All articles
12 min read

How to Test Business Integrations Before Go-Live

Two beige rotary telephones on a steel-grey bench connected by one coiled cord, with an unplugged patch plug, a clipboard and a small glowing amber indicator lamp nearby.

To test a business integration, run each real workflow end to end with realistic records, check that both systems end up with the same data, then deliberately break the connection to see how it recovers. Do it in a test environment first, repeat a short smoke test in production after release, and keep the checks written down so they can be rerun whenever either system changes. This guide gives you the order, the checklist and the points where integrations usually fail.

It is written for founders, owners and teams who connect a shop, CRM, ERP, accounting tool or internal app and want to know it works before customers or staff depend on it. You do not need to be a developer to run most of it.

TL;DR

  • An integration test checks the handoff between systems, not each system alone. Both can be fine on their own and still fail when they exchange data.
  • Test whole business workflows (order to invoice, new customer to CRM) at least five times with different records, then compare the data in both systems.
  • Test failure on purpose: timeouts, expired logins, duplicate messages, rate limits and downtime.
  • Write the checks down as a checklist, so a person or a script can repeat them after every change.
  • Senior eyes on the test plan matter most at the start, when the cheapest fixes are made.

What an integration test actually checks

An integration test checks how two or more components behave once they are connected. A Swiss training module on testing levels describes it as checking the interplay of components after they have been integrated, aiming at errors at the interfaces or in the combined behavior, which can appear even when the individual component tests found no defects. A typical case: both systems work, but one sends a date in one format and the other reads it in another.

In business terms, you are asking three questions:

  1. Does the right event in system A trigger the right action in system B?
  2. Does the data arrive complete, correct and without duplicates?
  3. What happens when something goes wrong in between?

That is different from checking that your web shop takes an order, or that your accounting tool issues an invoice. Those are tests of the parts. The integration test is about the seam.

How it differs from unit, system and acceptance tests

The terms get mixed up, so here is the short version. Deloitte’s description of the four levels is a good reference:

Level What it checks Who usually runs it
Unit test Each component works according to its design Developers
Integration test Interactions between integrated components or systems Developers, with business input
System test The complete system against functional and non-functional requirements Test team
User acceptance test (UAT) Business users confirm the system meets business requirements and is ready for release People who will use it

The International Software Test Institute places integration testing after unit testing and before system and acceptance testing, and notes that acceptance testing includes real users before launch.

One caveat for small teams: the line between integration testing and system integration testing, which includes external systems such as a payment provider, is not always sharp and is decided project by project. Do not spend energy on the label. Decide which connections your business depends on, and test every one of them.

Step 1: List every connection and the workflow behind it

Before testing anything, write a one-page map. For each connection note:

  • the two systems and the direction of data (one-way or both ways)
  • the business event that starts it (a paid order, a new lead, a stock change)
  • what should happen afterwards (an invoice, a CRM record, a shipping label)
  • who notices if it fails, and how fast

This map becomes your test plan. It also exposes the connections nobody owns. Many small companies find an old Zapier-style link or a script that a former employee built, and that still moves money or customer data.

Where it goes wrong: teams test the connection they just built and forget the ones around it. A new shop integration can change the data your ERP receives, which breaks a report nobody thought about.

Step 2: Check the data mapping field by field

Most integration bugs live in the mapping, meaning which field in system A fills which field in system B. Check four things for each field:

  • Correct pairing. The customer’s company name does not land in the contact name field.
  • Matching data types. A date written day-month-year in one system and month-day-year in another silently swaps days and months for every date whose day is 12 or lower. Nothing errors. The data is just wrong.
  • Required values. If the destination demands a field, such as a VAT number or a postcode, decide what happens when the source leaves it empty.
  • Length limits. A 60-character address squeezed into a 40-character field gets cut off without a warning.

Also check amounts, currencies and time zones. Switzerland adds its own wrinkles: addresses in four national languages, umlauts and accents in names, and CHF and EUR often living side by side in one shop. Put a customer called “Müller-Béguin” with a long street name through the whole chain and see what comes out.

