Eight UAT Mistakes That Push ERP Go-Live Back, and How to Test Instead
Most ERP go-live delays trace back to a UAT plan that tested the wrong things. Here are eight common mistakes, the fix for each, and exit criteria you can measure.
Happy-path scripts prove the demo works, not your business
If your go-live date is slipping, check the UAT plan before you look anywhere else. Delays in ERP user acceptance testing rarely come from bad software. They come from a test plan that checked how the system behaves on a good day and never how it fails on a bad one. Defects that should surface in week three of UAT surface in week one of production instead.
The first mistake is scripts that cover only the happy path: a clean sales order, full stock, correct price, on-time shipment. Real operations aren't like that. Customers short-pay, suppliers send partial deliveries, credit holds kick in, and someone keys the wrong unit of measure.
What to do instead: for each end-to-end process, write ERP UAT test scripts for three kinds of case: the standard flow, the common exceptions (partial receipts, backorders, returns, price overrides, credit holds) and deliberate negative tests where the system should block the user. As a rule of thumb, write one to three exception scripts for every happy-path script in order-to-cash, procure-to-pay and record-to-report.
Exit criterion: every in-scope process has signed-off exception and negative scripts, and the business process owner, not the project team, has confirmed the exception list matches what happens on the floor.
Consultants signing off the configuration they built
When the implementation partner writes the scripts, runs them and records the results, the test shows only that the configuration matches the partner's design. It doesn't show that the design fits your business. This isn't bad faith. It's the same problem as letting a developer approve their own code.
What to do instead: consultants can draft scripts and coach testers, but real end users run the tests: the AP clerk, the warehouse supervisor, the plant planner. Sign-off belongs to named business process owners. Budget for it honestly. Key users often need a large share of their time, commonly a third to a half, for several weeks of UAT, and their day jobs need backfill. Without it, testing gets squeezed into lunch breaks and nobody logs defects.
Watch the cost. Partner day rates for UAT support vary widely by region and seniority; typical 2026 ranges run from a few hundred to roughly $2,000 per consultant-day, and contracts differ. Confirm with your partner. Every consultant hour spent executing tests is one you pay for twice: once to build the configuration and once to approve it.
Exit criterion: client staff performed the large majority of test executions (aim for 90% or more), and each process area has a signed acceptance from its business owner.
Synthetic data hides the problems your migration will create
Test records a consultant built by hand are clean. Your legacy data isn't. It has customers with three ship-to addresses and no tax ID, items with inconsistent units of measure, years-old open purchase orders, and vendors set up twice. Synthetic data can't surface the defects that matter most at go-live, because those defects live in the data.
What to do instead: from the second UAT cycle onward, test on data produced by your actual migration tooling, from a recent production extract. Use full or near-full volumes for masters (customers, vendors, items, BOMs) and representative open transactions. Where data-protection rules apply (GDPR in Europe and the UK, India's DPDP Act, Brazil's LGPD, and comparable laws in the Gulf and Southeast Asia), mask personal fields while keeping the structure intact. Ask your privacy counsel whether a test environment holding production-derived data needs its own lawful basis or a data processing agreement with the partner.
Exit criterion: UAT ran on a migration load only a few weeks old, and reconciliations for open AR, open AP, inventory value and GL trial balance tie back to legacy within agreed tolerances.
Stubbed integrations that fail on day one
Mocking the WMS, the carrier API, the bank file or the e-commerce storefront speeds up early testing. Leaving those stubs in place through UAT is how companies discover at go-live that the 3PL rejects the ASN format or the bank won't accept the payment file.
What to do instead: treat ERP integration testing as part of UAT, not a technical task the business never sees. By the final cycle, every in-scope interface should connect to the real counterparty's test environment: transport management, EDI partners, payment providers, banks, and government e-invoicing portals where they apply. Test failures too: duplicate messages, rejected messages, timeouts, and messages that arrive out of order. EDI partners and banks may need weeks of notice to provide test connectivity, so ask early.
Exit criterion: no stubs in the final UAT cycle. Every interface has run end to end at realistic volume, error handling has been shown to work, and a named person owns monitoring each interface after go-live.
Skipping a full month-end and tax-period close
Many projects test individual finance transactions and never run a complete close. The gaps appear at the first real month-end: accruals that don't reverse, intercompany entries that won't eliminate, FX revaluation posting to the wrong accounts, and tax reports that don't tie to the ledger.
