Manual test case writing brings something valuable to every project: a QA engineer's full context and judgment on each individual case. AI-assisted generation brings something equally valuable: speed, producing a complete first draft in seconds instead of hours. Both approaches have real strengths, and most QA agencies end up using some blend of the two. What's less common is actually measuring that blend, so we wanted to know exactly how much time and cost each approach takes for the same scope of work, and where the line between them should sit.
We measured it on a real e-commerce checkout flow project (product selection, cart, payment, order confirmation) and wrote up the complete workflow in How QA Agencies Use AI Test Generation to Deliver Faster. Here's the short version: just the arithmetic.
Where the 30-34 hours go with manual testing
A checkout flow redesign is a well-scoped project. Here's where a QA engineer's time typically goes when writing test cases by hand for a project like this:
- Reviewing requirements and acceptance criteria: 2 hours
- Writing 50-100 test cases across happy paths, negative flows, edge cases, and boundary conditions: 16-20 hours
- Reviewing and refining with the product owner: 4 hours
- Executing tests and logging defects: 8 hours
That's 30-34 hours total, with 16-20 of them dedicated to writing test cases before execution even begins. Those hours buy something real: test cases shaped by product knowledge and judgment about what's likely to break.
The napkin math
At a fully-loaded QA engineer cost of $25/hour (a typical blended rate for a team based in Eastern Europe, for example), writing 50-100 test cases by hand costs $750-850 in labor. That's the investment in a thorough, well-considered first draft of your test suite.
An AI-assisted pass on the same scope works differently: about 30 minutes describing the feature, generation in seconds, then two hours reviewing and refining the output. Labor cost: roughly $150.
| Manual | AI-assisted | |
|---|---|---|
| Time | 16-20 hours | ~2.5 hours |
| Cost | $750-850 | ~$150 |
Same scope. Same 50-100 test cases covering the same flows. An 18-hour, roughly $1,350 difference: time you can reallocate to execution, defect analysis, and deeper exploratory work.
That difference compounds well. If your agency runs five similar projects a year, that's roughly 90 hours and $6,750 in labor freed up for the work clients notice most: deeper exploratory testing, faster turnaround, or taking on an additional project without hiring anyone. It's a standing margin improvement on every project of similar shape going forward.
What each approach is best at
Manual test writing and AI generation aren't competing for the same job. Each is strongest in a different place.
Where AI-assisted generation excels:
- Functional testing of well-defined features and forms
- Regression testing after product updates
- API and backend testing with consistent, rule-based scenarios
- Happy paths, negative scenarios, and standard boundary conditions
Where manual QA judgment adds the most value:
- Exploratory testing: finding the cases nobody thought to write down
- UX and usability judgment calls
- Complex workflows spanning multiple systems
- Security, compliance, and accessibility testing
The strongest setup combines both. AI-assisted generation covers the structured, high-volume ground quickly, and the hours it frees up go straight into the exploratory and judgment-based testing that benefits most from a human. Teams getting the biggest margin benefit tend to be the ones who moved their QA engineers from writing to reviewing, then reinvested the extra hours into the testing that genuinely needs a person doing the thinking.
What clients actually respond to
Clients care about outcomes more than methodology. Whatever mix of manual and AI-assisted testing produces the result, what matters to a client is turnaround time, quality, and cost.
In practice, that means leading with the outcome works well: "We deliver test results in 48 hours instead of five business days, at the same quality." It keeps the conversation focused on what the client actually came for, regardless of which methods produced the result.
Try it in a week, not a quarter
You don't need to overhaul your process to see how this compares for your own team. Pick one feature you'd test this week anyway. Write the acceptance criteria the way you normally would. Generate half the test cases with AI and write the other half by hand, then have someone senior review both sets without knowing which is which. Compare time spent and defects found.
That's the whole pilot: a controlled comparison on real work you were already planning to do. It gives you your own numbers, on your own codebase, with your own team's actual review standards, which matters more than any published benchmark including this one.
Takeaways
- Manual test case writing and AI-assisted generation each bring something different: judgment and context versus speed and volume.
- Combining both, using AI for structured coverage and reinvesting the saved hours into exploratory and judgment-based testing, is where the real margin benefit shows up.
- Clients respond to outcomes such as faster turnaround and consistent quality, whatever mix of methods produces them.
We go deeper on this in How QA Agencies Use AI Test Generation to Deliver Faster: the full breakdown of where the hybrid model pays off, how to introduce it to your own team, and a real 3-week pilot plan for testing it on your next client project.
