Order integration acceptance testing: what to check before website launch

Order integration acceptance testing: what to check before website launch

During a demonstration, a request from the website appears in the business management system, and everything looks ready for launch. But a buyer may click the button twice, send a file without a quantity, or amend a request while a sales manager is checking the previous version. Order integration acceptance testing is needed so that, before paying for the final stage, you see not only a successful transfer but also predictable behaviour in these situations.

For a B2B store owner, this is a practical question: what exactly should you require from the contractor so that sales managers can work with the new request workflow? You should check the entire journey, from the received message to a confirmed document in the ERP, the enterprise's business management system. Below is a way to organise acceptance testing without substituting an attractive screen for a business outcome.

First agree on what you are accepting

The phrase “orders are transferred to the ERP” leaves too much room for different expectations. For a developer, it may mean delivering a message to an intermediary service. For the sales department, it means a document created with the correct buyer and products. Write down a shared definition: which data comes from the website, where it is validated, who approves exceptions, and which record confirms completion.

Distinguish receiving a request from accepting an order for fulfilment. A “Request received” message may appear before the ERP confirms the document. The website must not describe this intermediate state as a completed order. Also agree on whether stock is reserved and whether a confirmation is sent to the buyer at this stage: these actions have consequences of their own.

If you are still designing a B2B self-service portal, first define its roles and the actions available to buyers. An acceptance session has a different purpose: to demonstrate that the agreed workflow follows the rules even when data is invalid or failures occur. Do not confuse it with a presentation of every feature in the customer account.

Turn one request into a set of clear checks

A hypothetical example, not a GL.ua case study: a dealer uploads request Z-318 through the website for four boxes of size M gloves. In the agreed reference data, a box contains 25 pairs, so the expected quantity is 100 pairs. The buyer, product, unit of measure and permitted price list have been created in advance for the test. All identifiers and quantities here are illustrative.

Give the contractor not only the file but also the expected result. Otherwise, the team may successfully implement its own assumption about your request. The product reference data owner confirms the product and packaging match; the sales department confirms the buyer and commercial terms. If these people have not agreed on the interpretation, the test is not yet ready to run.

Input situationWhat the client should seeWhat to check in the business system
A complete, valid requestA draft linked to the source and containing the agreed dataOne document after authorised confirmation
The request has no quantityA clear reason for stopping and an action to resolve the missing informationNo order created using a substituted value
The same submission is repeatedThe current result of the same operationNo second, independent order
The ERP response does not arrive in timeA state requiring reconciliation, rather than apparent completionThe actual presence or absence of the document is established

This expected results matrix must be agreed before the demonstration. Add an owner and evidence for each row: the sales manager's screen, an operation identifier, and the result in the test ERP. The entries in the table are requirements for your project, not a promise that every off-the-shelf system behaves this way.

Make errors visible rather than “correcting” them at random

Introduce several different errors into the test request. Remove a required field, submit text instead of a number, a negative quantity, an unknown product code and an incompatible unit. Do not limit the check to the button in the browser: the component receiving data on the server must validate the value. Otherwise, another client application or an external service may bypass the on-screen restrictions.

OWASP distinguishes valid format from valid meaning. A number may be correctly written but fail to meet the permitted quantity or packaging multiple. That is why business rule validation does not end with reading fields. For request Z-318, the system must know what “box” means, not merely convert a word and a number into a data structure successfully.

If AI recognises the data first

Document recognition can suggest values, but it does not define your sales rules. If one customer product code corresponds to several products, clarification or an agreed mapping is needed. The model's confidence does not replace that decision. Check not an impressive recognition percentage but whether the workflow can stop a specific unusable request.

You do not have to start with AI. A structured form or an agreed import template may be enough for a stable dealer request format. A chatbot or voice assistant is simply another possible way to collect data. Acceptance requirements for the final record do not disappear when the interface changes.

What the sales manager does with a stopped request

Ask to see the clarification queue: the original material, the problematic field, the reason for stopping and the next permitted action. “Processing error” without an explanation does not help the sales department. The sales manager must understand whether to contact the buyer, correct the reference data or refer the issue to the person responsible for integration.

