The phase where the business confirms the system does its job — and the one most often reduced to a signature.
User acceptance testing is where the people who will use the system run their real processes against it, with real data, and confirm it works. It is not a bug hunt — defects should have been found in system testing. It answers a different question: can this business run on this configuration?
UAT is routinely compressed into the last two weeks and reduced to a sign-off. When that happens the business discovers during the first live month what it should have discovered in UAT, at far higher cost and with customers watching. The value is not in the testing hours; it is in the business committing that the configuration matches how it actually works.
"Process a credit customer's rush order with a partial despatch" — a real situation, not a sequence of button presses.
Testing against sample records proves nothing about whether your actual data works in the configuration.
Returns, credit notes, part-deliveries, cancellations. Happy paths rarely fail; exceptions do.
Defects separated from change requests. Conflating them is how scope grows during UAT.
Defined before UAT begins — for example, no open critical defects and every core scenario passed by a business owner.
A UAT cycle completes with 94 percent of scripts passed and sign-off given. Two weeks after go-live the business is struggling, and the obvious question is how testing missed it.
It did not miss it; it was not looking. The scripts covered the standard path of each process — a clean order, a standard receipt, a normal invoice. What broke in production were the variations: part deliveries, credit notes against partly paid invoices, orders with mixed tax treatment, the month-end sequence under time pressure. None of those were scripted, because scripts were written from process documentation rather than from what actually happens.
Two practices prevent this. Write scripts from real transactions pulled out of the legacy system, including the awkward ones, rather than from the documented happy path. And test the periodic processes — month-end, payroll, statutory filing — at least once in full, because they are the highest-stakes and least-rehearsed activities in the entire system. A UAT that never closed a period has not tested the thing most likely to go wrong.
| Term | What it means |
|---|---|
| Test script | A documented scenario with steps and expected results, written from a business process. |
| Entry criteria | What must be true before UAT starts — build complete, data loaded, environment stable. |
| Exit criteria | What must be true before sign-off, including severity thresholds for open defects. |
| Severity | How badly a defect affects the business, as distinct from how quickly it will be fixed. |
| Regression | Re-testing what previously worked, because fixes break things. |
They know the intended path and will unconsciously avoid the edges. It has to be the users.
The problems live in your real data's history and oddities.
A missing feature that was never in scope is a change request. Mixing the two makes the defect count meaningless.
A signature obtained to hold a date transfers the risk to the first live month.
| Metric | What it tells you | What distorts it |
|---|---|---|
| Script coverage of real scenarios | Whether tests reflect what the business does, including the exceptions. | |
| First-pass rate | Build quality. Very high rates early often mean scripts are too easy, not that build is excellent. | |
| Open defects by severity at exit | Against agreed criteria. Going live with open high-severity defects should be a recorded decision. | |
| Defects found post-go-live vs in UAT | Testing effectiveness, measurable only afterwards but worth knowing for next time. |
We build and support the systems this page describes. Start with a free working proof of concept — no cost, no commitment.