Regulations and compliance · Practical guide

EU GMP Annex 11 data transfer and interface checks

An interface can deliver a message successfully and still trigger the wrong business action. A quantity loses its unit, a status is mapped incorrectly or a retry posts the same transaction twice. Follow the data through to its use in the target process. This guide covers ongoing data exchange in GMP for human medicinal products. A one-off migration also needs evidence for the defined transfer and its completion.

Table of contents

At a glance

How do you verify that an interface preserves the meaning of GMP data?

Section 5 of Annex 11 addresses appropriate built-in checks when computerised systems exchange data electronically. Define the expected identifier, value, unit, status and permitted effect in the receiving system. Test rejection, retransmission and partial processing as well as a normal transfer. A success code shows the technical step it represents. Check the receiving business record to verify the actual outcome.

  • Follow the data from the source to the resulting action in the target process.
  • Give each mapping rule an expected result defined independently of the implementation.
  • Include errors, retries and unresolved intermediate states in the evidence.

1. Define the scope of the ongoing data exchange

Section 5 of the January 2011 Annex 11 covers electronic data exchange. Section 4.7 supports selecting appropriate test cases, including limits and error handling. Section 4.8 covers preserving value and meaning when migrating data into a different format or system. Some controls overlap, but the operating questions differ.

For an ongoing interface, establish what triggers an event, how it is processed and what the target system is allowed to do. We developed the tables below to support this analysis. They prescribe no particular communication protocol or architecture.

Sources: [1] [4]

Further reading:

2. Specify what the exchanged data mean

Identify the source, target and trigger. Describe the meaning of each field. Which identifier refers to the batch? Is the quantity per container or for all the stock? Does a status mean “tested”, “released internally” or simply “message processed”?

Specify permitted values, formats, units, mandatory data and what happens when information is missing. If two systems use the same status term differently, ask the process owner to confirm the mapping. A developer can implement the decision but cannot reliably infer it from a field name.

Sources: [1] [3]

3. Make mapping rules testable

This stock notification example uses kilograms in the source system and grams in the target. A status mapping controls whether material may be used in the example. It illustrates an assumed function and does not describe a particular product or statutory batch certification.

Use the agreed rules as a test reference. Repeating the program's transformation logic for the comparison can reproduce its error. Define expected values and states independently for selected critical cases. Choose format and rounding rules to match how the data are used.

Example mapping rules for a stock notification
SourceTarget ruleIndependently defined acceptance criterion
Batch B-027 Assign the record to the same batch. Do not match on a similar name alone.
Quantity 12.5 kg Multiply by 1,000 to convert to grams. 12,500 g for the same stock.
Status Q In this example, quality inspection is pending. Material use is not permitted.
Status F In this example, the material is released internally for use. Apply this status only to the corresponding stock.
Event E-084 Keep the unique event reference. The processing status can be retrieved for this specific event.

Sources: [1] [3]

4. Distinguish message receipt from completed processing

A positive acknowledgement can mean that a server has received the message. It may also confirm completion of a business posting. Find out exactly what your acknowledgement establishes. The source process may report success while the target still has a queued message or an unresolved processing error.

Define the relevant intermediate states and who monitors them. For a deferred record, specify whether to retry, investigate manually or terminate processing. A warning without an owner can be as consequential as a silent failure. Match monitoring to data risk and process needs.

Sources: [1] [3]

Further reading:

5. Trace failure cases to the outcome in the target

The example starts with no posting for E-084. A valid message for B-027 and 12.5 kg must produce one stock entry for 12,500 g against the right batch. Under the assumed status Q, material use remains blocked. Define these conditions before running the test.

Resending E-084 must not increase stock to 25,000 g. An interface can detect a repeat or handle it through another controlled method. Test the expected business outcome and verify that errors can be traced. The example leaves the implementation open.

Expected results for a defined starting state
Test inputExpected result in the target systemEvidence needed
Valid E-084, 12.5 kg, status Q 12,500 g posted once. Material use remains blocked. Source record, message and resulting target state
Send the identical E-084 again No additional stock. The repeat and its handling can be traced.
Unit missing No silent assumption that the value is in grams. Error or documented resolution
Unknown batch No posting against a substitute batch. Visible batch assignment error
Data transfer interrupted No transaction with an unresolved state is reported as complete. Evidence of the processing state and restart

Sources: [1] [3]

6. Consider message order and partial processing

Related messages may arrive late or be processed only in part. Define what happens when a release arrives before the stock record it refers to, or when an updated status overtakes an older message. The required behaviour follows from the business process. It does not prescribe a particular message broker.

Consider concurrent updates and manual correction. Who resolves a missing reference? How does a restart avoid inconsistent states? Interrupting the process deliberately in a suitable test environment may reveal more than repeating the same successful transaction. Choose the environment and tools for what the evidence needs to establish.

Sources: [1] [3]

7. Carry the agreed rules into operations and change control

An initial successful test cannot keep an interface under control indefinitely. Assign responsibility for monitoring, errors and changes. A new status code, unit or field length can change meaning while the connection still works.

Assess changes to the source system, the target system and the transfer between them. Select regression tests from the rules and dependencies affected. Determine whether full daily reconciliation is appropriate for the process. The controls and the response to discrepancies must be traceable.

Sources: [1] [3]

8. Prepare the tests with the process owner and IT

The process owner defines meaning and permitted outcomes. IT explains transport, transformation and error states. QA links these to the evidence required by risk. Waiting for a failed import before bringing the three perspectives together often leaves the team without an agreed expected result.

