Software QA Process: A Weekly Release Checklist for Teams
A repeatable software QA process lets teams ship every week without dreading release day, from agreed acceptance criteria and production shaped test data to a named release owner and a rollback plan decided in advance.
The Short Answer
Shipping every week changes what quality means. When releases go out once a quarter, a long testing phase at the end feels manageable. When they go out every Friday, that phase becomes a bottleneck, and, sooner or later, a step the team quietly skips. A working software QA process solves this by spreading quality checks…
Shipping every week changes what quality means. When releases go out once a quarter, a long testing phase at the end feels manageable. When they go out every Friday, that phase becomes a bottleneck, and, sooner or later, a step the team quietly skips. A working software QA process solves this by spreading quality checks across the whole week, so release day is a confirmation, not a gamble.
This guide describes that process in practical terms: what gets checked before every release, who owns each check, and how standards hold when artificial intelligence writes a growing share of the code.
Why a software QA process beats a testing phase
A testing phase treats quality as something you inspect at the end, like checking a finished product on a factory line. The problem is that software defects are cheapest to fix where they are created. A misunderstood requirement caught in a review meeting costs minutes. The same misunderstanding found by a customer after release costs a hotfix, a support queue, and trust.
A process, by contrast, puts small checks at every stage. Acceptance criteria are agreed before coding starts. Test data is prepared alongside the feature, not after it. A named person owns the release checklist. A rollback decision is made in advance, while everyone is calm, rather than during an incident, when nobody is.
There is a second reason this matters now. AI coding tools have made code cheap and review expensive. Pull requests that once held fifty lines now hold five hundred. Sonar's 2026 State of Code survey of more than 1,100 developers found that 96 percent do not fully trust AI generated code, yet only 48 percent always check it before committing.
The software QA process in four gates
Think of the weekly cycle as four gates. A change passes through each one, and each gate has a named owner and a clear pass or fail outcome. If a gate fails, the release waits.
Gate 1: Agree what done means, before coding starts
Every ticket needs acceptance criteria written in plain language before a developer opens an editor. Good criteria describe observable behaviour: "A returning customer sees their saved address prefilled at checkout, and changing it updates the order before payment is taken." Vague criteria, such as "improve checkout", cannot be tested, so they cannot be passed.
The reviewer for this gate is usually the product owner or the ticket author. Their job is to confirm that the criteria cover the happy path, at least one failure case, and any edge case the business cares about, such as expired cards or out of stock items.
Gate 2: Build test data that matches production shapes
Most bugs found in production were never visible in testing because the test data was too tidy. Real users have long names, empty address fields, expired discount codes, and carts with forty items. Test data should mirror those shapes, including the awkward ones.
A practical rule is to seed staging with anonymised, production shaped data: real category structures, real price ranges, real variation in user records. Then add the specific cases the new feature touches. If a change affects checkout, the test set needs a guest checkout, a logged in checkout, a discount code checkout, and a failed payment. A structured app testing checklist is a useful companion here, because it forces the team to cover cases it would otherwise forget.
Gate 3: Review behaviour, not just code
Code review should answer one question above all: does this change do what the acceptance criteria say, in the situations real users will create? Reviewers should run the feature, not only read the diff. For AI generated code this matters even more. Research from CodeRabbit, based on 470 open source pull requests in December 2025, found AI coauthored changes carried about 1.7 times more issues than human written ones. Stack Overflow's 2025 Developer Survey found trust in AI tool accuracy had fallen to 33 percent, down from 43 percent the year before.
The practical response is not to ban AI assistance. It is to review the behaviour: run the feature with realistic data, check the failure paths, and confirm the tests actually test the new rule rather than merely existing.
Gate 4: The release checklist, with a named owner
On release day, one person owns a short written checklist. Not the team in general, one named person. The checklist confirms the build version, the migration plan, the feature flag states, the monitoring dashboards, and the rollback decision. It also records the sign off.
Crucially, the rollback decision is made before the deploy, not after something breaks. The owner writes down the trigger ("roll back if checkout errors rise above the agreed threshold within thirty minutes") and the method (a feature flag switch, a previous build, or a forward fix). Teams that deploy with a rehearsed rollback strategy, such as the blue green, rolling, and canary approaches compared in this guide to zero downtime deployment strategies, recover faster because the decision is already made.
A checkout release, walked through
Consider a concrete release: a mid sized online store adds a "buy now" button that skips the cart and takes the customer straight to payment. The change looks small. It touches pricing, discounts, inventory, tax, and payment, which is exactly why it is a good test of the process.
Before coding, the team agrees acceptance criteria. A buy now purchase must apply the same discount rules as a cart purchase, reserve stock at the same point, and show the same final price the customer will be charged. The criteria also cover failures: if payment fails, the stock reservation must release, and the customer must see a clear error, not a blank page.
Test data is prepared to match production shapes: a product with limited stock, a product with a percentage discount, a product with a fixed amount discount, and a customer with an expired saved card. During behaviour review, the reviewer runs the buy now flow with the expired card and discovers the bug: the failed payment leaves the stock reserved, so the item shows as out of stock for the next customer even though nobody bought it. The code looked clean. Only running the failure path found it.