Run the corrected request separately: it must go through validation again while preserving the error history. Accept not only the stop mechanism but also a working route back into processing.

A repeated message is not a new purchase

Two identical sets of products can represent two intentional purchases. Conversely, different messages may refer to the same request. Comparing only the total or cart contents is therefore insufficient. Together with the contractor, define an order intent identifier: how does the system distinguish another attempt to transmit the agreed operation from a genuinely new order?

In technical documentation, this relates to idempotency: repeating a request does not produce another identical side effect. The AWS explanation of safe retries discusses the client request identifier and why identical parameters alone do not prove identical intent. The specific mechanism for your ERP must be agreed and demonstrated separately.

Reproduce clicking confirm again and redelivering a message. Then ask the technical team to run two attempts simultaneously: a sequential retry after completion can conceal a concurrency bug. After each test, inspect not only the sender's log but also the recipient's list of actual documents.

Clarify how long the integration recognises a repeat and how it behaves after that period ends. Do not transfer the limitations of one API to another system. Also test a new dealer order with the same set of products: duplicate protection must not silently discard a legitimate repeat purchase.

The most important pause is after sending but before the response

Imagine a continuation of the hypothetical test: the sales manager confirmed Z-318, the integration sent the data, but the connection broke before the response arrived. The ERP may already have created the document. A red message on the website does not disprove that. If the sales manager immediately creates a new request manually, two orders may turn up later.

This requires a separate unknown outcome state. It means not “there is no order” but “the outcome is not yet confirmed”. Ask the contractor to show how the system reconciles the status using the operation identifier, who investigates incomplete cases and when another attempt is allowed. The procedure depends on the capabilities of the specific ERP interface.

In the test environment, reproduce the loss of the response after the document has actually been created. The expected result is finding the existing document and restoring its link to the request, not creating a new record. Separately reproduce a situation in which the submission never happened. These cases may look similar to the user but require different actions.

Also test restarting the intermediary service between these events. The record of the received intent and processing result must not exist only in RAM. The technical team must explain how the operation remains connected after a failure, and the client must see evidence in a reproduced scenario.

Assess the sales manager's work, not just the number of successful requests

During acceptance testing, put the person who will process requests after launch in front of the working interface. Let them find a stopped document, clarify the data and explain the current status to the buyer without prompting from the developer. If every case requires reading a technical log, the interface does not yet solve the operational task.

For each scenario, retain evidence of the end-to-end result: the input data, applied rules, sales manager's screen and ERP record. Identifiers must connect these parts. Do not copy complete customer documents into every log: agree on the data to be included and access permissions, and use approved anonymised materials for tests.

Summarise how many requests were completed, how many were stopped for justified reasons and how many remain without a definite outcome. Record manual clarifications and active working time separately from waiting time. A high overall figure must not conceal a critical defect: examine an extra document or an incorrect product quantity separately.

End acceptance testing with a decision that can be put into practice

Run the session in the agreed order:

  1. Compare a normal request with the expected ERP document.
  2. Check data errors and the return to processing after clarification.
  3. Reproduce retries and connection failures.
  4. After fixing a defect, repeat the failed scenario and the related valid scenario.

Record the integration version and the rules used in the results.

Preparing this acceptance process is work too. Timelines depend on the availability of the test ERP, agreed reference data, sample requests and participation by the responsible sales and accounting staff. In the estimate, separate implementation of the data exchange from the testing environment, failure reproduction, clarification interface and handover of instructions. Cost depends on the agreed scope; there is no universal integration price.

The launch decision must name the permitted channel, tested scenarios, open defects and the person responsible for uncertain outcomes. Add a way to stop the new workflow while retaining manual request intake. A finite set of tests does not prove the absence of all future errors, but it provides a clear basis for approving or postponing a specific launch.

GL.ua (Glyanets) includes integrations and testing in its full-cycle website development service. If you are planning this kind of store development, start the discussion of website development and integration with one request workflow and a list of outcomes the business needs to see during acceptance testing. This will help you assess an agreed scope of work rather than a general promise to “automate orders”.

Order a site now!

Just one step to your perfect website

Accessibility menu
Contrast settings
Font size
Letter spacing
Line height
Images
Font
Reset the settings