Table of contents
At a glance
How do you demonstrate that a GxP data migration is correct?
Combine quantity reconciliation with checks of critical field content and tests of connected business transactions. Data criticality and transformation logic determine the strategy. Matching record counts do not prove matching contents or correct relationships. Agree acceptance criteria and exception handling before the final migration.
- During the SAP cutover, volumes, keys, and critical field values between defined source and target states can be automatically compared.
Scope and Applicability
The article elaborates on the validation strategy for data migrations. It does not specify a general sample size or allowable error rate. Such values require a justified derivation based on the specific dataset in question.
explicitly limit the statement of each reconciliation
A quantity comparison verifies the number or sum within a defined scope. A content comparison evaluates values and transformations. A relationship comparison ensures that associated objects remain correctly linked. A business process test finally confirms that the received information is used as intended. These levels complement one another but are not interchangeable.
Define what the comparison should establish for each data area. Item master data need different checks from historical laboratory records linked to methods and attachments. Address the actual risks. Requiring a manual visual check of every item may be disproportionate. Sampling only straightforward routine cases at random may leave important gaps.
Annex 11 requires appropriate controls for data transfer to ensure that value and meaning are not altered. The specific combination of controls demonstrating this must be justified within your project. The tables below are standalone methodological aids. They do not establish a universal sample size or acceptable error rate.
Use data criticality and transformation as a basis for validation.
Capture which data influence a decision and how they are altered in the target. An unaltered string has different risks compared to unit conversion, status mapping, or consolidation of multiple objects. Particular attention should be paid to key parameters, limits, date references, and excluded values, where they impact the technical or scientific validity of the statement.
Describe existing verification mechanisms. An automated field-by-field comparison can cover large datasets, but requires appropriate rules and well-defined boundaries. Manual reconstruction can reveal relationships, yet often focuses on a limited set of cases. The strategy should explain which risks are addressed by which form of evidence and which gaps remain.
| Risk | Appropriate internal comparison | Additional technical question |
|---|---|---|
| Missing data set | Quantity and key reconciliation | Was the record deliberately excluded? |
| Incorrect value | Critical Field Comparison | Is a transition planned? |
| Incorrect relationship | Reconstruct linked objects | Is the historical statement clear? |
| Incorrect status | Map and review downstream processes | What usage is permitted thereby? |
| Hidden Exception | Targeted Boundary and Special Cases | Is the rare critical situation covered? |
Example: 1,000 samples are correct, two results are not correct
Source and target each contain 1,000 samples. Two result assignments were swapped. The quantity comparison is valid, as is the sum of the result values. Only a key and relationship reconciliation identifies that the results are incorrectly attributed to the wrong samples. The number 1,000 is an example, not a requirement for your test batch.
The in-house testing strategy therefore links all expected sample keys with their result references. For selected complete assays, the method, unit, and evaluation are additionally reconstructed. A separate test set intentionally includes swapped relationships. This verifies whether the matching process can detect such errors.
If a difference is identified, not only the individual row is corrected. The root cause may lie in an inappropriate key logic and could affect additional objects. Determine the potential scope, amend the rule, and rerun the affected comparisons. Document the new inspection statement that results therefrom.
| Reconciliation | Result in example | Statement |
|---|---|---|
| Sample size | 1,000 in both systems | Quantity is correct, relationships unverified |
| Summary of results | Same total sum | Swapping remains invisible |
| Key and Relationship | Two incorrect assignments | Professional error identified |
| Complete process | Probe, Method, Result and Assessment | Context specifically verified |
| Breach error | Test tool detects known swap | Detection capability of the inspection logic verified |
Challenge automated comparison rules themselves
Script or tool, input files, and relevant settings shall be versioned. Ignored fields, normalizations, and tolerances shall be clearly named. A comparison may intentionally standardize whitespace; however, it must not inadvertently equate a critical identifier or unit without deliberate recognition. The rules must align with the intended technical and scientific meaning.
Use known positive and negative test data. Introduce missing rows, duplicate keys, inverted relationships, and erroneous transformations, to the extent such risks exist. The comparison should detect and accurately report expected differences. A flawless run without such challenges provides little insight into the system's ability to identify real-world errors.
Also verify report completeness. A tool can detect differences, but a truncated output cannot fully illustrate them. Document totals, error categories, and the capability for individual clarification. A green status is only meaningful when both the processed volume and excluded data are known.
Sample selection and targeted justification
A sample must have a clearly defined purpose. Is it intended to support the execution of a common, routine rule, or to evaluate rare, critical relationships? Random selection and targeted risk cases serve distinct functions. Both can be combined, provided the boundaries of conclusions remain clearly defined.
