial suite, how it fits into development and who maintains it. We can also assess applications built by another supplier before recommending an appropriate scope.

Discuss application testing

10Monitor the suite’s value and running cost

Review critical journey coverage, feedback time, maintenance effort, recurring false alarms and defects that reached production. Test count and the percentage of green runs are not the objective. For a serious escaped defect, ask whether the missing piece was a scenario, an assertion, the environment or a response to the result. Include test maintenance in agreed ongoing website care.

11Frequently asked questions

Are automated tests worthwhile for a small application?

Yes, when they repeatedly protect an important function that changes. Start narrowly. A largely static site may benefit more from a simple check of its main form than from a large testing project.

Will automation replace a tester?

It takes over some repeated checks. People still need to design scenarios, investigate new behaviour, diagnose failures and maintain data and tests. Accountability for quality remains.

Does everything need a browser test?

No. Rules and calculations are often better checked directly, and component interactions through integration tests. Reserve browser tests for selected important journeys. Choose the combination around the application’s risks.

Does 100% coverage guarantee a defect-free product?

No. A coverage metric may describe executed lines or branches rather than correct expectations, all possible inputs or usability. Understand precisely what it measures and which business scenarios remain unchecked.

What does Listify need for an initial proposal?

A description of the application, important roles and tasks, release frequency and recurring problems. Tell us whether a test environment and existing tests are available. Access can then be arranged through an agreed secure process.

LISTIFY teamWebsites, apps and marketing from Prague since 2008

More articles

All articles →

Share this page

By email

Got an idea?

On a short call, we'll find out what you need and suggest the next step. Then you'll get a proposal with a fixed price and a timeline.

+420 771 166 199Mon to Fri, 8:30 a.m. to 4:00 p.m. (Prague time) · info@listify.cool

When should we call you?

Pick a day and a time window. We'll call you, and it takes about 15 minutes.

Day