Through AI-assisted quality engineering we reduced the volume of manual regression test cases by roughly 3× and cut the average time per test case by 2×. The capacity freed up is now directed at testing new product features rather than maintaining the same regression baseline.
— QA Lead
The client is a cybersecurity company offering a unified protection platform managed from a single console. They came to a1qa to free up QE capacity for new features — meeting the business’s growth plans without letting a widening regression scope slow releases.
Because one product console aggregates many distinct functional areas, the regression scope was wide and growing. The QA team needed to decrease the cost of maintaining that scope without reducing the depth of regression test coverage or blocking capacity for testing new features being delivered at an increasing pace.
Delivery followed a phased model that ran from a small proof on one slice to full rollout, funding each phase only on confirmed results from the previous one.
The client used an Agile delivery model, with QA integrated into product development cycles. a1qa’s quality engineering team of five worked alongside the client’s specialists: two QE automation engineers, two AI engineers, and one QA lead coordinating each initiative from the Pilot to broader application after each phase produced a measurable cost reduction.
Manual test authoring was replaced by a structured workflow using AI agents. One agent reads requirements from the project knowledge base and converts them into test scenarios. A second agent uses Playwright via MCP to open the product in a browser, log in, and map the actual UI elements and locators — the same way a human tester would navigate the interface. A third agent combines both inputs, checks for existing tests, applies framework rules to avoid duplicates, and generates test code including page objects and locators. The resulting tests run from the point of generation with only stabilization and targeted debugging remaining for the engineer.
A pre-defined framework structures the project by domain, with acceptance rules written so that both AI models and human engineers can read them. When a new test is added, the model reads the framework rules, identifies the correct domain and application area, finds analogous existing tests in the registry, and adds the required fixtures, models, and supporting code. The engineer reviews and debugs the specific function rather than writing the full test from scratch.
Previously, when a test failed, an engineer manually exported logs, read them line by line, unpacked product-level logs separately, determined whether the failure was an automation defect or a product bug, and then filed a Jira ticket. Each failure required individual handling and a non-trivial time investment. With the new workflow, the failure record and logs are passed to an AI model, which collects full context, identifies the root cause, classifies the failure as a product bug or automation issue, and produces a structured report with a complete log trace — without engineer involvement in the triage step.
By combining detection of gaps in test coverage with risk-based selection, the team identified which manual regression test cases were genuinely necessary for each cycle and which could be automated or removed from the scope. This reduced the number of manual test cases requiring execution each regression testing cycle from approximately 185 at the point of AI adoption to 78 by the final measured period — a reduction of more than 3×.
The team restructured regression planning using data signals from the initiatives above, aligning scope to actual product change risk rather than maintaining a fixed legacy case list.