Step 3: Run the core workflows end to end

Now test the business flows, not individual calls. A sensible set for most small companies:

  • new customer in the CRM appears in accounting
  • order in the shop creates an order or invoice in the ERP
  • payment status updates back in the shop
  • stock change in the ERP updates the shop

For each flow, run it from start to finish with several different records. A commonly cited checklist suggests at least five runs per workflow with different data, and that is a reasonable floor: one run proves it can work, five start to show whether it works for a normal customer, a customer with a long name, a foreign address, a discounted order and a refund.

For each run, verify:

  1. The trigger fired.
  2. The data arrived with correct values.
  3. The record is linked to the right customer or order, with no duplicate.
  4. Follow-on actions fired, such as an email, a task or a status change.
  5. Both systems show the same state afterward.

Write down what you entered and what you saw. If you cannot show it, you cannot rerun it later.

A worked example

Say you run a Zurich online shop connected to an ERP. Test order 1: a standard Swiss customer paying by card, one item. Test order 2: a customer in Germany with a discount code and two items. Test order 3: an order with a very long delivery address. Test order 4: a cancelled order. Test order 5: a partial refund. After each, open the shop and the ERP side by side and compare totals, VAT, address, status and stock. A single mismatch in the VAT on order 2 is exactly the kind of fault that shows up in a quarterly accounting review months later if you skip this step.

Step 4: Break it on purpose

The happy path is the easy half. Real integrations spend their lives dealing with slow servers, expired credentials and messages that arrive twice. Test these cases:

  • Timeouts and downtime. Switch off the receiving system for ten minutes during a run. Does the sender queue the data and retry, or does it lose it?
  • Authentication expiry. Tokens and API keys expire. Does the integration refresh them, and does someone get an alert when it cannot?
  • Duplicate messages. Systems often resend a message when they do not get a reply. Sending the same order twice should not create two invoices. Engineers call this idempotency: repeating an action leaves the same result as doing it once.
  • Rate limits. If you import 5,000 products at once, does the other system start refusing requests, and what happens to the rest?
  • Bad data. Send a malformed record. It should be rejected with a clear message, not crash the whole batch.
  • Sensitive data. Check that personal data does not show up in logs or error messages. Under Swiss data protection law this matters, so use anonymized or invented customer data in test systems rather than a copy of your live customer list.

Each of these should end in one of two outcomes: the integration recovers by itself, or a named person is told. The worst outcome is silence.

Step 5: Use a sandbox and a real-world acceptance run

Most payment, shipping and cloud providers give you a test environment, often called a sandbox, where nothing real is charged or shipped. Use it for everything above, then run a final acceptance pass with the people who will live with the result: the person who books invoices, the person who packs orders.

Some providers formalize this. SIX Payment Services’ integrator test script for attended point-of-sale terminals tells integrators to execute and sign the described tests, document deviations by test number, and submit the script and related receipts for acceptance. Swiss public transport does the same for data interfaces: the sector publishes a baseline test protocol for connecting a VDV 453/454 interface. The lesson applies to any company: a signed, numbered test record turns “it seemed to work” into evidence.

After go-live, run a short smoke test in production, a handful of real but harmless transactions, to confirm that production keys, URLs and permissions are correct. Test and production settings are different, and this is where “it worked yesterday” usually breaks.

Step 6: Automate what you will repeat

Integrations are not finished at launch. A vendor updates an API, a field is renamed, a new product type appears. Every change is a reason to rerun your tests, and that is cheap only if the tests are automated.

Postman’s official guide says integrations can be tested manually, on a schedule with notifications, or in a CI/CD pipeline after code pushes. CI/CD means the tests run automatically each time the code changes. The same guide shows how to validate response codes, headers and body data, order requests into collections that mirror a workflow, and use a create request followed by a retrieve request to confirm the record really exists. It also describes using environments to run the same suite against different URLs or settings, which maps neatly onto test versus production.