Consider data distribution and transformation variants. A migration dataset may contain multiple years, multiple locations, status models, or historical formats. Selecting only new data sets does not automatically capture older specificities. Identify relevant groups and select appropriate cases for each. Statistical conclusions must be based on an appropriate methodological foundation.
Document the rationale for combining an automated full-field comparison of specific fields with manual reconstruction of selected process steps. The term "full comparison" refers exclusively to the completely defined scope of comparison and does not constitute assurance that all relevant quality attributes have been evaluated. This boundary should remain clearly visible in the batch release record.
Evaluate differences and provide final acceptance and release.
Classify differences by root cause and potential impact. A planned format change may be acceptable if the intent remains intact. An unexpected status change or lost reference requires clarification. Establish acceptance criteria prior to final run, rather than adjusting them after reviewing errors.
Link each relevant difference to investigation, correction, and retesting evidence. A consolidated note "canceled" is weak if the scope of the issue and root cause remain unclear. Assess systematic effects beyond the originally identified individual case. Release refers to the actually corrected and reviewed status.
Obtain source data, mapping, tool version, comparison reports, and decisions. Changes arising between general release and final acceptance require an appropriate delta comparison. An earlier performed test is not a substitute for verification of the final data set and its authorized use.
Three tests address three distinct questions
Map test cases to the actual implementation rules. A verbatim text copy requires different controls than a unit conversion or the consolidation of multiple source objects. Rare but critical exceptions must not be excluded from testing solely because a random sample does not include them.
Automated comparisons can support large volumes of data. However, their rules, excluded fields, and error reports must be traceable. A green result is meaningless if the comparison deliberately overlooks critical parameter values. Verify the comparison logic using intentionally embedded test deviations.
Draft: Document inspection rules and their detectability
Document the purpose, key parameters, and justified exclusions for each comparison rule. The expected reference result list for deliberately altered test data must be independently determined, irrespective of the specific comparison logic being applied. Otherwise, the same error could bias both the comparison and its apparent validation.
| Test rule | Example entry | Acceptance question |
|---|---|---|
| Purpose | Associate the probe with the corresponding result correctly. | Is the relevant relationship verified? |
| Key | Unique source and target references. | No ambiguous ambiguity? |
| Reference error | Two result associations are swapped. | Does the tool detect both differences? |
| Exclusion | The comparison ignores a layout field that does not affect the decision. | Is the boundary justified for the process? |
| Report | Individual findings and processed volume. | No silent truncation or missing data? |
| Link to the run | Tools, input files, and rules are identified. | Is the statement reproducible? |
Own sample calculation: A rare error often remains invisible
Assuming a dataset contains exactly 1,000 records, one of which contains a relevant error, a randomly selected sample of 20 distinct records without replacement has a 2% probability of including the erroneous record—calculated as 20 divided by 1,000. Consequently, there is an 98% probability of missing the error. This deliberately simplified calculation does not constitute a recommendation for sample size and is not an indication of any specific actual error rate.
The case illustrates why a purely random, small sample fails to reliably capture rare critical scenarios. Supplement targeted cases based on risk assessment: historical format, unit change, corrected result, many-to-one relationship, or missing reference. A targeted selection does not address the statistical estimation question for the entire population. Clearly distinguish between these two objectives. The determination of scope and methodology is justified by the required level of assurance and its limitations.
| Method | Possible statement | Open Boundary |
|---|---|---|
| Compare row counts | The number of identified records matches | Incorrect values and reversed relationships may remain possible |
| Comprehensive comparison of specific fields | All designated fields compared according to applicable guidelines | Semantic content not included remains unverified |
| Random selection | Statement only with appropriate statistical basis | Rare critical variants may be missing |
| Targeted Risk Cases | Certain relevant error classes challenged | No generic error rate estimation |
| Reconstruct historical process flow | Required relationships and significance assessed | Scope remains explicitly limited |
Findings do not hide behind a high compliance rate
When 999 out of 1,000 compared records match, the simple similarity rate is 99.9 percent. This figure alone does not determine the suitability of the migration. A single discrepancy may impact a critical batch or result assignment. Assess error type, scope, root cause, and professional implications. A pre-determined blanket percentage value must not neutralize a critical finding without proper justification.
Before the final adoption, define what happens in case of a difference. Will the rule be corrected, the affected population re-examined or the population expanded? Check in particular systematic causes. A single detected mapping error can affect several data classes. After correction, the specific error and its possible range are checked appropriately. Accepted differences remain validly justified individually or according to clearly defined classes; they do not disappear implicitly from the denominator.
SAP Migration: Align functional responsibilities and assign service provider contracts
When transitioning from SAP ECC to S/4HANA or another SAP data migration, implementers and CSV responsibilities should apply the same object and process boundaries. Identify affected materials, batches, test records, and open activities, to the extent they are relevant in the specific operational context. Clarify which data is actively required in the target system and which historical data can be retained through an appropriate archive. The list follows your actual usage and is not a universal requirement for every SAP project.
