Skip to content
  1. Home
  2. Articles
  3. What Is UAT? What the Client Should Define Before Signing Off a System
115
Enterprise

What Is UAT? What the Client Should Define Before Signing Off a System

What Is UAT? What the Client Should Define Before Signing Off a System

UAT is testing by the people who will actually use a system, before the client signs acceptance. This article covers how UAT differs from SIT, who should test, how to write pass criteria and acceptable defect levels into the TOR, and what a sign-off document should contain.

The acceptance form on your desk has one line for a signature. What that signature confirms is that the system works as agreed. The questions to ask before you pick up the pen are what exactly was agreed, and who proved it. If the answer is "the developer says it has been tested", you are signing for the other side's assurance, not for your own test results.

User acceptance testing is where those questions get answered. This article is written for the client side: the IT lead, the system owner and whoever drafts the TOR (terms of reference). We will cover how UAT differs from the developer's own testing, who should do it, how to write pass criteria, and which documents should exist before anyone signs.

What UAT is, and how it differs from SIT

UAT stands for User Acceptance Testing: testing by the people who will actually use the system, to decide whether it is ready to be accepted. The ISTQB Certified Tester Foundation Level Syllabus v4.0.1 (2024) describes acceptance testing as focused on validation and on demonstrating readiness for deployment, and says it should ideally be performed by the intended users. UAT is one of several forms, alongside operational, contractual and regulatory acceptance testing.

SIT, or System Integration Testing, sits at a different level. It checks that the system talks to other systems correctly: data reaches the accounting system in full, or a call to an outside agency's API returns what the specification says it should. SIT is usually run by the development or test team, and the question it answers is whether the system was built to the design.

UAT answers a different question: whether a system built to the design can carry the organization's real work. A system can pass every SIT case and still make a clerk click through six screens to approve one document, or produce a month-end report that does not match the form management actually uses. Problems like these are not in the developer's test cases, because the developer does not do that work every day.

Meeting the spec and being usable are two separate things.

The order matters. The developer finishes its own testing first, and only then hands the system to users for UAT. The NHS Business Services Authority, a UK public body, states in its testing playbook (published 2026) that the test team should sign off before user acceptance testing begins. If users keep running into basic defects the developer should have caught, UAT time goes on reporting the same bugs instead of checking whether real work flows.

Who should run UAT

UAT should be run by the people who will use the system every day after go-live: the staff who enter data, the supervisors who approve, and whoever pulls the reports others depend on. They know which cases happen constantly, which happen once a year but must never be wrong, and which odd-looking records are perfectly normal for the business.

Their time has to be planned. Staff pulled into UAT still have their regular jobs. If no time is set aside, UAT gets squeezed into late afternoons, testers click through only the common paths, and nobody tries the hard cases. The head of the unit should approve the testers' time in writing when the project plan is drawn up.

The organization's IT team coordinates and holds the criteria; it does not test on behalf of every user. The developer should be present during UAT to take issues, explain behaviour and fix things, but should not be the one doing the testing. Developers know how the system was designed to be used, so they tend to walk the paths it already supports without meaning to.

Acceptance criteria belong in the TOR from the start

ISTQB defines acceptance criteria as the conditions an implementation must meet to be accepted by stakeholders. The definition sounds ordinary, but it matters to the client. If the conditions are not written down before development starts, acceptance turns into a negotiation over what counts as the agreed work.

For Thai government agencies, clause 175 of the Ministry of Finance Regulation on Public Procurement and Supplies Administration B.E. 2560 (2017) requires the inspection committee to check that supplies are correct and complete against the evidence agreed in advance, and allows it to invite experts when technical testing is needed. UAT criteria written clearly into the contract are that evidence. Without them, the committee has little to compare against except the contractor's own delivery documents.

Usable criteria have one thing in common: each can be answered pass or fail. "The system is easy to use and supports the unit's work" cannot. "A clerk can record one request, from opening the screen to submitting it for approval, without leaving the system to enter data elsewhere" can. The TOR should contain at least three parts: a list of business scenarios that must pass, criteria for how many defects of which severity are acceptable, and the positions authorized to sign.

The scenario list does not need to cover every button. Start with the work the unit really does each week, then add the work that is rare but must not go wrong, such as closing the financial year or issuing a certificate with legal effect.

Which defects can go live

A system with no defects left on acceptance day hardly exists, and a zero-defect criterion tends to end one of two ways. Either the project slips indefinitely, or everyone quietly stops counting some kinds of defect so the form can be signed. Both are worse than admitting up front that some defects will remain, and agreeing which levels are acceptable.

ISTQB notes that testing can end before the criteria are met if stakeholders have reviewed and accepted the risk to go live. That sentence does not license skipping the criteria. It says accepting risk has to be a decision someone signs, not something that happens because time ran out.

The common approach is to grade defects by severity and require some grades to be at zero before sign-off. NHSBSA's criterion is that no Priority 1 or 2 defects remain outstanding; any remaining defects are recorded in the test completion report with their severity, user impact and expected fix date. Those levels are that organization's choice and you do not have to copy them, but the structure transfers well.

