QA Outsourcing Often Buys Hours, Not Quality
QA outsourcing promises better software at a lower cost. Many contracts deliver something narrower: a number of testers, billed by the month.
That model made sense when testing meant people working through scripts by hand. Once AI can run routine tests, paying per tester stops making sense: the vendor earns more when the work stays manual.
The headcount model has not gone away. In Deloitte’s 2024 Global Outsourcing Survey of more than 500 business and technology leaders, 29% were still using staff augmentation in their outsourcing: renting people rather than buying a result.
Whether QA outsourcing works depends less on the day rate than on two questions. Who designs the tests? And who owns them when the contract ends?
Five Questions to Ask Any QA Vendor
These five questions separate a vendor selling hours from one selling outcomes. Ask them before price comes up. The rest of this piece explains why each one matters.
- Who owns the test cases, scripts and tooling when the contract ends? The answer should be the buyer, in writing, with the assets stored in the buyer’s environment throughout.
- How do you measure coverage? Look for coverage against acceptance criteria or requirements. A count of test cases says how busy the vendor was, not how safe the release is.
- Is your price tied to headcount or to outcomes? A headcount price is not wrong in itself, but it should come with a plan to automate and a reason for the vendor to follow it.
- What is your defect leakage rate on current clients? Defect leakage is the share of defects that escape to production. A vendor that does not track it cannot improve it.
- Where does AI sit in your process, and who checks its output? AI output still needs a human check, so the answer should name the people who review what AI generates, not only the tools.
Automation Is Still the Exception
Survey data shows that most teams have not yet realized the full potential of AI in testing. The World Quality Report 2025-26, an annual survey of 2,000 executives in 23 countries by Capgemini and Sogeti, puts numbers on the gap.
On average, 33% of test cases are automated. Only 8% of organizations have a fully established enterprise test automation strategy. And 59% say their quality engineering is not automated enough to keep up with agile delivery.
Generative AI has not closed the gap on its own. Teams using it report an average productivity gain of 19%, and a third saw very little improvement. Meanwhile, 60% worry about hallucinations and the reliability of what AI produces.
For a buyer, this cuts two ways. AI has not yet made testing cheap by default, so a vendor’s AI claims need checking. And a vendor paid by the hour has little reason to push that 33% higher.
Outsourcing Fails When Knowledge Leaves With the Vendor
Outsourcing decisions have moved beyond price. In Deloitte’s 2020 survey, 70% of organizations named cost savings as the main reason to outsource; by 2024 that had fallen to 34%. What enterprises need from a QA partner now is direction on what, when, why and how to test their products and platforms. The risk is that this knowledge ends up outside the company.
It happens gradually. Test cases live in the vendor’s tools and in its testers’ heads. Coverage gets reported as a count of tests run, not as a share of requirements checked. The World Quality Report found that 37% of organizations still rely mainly on basic indicators such as defect counts and automation coverage, and only 25% link quality metrics to business outcomes.
The decision to stop a release then falls between two teams. The vendor reports what it ran; the product team assumes that means the release is safe.
Buyers notice eventually. In Deloitte’s survey, 70% of organizations had brought part of their previously outsourced work back in-house in the past five years. Better control over service quality was the most common reason, cited by 68%.
The Cost Shows Up at Release Time and at Exit
Knowledge held by a vendor costs twice: while the contract runs, and when it ends.
While it runs, the cost is speed. Releases wait in the testing queue because only the vendor knows which tests matter for a given change. Defects that the test counts never measured reach customers, and each one is found later and fixed at a higher price.
When it ends, you pay again to rebuild knowledge you already paid for. A new vendor, or an in-house team, has to recreate test cases that only the old vendor understood. In Deloitte’s survey, 46% of organizations named poor vendor performance during transitions as a challenge in their outsourcing programs.
Neither cost appears on the invoice. Both show up in delivery.
Outsource the Running, Keep the Design
Our recommendation: enterprises must own their testing strategy, even as they outsource day-to-day testing to vendors.
Two models work in practice. In the first, the vendor runs testing end to end, measured against the company’s own acceptance criteria rather than a count of tests. In the second, the vendor builds the testing capability, runs it until it works, then hands it over. In both, the test cases, scripts and tooling sit in the company’s environment from day one.
We have seen both work. A fast-scaling, cloud-based point-of-sale (POS) platform moved to AI-first, in-sprint testing tied to each release. Defect leakage fell from 20% to less than 2%, and the testing team was reduced from 30 to 6 as AI absorbed routine test execution (case study). An international airline took the second route: we built its AI testing capability with a defined path for transferring the intellectual property, so the airline owns the system outright (case study).
The common thread is ownership. When the test design belongs to the buyer, the vendor is paid to run it well, not to keep it.
Skip These Three Steps and You Will Fall Behind
- Write down acceptance criteria for your next three releases. They become the yardstick any vendor, or any AI tool, is measured against. We explain why test design has to stay governed in From AI-Generated Tests to AI-Governed Test Design.
- Measure defect leakage before you buy. Without a baseline, no vendor’s claim can be checked, including ours.
- Put ownership and exit terms into the contract. Name the assets, where they live and how they transfer, before work starts.
If you want to see how AI-first testing looks on a product like yours, our quality engineering team runs a 20-minute walkthrough.
Buy quality outcomes, not man-hours.