The fix is small, releasing the reservation on payment failure, and a regression test is added for that exact case. On release day, the owner runs the checklist, confirms the feature flag default, and notes the rollback trigger in advance: if buy now errors rise above the threshold in the first hour, the flag switches off and the normal cart flow returns. The release ships, and no customer ever sees the bug. After this release, "payment failure releases stock" becomes a permanent item on the checkout checklist.
How the software QA process scales when AI writes most of the code
AI assistance changes the shape of quality work in three ways. First, volume rises: more code per developer per week means more behaviour to verify. Second, polished looking code makes reviewers skim. Third, failure modes shift toward plausible but wrong logic, such as a discount applied twice or a rule that works for the common case and breaks on an edge case.
The process adapts rather than expands. Acceptance criteria become more important, because they give reviewers something to check the output against. Behaviour review replaces line by line reading as the main defence. And the team keeps a rule that AI generated tests are reviewed with the same care as AI generated features, since a test written by the same tool that wrote the code can share the same misunderstanding.
The evidence supports this caution. A May 2026 survey of 273 software leaders by SmartBear found that 70 percent said application quality had already degraded as AI accelerated development, and 60 percent reported quality issues in the past year because code creation outpaced testing capacity. The 2025 DORA report similarly found that AI adoption improved delivery throughput while continuing to increase delivery instability. The lesson is to make verification keep pace with generation, which is what a repeatable process does. For teams that want outside help, professional software testing and QA services can supply the structure, test coverage, and release gates that a busy in house team struggles to maintain alone.
The weekly release QA checklist
Use this checklist every release week. Keep it short enough that the team actually completes it, and update it whenever a production bug reveals a missing item.
- Acceptance criteria agreed and written before coding started (Product owner)
- Test data prepared, matching production shapes and edge cases (Developer)
- Automated tests pass in CI, including a test for each new rule (Developer)
- Behaviour reviewed by running the feature, not only reading the diff (Reviewer)
- Critical user paths tested end to end on the release candidate build (QA lead)
- Data migrations checked, and the migration is reversible or additive (Developer)
- Feature flag defaults confirmed for the production environment (Release owner)
- Rollback trigger and method written down before deploy (Release owner)
- Monitoring dashboards and alerts verified on the release build (Release owner)
- Sign off recorded, and a named person is on call for the first hours after release (Product owner)
A checklist this short only works if every item has evidence behind it: a green CI run, a test result, a written rollback note. A tick without evidence is a guess, and guesses are how bugs reach users.
QA habits that survive a fast release cadence
Five habits separate teams that sustain weekly shipping from teams that burn out on it. First, keep releases small. A release that cannot be described in five sentences is too large to review well, so split it. Second, protect the review slot: put a fixed review time in the calendar on release day, so it never competes with other work. Third, give the reviewer real authority to hold a release. A review that cannot stop anything quickly becomes a formality.
Fourth, learn from every escape. When a bug reaches production, ask which checklist item would have caught it. If none would, add one. If one would, ask why it was missed, and fix the reason, not the person. Fifth, prune the checklist regularly. A checklist with eighty items gets skipped; a checklist with ten gets done.
These habits also shape how a team builds software in the first place. Products developed with quality gates from the start ship faster over their lifetime, because less time is lost to rework. That is one reason many growing companies pair their internal team with custom software and enterprise development partners who build testing and release discipline into the delivery process itself.
Frequently asked questions
What is a software QA process?
A software QA process is the repeatable set of checks a team runs before, during, and after development to make sure each release behaves as intended. It covers acceptance criteria, test data, behaviour review, release checklists, and post release monitoring, so quality is built into the week rather than inspected at the end.
How is a QA process different from a testing phase?
A testing phase happens once, at the end of a project, and checks finished work. A QA process runs continuously, with small checks at each stage, so defects are found where they are cheapest to fix.
How much testing should a weekly release include?
Enough to cover the critical user paths, the new rules introduced by the release, and the failure cases for anything that touches money, data, or accounts. Automated regression tests should run on every change, while manual effort focuses on the new behaviour and the riskiest paths.
Who should own the release checklist?
One named person per release, usually the release owner or QA lead, with the authority to hold the release if a check fails. Individual items can have different owners, but accountability for the whole list must sit with one person, or items quietly go unchecked.
How do you handle QA when AI writes most of the code?
Review behaviour, not diff length. Confirm the output against written acceptance criteria, run the feature with realistic data, check failure paths, and review AI generated tests as carefully as AI generated features. Keep humans accountable for sign off, and use feature flags and staged rollouts so a defect reaches few users for a short time.
What should a rollback plan include?
Three things, decided before deploy: the trigger that causes a rollback, such as an error rate threshold or a broken critical path; the method, such as a feature flag switch, a previous build, or a forward fix; and the person authorised to make the call.
Weekly shipping does not have to mean weekly firefighting. With agreed acceptance criteria, production shaped test data, behaviour review, a named release owner, and a rollback decision made in advance, a small team can release every week and sleep well on Friday night. Start with the checklist above, adapt it to your product, and let every escaped bug make it sharper. If your team wants help putting this release discipline in place, talk to the Skills Technology Solutions team about building a QA process around your product and release cadence.
Related Articles
Keep reading — more guides you may find useful.
Related Services
Services connected to this guide.