What must be agreed precisely is the definition of each level. If the contract only says "critical" and "non-critical", users and developers will grade the same defect differently every time. Define levels by their effect on work. For example: the top level means core work cannot be done and there is no workaround; the next means it can be done, but only through a workaround staff have to remember; the lowest means appearance or wording is wrong without affecting the result.

The UAT environment and test data

UAT should run on an environment close enough to production. ISTQB says system-level test environments should be similar to the operational environment, and the same principle applies to UAT. If UAT runs on a developer's machine, or on a server not connected to the systems it must integrate with in real life, a pass tells you less than you think.

Test data is the other decision. Sample data created by developers tends to be too clean: no unusually long names, no records with empty fields, none of the oddities accumulated in the old system. Copying the full production dataset carries personal-data risk. The middle path is real data that has been masked, which we cover in our article on masking test data.

If the project migrates data from an old system, run UAT on migrated data at least once. Users will notice immediately when a balance or a transaction history does not match what they are used to seeing. How to check that migrated data is actually correct is in Complete Is Not Correct.

UAT does not test everything. Handling many simultaneous users needs a different method, such as load testing. Security needs specialist testers, and we cover how to procure them in our guide to procuring a penetration test.

What a UAT sign-off document should contain

A good sign-off document is more than a signature line and the word "passed". At minimum it should state which scenarios were tested, the result of each, which defects remain, at what severity, and who will fix them by when. The last part is the residual risk and who accepted it. NHSBSA requires sign-off to include residual risk, signed by the Product Owner, Technical Lead, Delivery Manager and Test Lead.

This document earns its keep on the day nobody remembers what was agreed. In the second year after go-live, when someone asks why a particular report has never come out right, the answer should be findable in the sign-off: whether it was logged as an outstanding defect, and when it was agreed to be fixed.

For Thai government agencies, a UAT document does not replace the formal inspection of supplies under the regulation. What it does is give the inspection committee concrete evidence that the system works against the agreed conditions. Clause 175 also says that if supplies forming a set are missing a component so that they cannot be fully used, they are treated as not yet delivered. A system whose core functions cannot yet carry real work deserves the same view.

Limits of UAT worth knowing in advance

Even well-run UAT lets defects through. There are few testers, limited time, and only the scenarios someone thought of get tested. Some work happens only at fiscal year-end, or only when hundreds of users log in at once on day one. Passing UAT lowers the risk considerably, but it does not mean the system will have no problems after go-live.

The pushback we hear most often is that the developer has already tested, so why should the organization spend staff time testing again. It is a fair question. The developer's testing checks that the system does what the specification says. The specification itself may have missed what the work actually needs, and the only people who can check that are the people who do the work. The two layers answer different questions, so they are not duplicates.

What UAT does not cover is the period after go-live. Whether a defect found after sign-off gets fixed depends on the contract, not on the UAT result. The warranty period, and the line between fixing defects and adding features, should be written clearly into the contract.

One more thing we should say plainly: this article relies mainly on ISTQB and NHSBSA. Thailand has TIS 29119 Part 1-2567 (2024), which presents the core concepts of the ISO/IEC/IEEE 29119 software testing series, but we have read only its notice in the Royal Gazette, not the full standard. If your organization needs to cite it in a TOR, get the full text from the Thai Industrial Standards Institute and review it first.

Where to start

If the project is still at the TOR stage, start with a list of ten to twenty business scenarios the system must handle, and ask the owner of each area to add to it. That list becomes both the scope of work and the UAT criteria.

If development has already started and the contract has no UAT criteria, it is not too late. Agree a written memo with the developer covering the scenario list, the defect level definitions and the signatories before the test cycle begins. Agreeing now is harder than at the TOR stage, but far easier than arguing on acceptance day.

If acceptance day is close, ask the developer for at least three things: evidence that its own testing is complete, the list of outstanding defects with their severity, and an environment users can test on by themselves, without a developer sitting next to them.


In short

UAT is where the organization proves for itself that the system can carry real work, before the acceptance signature shifts responsibility to the client side. What gives UAT meaning is not the number of test cases. It is criteria agreed in advance, and testers who actually do the work.

If you take one thing from this article, write the scenarios that must pass, the defect levels you will accept and the names of the signatories into the TOR from the first draft. Those three turn acceptance day into a comparison against criteria instead of a negotiation.

If your organization is drafting a scope of work, or is about to enter UAT, and you would like someone to review your acceptance criteria before you begin, we would be glad to talk. You can reach us at 088-983-9386 or [email protected], and our office is in Bangkapi, Bangkok. There is no deadline and no need to decide anything quickly.

FAQ: Frequently Asked Questions about This Article

A collection of questions and answers to help you better understand the content of this article.

Tags:

uatuser acceptance testinguat sign offsit vs uatacceptance criteria
Share:

Other Articles

Stay tuned for upcoming articles!