SAP differentiates for S/4HANA Cloud Private Edition between a new implementation, system conversion, and selective data transfer. This leads to a corresponding audit focus: During a new implementation, the transferred objects are reconciled against the designated source. In the case of a system conversion, additional comparisons before and after conversion, as well as appropriate regression testing of affected GxP processes and internally developed modules, are emphasized. For selective data transfer, further justification for the selection of transferred and excluded data sets is required. This represents a methodological derivation, not a universal SAP testing checklist.
A designated migration partner should be able to explain transformation rules and actual takeover outcomes. The technical assessment shall further verify whether the data retains the required attributes in the intended downstream process. A technically correct load may nonetheless result in a professionally inaccurate assignment. Therefore, the responsibility for deciding on mapping, cleansing, exceptions, and final usage must be explicitly assigned to the contract.
An example: A material-charging combination has a lock status relevant for the use in the source system. The comparison examines key, mapping and actual target status. For the specified follow-up process, the example specifies in advance: A batch locked there should not be used, but a suitable one already released. The process case checks this agreed effect in the actual target state. A set comparison alone would not recognize a wrongly assigned status safely. No universal lock effect of a particular SAP field is claimed.
The task matrix is a standalone planning aid. A suitable proposal does not only specify the migration tool being used, but also includes required test data, internal point of contact, appropriate validation evidence, and the handling of test results. Unknown data quality should remain visibly acknowledged as an open assumption. A pre-emptive analysis may then be more appropriate than an unverified commitment regarding overall effort.
| Transfer | Contribution of the Technical Partner | Technical Assessment |
|---|---|---|
| Objects in scope | Identify the records extracted and those excluded | Clarify required active and historical usage data |
| Mapping | Explain rules, default values, and transformations | Assess significance of critical fields and their relationships |
| Cleanup and Exemptions | Provide finding lists and proposed corrections | Source correction, transformation, or justified exclusion decided |
| Verification | Identified cycles and difference reports are provided | Assess appropriate references and subsequent processes |
| Final Acceptance | Handle latest source and delta | Actually accepted and authorized target state determines |
How do you automate the validation of a SAP migration before and after the cutover?
Automatically compare volumes, key fields, and critical field values between predefined source and target states. A deliberately erroneous run demonstrates whether the comparison and technical validation can detect deviations. QA evaluates the findings for batch release.
Before the cutover, define which keys, critical values, and process decisions must be retained. According to EU GMP Annex 11, Section 4.7, validation evidence of appropriate test methods and scenarios, along with a documented assessment of suitability for automated testing tools and test environments, is required. Section 4.8 mandates testing to ensure the integrity and meaning of data during transfer to a different format or system. The following work guide demonstrates how both requirements can be substantiated for a specific system transition.
In the rehearsal, use two synthetic batches of material MAT-017 at plant 1000. Begin with source batch CH-042 set to OPEN and change it to HOLD in the final delta. Leave CH-043 at OPEN. The approved example rule maps HOLD to the target status GESPERRT (blocked) and OPEN to VERWENDBAR (available for use). Set up these expected statuses as test references independently of the loading logic.
In the isolated negative test, retain the last Delta change upon loading. This results in a flawed target state in which both CH-042 and CH-043 are deemed USABLE. The quantity and key reconciliation must confirm two out of two datasets. The field comparison must flag a deviation for CH-042 from GESPERRT to USABLE. Subsequently, during the test process, attempt to select each batch for the same consumption step. For CH-042, rejection is required; for CH-043, approval is required. If CH-042 is approved in the erroneous target state, the process test must also generate a result. The status names and review dates apply only to this example.
After correction, perform the same test batch from the specified starting condition with the specified delta again. Verify the critical fields and both process scenarios. Allow QA, independent of technical execution, to assess whether the evidence supports the release decision for the specific target outcome.
| Assessment point | Before the Cutover | After the last charge cycle |
|---|---|---|
| Data version | Secure source data with snapshot ID, date, scope, and justified exclusions. | Compare the final source dataset including the latest delta with the target snapshot of the same data scope. Link both snapshot IDs in the report. |
| Rules and Versions | Map ID, approved comparison rules, independent setpoints, and acceptance criteria to be documented. Suitability of the inspection tool and environment to be documented. | Log the actual Build-ID, Mapping-ID, and version of the comparison rules during runtime. Verify keys in both directions and critical values according to approved rules. |
| Delta and Run | Define the start and end marks of the delta. Check for a status change at the end of the delta window and a repeat during the pilot run. | Assign Delta-ID, Run-ID, and time window. Investigate missing, duplicated, or rejected changes in the run report. |
| Technical Impact | Define for sealed and usable batches. Demonstrate error detection using a deliberately faulty test setup. | Confirm access restrictions and authorized usage with a controlled test run on the specified target system. Verify test results using key, required, actual, and run ID. |
| QA Decision | Independent responsibility and release criteria to be agreed. In the example, each unexplained critical deviation blocks release. | QA evaluates the run sheet, deviations, and repeatable evidence. The documented decision refers to the reviewed target value. |
Sources: [1]
Next steps
- Capture critical data types and transformation rules.