Start with a few cases that distinguish your actual rules and justify any further coverage. Our example table sets no regulatory minimum number of tests. Include other failure consequences that matter in your process.

  • Identify the source, target, trigger and owner.
  • Define identifiers, values, units and statuses in business terms.
  • Distinguish technical acknowledgement from final processing.
  • Define independent expected results for normal, boundary and error cases.
  • Assess retries, message order and partial processing.
  • Assign responsibility for monitoring, errors and changes.

Sources: [1] [3]

Project experience and collaboration

Check that data retain their meaning across the interface

We ask the owners of the source and target systems to explain the same transaction and then compare the transferred data. Are unit, status, identity and time still clear? With IT, we test the normal case, repeat messages, interruption and a delayed message. Targeted additional testing may suffice for a stable, unchanged connection. Changes to mapping or the intended use in the target require a wider assessment. We define the expected decision before the test so that successful transmission alone cannot establish that the business result was correct.

Frequently asked questions about Testing interfaces

Does a successful API call prove the data transfer was correct?

It establishes only what that response is defined to mean. Check the processed record and the resulting action in the target system. An accepted message may still be queued or may subsequently fail a mapping rule.

Sources: [1] [3]

Which fields should we compare?

Select the fields needed to establish identity, completeness and meaning. These can include identifiers, units, quantities, status, the relevant time and a decision reference. A count of records or a total quantity alone often misses errors that matter to the process.

Sources: [1] [3]

Do all interfaces need the same tests?

No. An informational extract and a transfer that determines whether material may be used can have different consequences. Choose methods and scope from the actual rules, failure consequences and existing controls. A common technology does not establish identical evidence needs.

Sources: [1]

Must the interface prevent every repeat transmission?

Not necessarily. It must handle repeats in line with the defined process. In this example, receiving the same event twice must not create a second stock entry. Test how your implementation detects or otherwise controls retries.

Sources: [1]

May data be buffered temporarily?

A buffer can be acceptable. Assess protection, completeness, processing state and subsequent transfer. PIC/S identifies risks associated with temporary storage. A controlled intermediate state must remain distinguishable from a fully processed record.

Sources: [3] [1]

Is a checksum sufficient?

A checksum can detect certain changes in transit or storage. It cannot by itself establish that a status was translated correctly or a result assigned to the right batch. Check transfer integrity and the meaning of the data received.

Sources: [3] [1]

How do we test a unit conversion?

Define the source unit, target unit and expected calculation independently. In this example, 12.5 kg multiplied by 1,000 gives exactly 12,500 g. Test invalid or missing units and relevant boundary cases. Matching digits alone cannot establish matching quantities.

Sources: [1]

Is every transfer a migration?

Ongoing exchange under section 5 and a defined migration under section 4.8 address different needs. Both can involve transformations. A system replacement also needs a controlled migration dataset and closure. Ongoing exchange needs operating controls and error handling.

Sources: [1]

What evidence should an error test produce?

Record the error you triggered, the observed response and the resulting data state together. A warning screenshot alone cannot show whether an incorrect posting also occurred. Capture the relevant outcome in the target and any controlled follow-up.

Sources: [1] [2]

When is retesting needed after a change?

Assess the rules, components and dependencies that changed. A new status value can require different checks from a wording change. Explain the coverage you select. The term “interface change” alone justifies neither a complete retest nor skipping tests.

Sources: [1]

Sources and review date

Sources were checked on 7 October 2026 for GMP manufacture of human medicinal products. EudraLex listed the January 2011 versions of Annex 11 and Chapter 4. The 2025 consultation draft is identified separately. We developed the worked examples and tables as practical aids.

  1. EU GMP Annex 11: Computerised SystemsJanuary 2011 revision, in operation since 30 June 2011. Accessed 7 October 2026. Sections 1, 4.3–4.4, 4.7–4.8, 5 and 10–13.
  2. EU GMP Chapter 4: DocumentationJanuary 2011 revision, in operation since 30 June 2011. Accessed 7 October 2026. Sections 4.1 and 4.8–4.10.
  3. PIC/S PI 041-1: Good Practices for Data Management and IntegrityPI 041-1, 1 July 2021. Accessed 7 October 2026. Sections 9.1.5–9.1.6 and section 9.4, pages 38–39.
  4. European Commission, EudraLex Volume 4On 7 October 2026, the human medicines section listed the January 2011 editions of Annex 11 and Chapter 4. Accessed 7 October 2026 under Human medicines, Annex 11.
  5. European Commission, 2025 consultation on Chapter 4, Annex 11 and Annex 22The consultation ran from 7 July to 7 October 2025. It was closed when checked on 7 October 2026. Under “Details” and “Revision of Annex 11”, the links explicitly identify draft guidelines. We cite the consultation to establish status.

Sources checked: .

Daniel Herrmann, Daniel Herrmann Consulting

Daniel Herrmann Consulting

Discuss the controls and evidence your interface needs

We help you apply this approach to your actual systems, assess gaps and prepare the evidence. We agree scope and responsibilities with the people who operate the process.

  • Computer system validation →

    Bring your GxP systems into validated operation, from laboratory software, MES and eQMS to ERP. We define the scope, plan risk-based tests using GAMP 5 and hand the evidence to QA for release.

Send a project enquiry Customer testimonials →