What to do instead: run at least one full simulated period close in UAT, ideally two, following the real close calendar: subledger cut-off, inventory valuation, depreciation, accruals, intercompany, revaluation, consolidation and management reporting. Then run the tax period the same way. Tax rules differ widely by country:
- VAT returns and digital record-keeping in the UK and EU.
- GST returns in India, plus e-invoicing and e-way bill integration.
- Clearance-model e-invoicing such as Mexico's CFDI, Brazil's NF-e and Saudi Arabia's ZATCA requirements.
- Sales and use tax engines in North America.
Have your tax advisor review the outputs, especially where you're switching tax engines or e-invoicing providers.
Exit criterion: the simulated close finishes within the target number of working days, the trial balance reconciles to migrated opening balances, and the controller and tax lead have signed off the tax reports.
Testing with superuser access instead of real roles
Testers often run UAT with broad admin access so they don't hit authorization errors. Then on go-live morning, half the warehouse can't confirm picks and finance can't release payments. Or worse, users can do things they shouldn't.
What to do instead: assign production security roles to testers by the second cycle and run every script under those roles. Add negative tests for segregation of duties. For example, confirm the person who creates a vendor can't also approve a payment to that vendor. Auditors will look for this, whether you report under SOX-style controls, local statutory audit or customer audits.
Exit criterion: every script has passed under a production role, the role-to-user mapping is signed off, and the segregation-of-duties conflict report shows no conflicts without documented mitigation.
Vague exit criteria that let the calendar decide
If "UAT complete" just means the scheduled weeks have passed, the steering committee ends up making a gut call under deadline pressure. Set objective gates before testing starts, while nobody is under pressure. A workable ERP testing strategy defines defect severity and a threshold for each level.
| Severity | Definition | Typical go-live threshold |
|---|---|---|
| Critical | Stops a core process (ship, invoice, pay, close) with no workaround | Zero open |
| High | Core process affected; manual workaround exists but is costly or risky | Zero open, or a few with workarounds the process owner has signed off |
| Medium | Non-core issue, or a workaround that's easy to sustain | Allowed with a dated fix plan |
| Low | Cosmetic or convenience | Allowed, logged to backlog |
Add coverage gates: 100% of in-scope scripts executed, and a final-cycle pass rate your steering committee agrees in advance. Many teams set it high, around 95%, with every failure explained.
Exit criterion: the go/no-go meeting reviews data against criteria documented and signed before UAT began. Nobody negotiates the criteria in the room.
Treating UAT as the finish line instead of a gate to cutover
The last mistake is assuming that once UAT is signed off, go-live is ready. Defect fixes made late in UAT can break things that already passed. And nobody has timed the actual cutover: the final data load, reconciliations, opening stock counts and interface switch-over.
What to do instead: after every fix release, re-run a regression pack of your most critical scripts. Then hold at least one full cutover dress rehearsal, timed hour by hour, with the real people and tools. ERP cutover readiness means you know how long cutover takes, the last point from which you can roll back, and who signs each checkpoint. A rehearsal that overruns the planned downtime window is a finding to fix, not something to explain away.
Your next step: check your current UAT plan against the eight criteria above. If three or more are missing, don't defend the date. Add a test cycle now. It will usually cost less than the first month of production firefighting.
Common questions
How long should ERP user acceptance testing take?
For a mid-market cloud ERP, plan for two to three UAT cycles over roughly four to eight weeks, plus time for regression and a cutover rehearsal. Multi-country or multi-entity rollouts usually need longer because each tax and close process has to be proven. Set the duration by scope and exit criteria, not a fixed calendar.
Who should write ERP UAT test scripts?
The partner can draft scripts from the process design, but business process owners should review them and add the exceptions they see every week. End users should execute them. Each process owner, not the implementation partner, signs off their own area.
Can we use production data in a test environment?
Often yes, but data-protection laws such as GDPR, India's DPDP Act and Brazil's LGPD may require masking personal data, restricting access, and having contracts in place with the partner. Rules differ by country, so confirm the approach with privacy or legal counsel before loading any data.
What's the difference between integration testing and UAT?
Technical integration testing checks that messages move correctly between systems. UAT checks that the business process works end to end, including those integrations. By the final UAT cycle the two should converge: real connections, real data, run by business users.
Written by the Carrier Mesh HQ editorial team. Prices, features and rules change often - confirm details with the vendor or a licensed professional before you decide.