- Define validation levels and provide justified selection criteria.
- Execute automatic rules and professional reconstructions.
- Resolve every relevant difference and issue and issue final acceptance and release.
Project experience and collaboration
Design migration checks around the errors they must detect
Before migration, we agree with the process owners what each comparison needs to establish. IT describes the transformations and relationships. The responsible specialists define expected values and specific failure cases. Comparing record counts can reveal missing records while overlooking a result assigned to the wrong record. We therefore try the proposed checks against deliberately incorrect test data. The mix of automated comparisons and judgement-based sampling depends on the data. Each unresolved difference remains visible, even when the overall match rate is high.
Frequently asked questions about Migration sampling and full reconciliation
When is an automated full comparison conclusive?
When the comparison rules actually assess the required technical attributes and their detection capability has been demonstrated. A comparison of all rows may still miss incorrect relationships, overlooked status values, or erroneous conversions. Define the scope and exclusions and validate the tool using independently defined expected outcomes. Additional technical evaluations shall be based on remaining risks; the volume of data processed alone does not determine the outcome.
Can a sample only contain normal cases?
What happens when a difference is identified?
Is there a general acceptable error rate during GxP migrations?
A general acceptable error rate is not specified by Annex 11. Section 4.8 requires verification that value and meaning are preserved during migration. Justify acceptance criteria based on the affected data and potential consequences. A non-critical deviation in a display issue should be evaluated differently from a misassignment of a result or release decision.
Why must we verify the comparison tool ourselves?
A faulty filter or missing field may generate false green results. Known positive and negative reference cases demonstrate whether the required differences are detected. Also evaluate the report and processing volume. A successful technical run does not automatically validate the correct test conclusion.
How do we document deliberately accepted differences?
Name the transformation, technical rationale, and appropriate acceptance criteria. A format change may be suitable if value and meaning are preserved. A previously unknown status difference, however, requires clarification. The decision should address specific data and impact, not merely cite a general allowable error rate.
Sample or 100 percent: What does a GxP data migration require?
The scope aligns with the required assertion and risk profile. A full comparison refers solely to the actually reviewed fields and rules. A sample depends on the selection methodology and statistical basis. In the example presented in this article, random sampling of 20 out of 1,000 identifies a single erroneous record with only a 2% probability. Combine targeted full comparisons, focused cases, and subject-matter reconstruction.
Is 99.9% similarity sufficient as a release criterion?
Not solely. Assess the nature and sequence of differences, as well as the coverage of the comparison. A single critical relationship error can be significant despite a high overall error rate. Define appropriate criteria, conduct prior investigation and verification. A general tolerable GxP error rate is not justified by this figure.
How do we engage support for an SAP data migration under GxP?
Define boundaries for objects, data version, transformation rules, and the intended downstream process. Agree on technical handover and technical alignment as separate outcomes. Assign responsibility for cleansing, mapping, exceptions, and determination of the target state. A tool name or successful load job does not fully describe this scope.
Sources and review date
Examples, selection criteria, and tables are standalone methodological aids. Justify their suitability for the specific dataset.
- European Commission: EU GMP Annex 11, Computerised Systems (2011)§ 4.8: Data transfer with retention of value and meaning; §§ 4.7 and 5 as supplementary testing and data framework. Current version as of EudraLex on 14.10.2026; no universal project or deadline specification.
- PIC/S: PI 041-1, Good Practices for Data Management and Integrity (2021)Inspection Guidance on Data Integrity; in particular electronic data, transmission, access, and review.
- SAP Learning: Customer Transition Paths to SAP S/4HANA Cloud Private EditionManufacturer description of transition paths for Private Edition. CSV impact and effort packages in the article are individual derivations.
Sources checked: .
Perform a data reconciliation exercise prior to takeover.
We support you in assessing the current situation, clarifying responsibilities, and implementing appropriate evidence within operations. The type of support that is most effective depends on your system, your team, and the specific objective.
- SAP S/4HANA validation →
Support an SAP go-live with evidence your team can defend at audit while keeping validation on schedule. We take a risk-based approach from the template to the interface for greenfield or brownfield projects and existing ECC systems.