A practical rhythm for a small team:

  • on every code change: the automated workflow tests
  • every night or week: a scheduled run with an alert on failure
  • after any vendor announcement: a manual review of the affected mapping

Deloitte adds an organizational point worth copying: in every testing phase, development, testing and business should stay connected so defects are resolved within agreed timeframes, business questions get answered and progress is reported regularly. For a ten-person company that can be as simple as one shared list of open defects and a weekly fifteen-minute check.

An integration testing checklist you can reuse

Copy this into your project tracker and tick it per connection.

Setup

  • Connection map written: systems, direction, trigger, owner
  • Test environment uses invented or anonymized data
  • Credentials and scopes confirmed for test and production

Data

  • Every field paired and checked for type, format and length
  • Required fields handled when empty
  • Dates, time zones, currencies and VAT verified
  • Special characters and long values tested

Workflows

  • Each core workflow run at least five times with different records
  • No duplicates, correct links, follow-on actions fired
  • Both systems show the same final state

Failure

  • Timeout, downtime and retry behavior tested
  • Duplicate messages produce one result
  • Expired credentials trigger an alert
  • Rate limits and bad data handled
  • No personal data in logs or error messages

Release

  • Results recorded and signed off by a business user
  • Production smoke test passed
  • Logs show what was sent, received and which error occurred
  • Tests scheduled to rerun, and a person named to receive alerts

Where to bring in senior help

You can run most of this yourself for a simple connection between two well-documented tools. The effort grows with the number of systems, with custom or legacy software in the mix, and with the cost of an error, as with payments, stock or customer data. Three signals suggest it is time to involve an experienced engineer: the integration touches money or legal records, one of the systems has no test environment, or earlier fixes keep breaking something else.

The cheapest moment for that help is the planning stage, when the connection map and test plan are written, because a wrong data model is far easier to fix before it is built than after.

Ampersand is a small Zurich studio that builds system integrations and API connections for shops, CRM and ERP, and every project is led by senior engineers from the first consultation to handover. That continuity is meant to avoid mid-project rewrites and dependency traps, and our aim is that you can run what we build without being locked in. If you want an integration built, migrated or kept running, see the Ampersand pricing and services page: there are plans for a single connection, an integration suite and ongoing monitoring, and our Run & monitor plan is 750 CHF per month.

Common mistakes to avoid

  • Testing only the happy path. The first real outage will be the actual test.
  • Using one clean sample record. Real data is messy, long and full of exceptions.
  • Copying live customer data into test systems. Use invented or anonymized records.
  • No owner for alerts. A retry log that nobody reads protects nobody.
  • Treating launch as the end. Rerun tests whenever either system changes.
  • No written record. If the results live in one person’s head, the next change starts from zero.

FAQ

How many times should I test each integration workflow?

Run each core workflow at least five times with different records: a standard case, one with special characters, one with a discount, a cancellation and a refund. One run shows the flow can work. Several varied runs show whether it holds up for the customers you actually have.

What is the difference between integration testing and user acceptance testing?

Integration testing checks that connected systems exchange data correctly, and it is mostly technical. User acceptance testing has business users confirm the system meets business requirements and is ready for release. Do the integration tests first, so users are not the ones discovering broken data.

Do I need special tools to test business integrations?

For a first round, a spreadsheet of test cases and a side-by-side comparison of both systems is enough. Once you repeat the tests often, an API testing tool such as Postman can run request collections on a schedule or in a CI/CD pipeline and alert you on failure.

What should I test after go-live?

Run a short production smoke test with a few real but harmless transactions to confirm keys, URLs and permissions, then keep the automated checks running. Rerun the full workflow tests whenever either system is updated or a mapping changes.

Sources

Talk to us

Have a project this touches on?

A free 10-minute call is the fastest way to find out whether we are the right studio for it.

Book a free 10-min call