Skip to content

Systems and records

Carve-out shared systems export: separating records and proving it

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

Scope each shared system by its separation key: find the field that ties records to an entity, confirm every record has it, and give tables without one, such as shared customer or contact masters, an explicit rule. Then prove the export twice: it reconciles to totals the business already trusts, and an inverted filter finds none of the remaining entity's records.

System export

Carve out shared systems

This page covers carve-out shared systems export: separating records and proving it.Illustrative

Key takeaways

  • Every shared system has its own separation key and its own shared tables; read the vendor documentation for the version you run.
  • Export shared master records only through the carved-out entity's transactions that reference them.
  • Records both entities touched need a written tie-break rule and a held list, decided before extraction.
  • Completeness is proved by reconciling to independent figures such as the ledger, not by a plausible row count.
  • An inverted filter that returns zero rows is the simplest proof that nothing from the other entity slipped in.

How should shared systems be scoped in a carve-out dataset export?

Scope a shared system by its separation key, not by the deal perimeter. For each table or object you plan to export, find the field that ties a record to a legal entity, business unit or company, confirm that every record has a value in it, and treat tables with no such field as shared data that needs its own rule. Then prove the result twice: the export must be complete for the carved-out entity, reconciling to totals the business already trusts, and it must contain nothing from the entities that stay behind.

That two-sided test is what makes carve-out exports harder than they look. A filter that returns the right row count can still miss records filed under the wrong unit, and pull in customers, products and contacts that sit in shared master tables. The checklist later in this article turns both directions into checks with a pass condition and an evidence file for each.

This is general information for planning an extraction, not legal advice. Which entity may use or license each record depends on the purchase or separation agreement, any transition services agreement and the privacy terms the records were collected under, and counsel should read those documents before any export is licensed.

Where does the entity key live in each kind of shared system?

Most multi-entity business systems have an entity key, but it behaves differently in each product, and some tables are deliberately shared across every entity. Read the vendor's documentation for the version you run before writing the filter.

Where does the entity key live in each kind of shared system?
System patternSeparation keyThe trapWhat to do
Dynamics 365 finance and operationsdataAreaId on tables where SaveDataPerCompany is YesTables with SaveDataPerCompany set to No have no dataAreaId and are shared across companies; views that allow cross-company reads can return several companies' rowsList which tables are shared and filter transactional tables by dataAreaId, never views alone
Dataverse and Dynamics 365 customer engagementOwning Business Unit column on each recordWhen record ownership across business units is on, a user can own records in units other than their own, so filtering by owner misplaces recordsFilter on the owning business unit column, not on the owning user
Odoo multi-company and branchesCompany field on recordsProducts and contacts are shared by default when the Company field is emptyTreat blank-company records as shared and export them only through transactions that reference them
Ticketing and email shared by two unitsOften none: queue, brand, group or mailbox nameQueues are renamed and reassigned over the years, so a queue name is not a stable keyBuild a dated mapping of queue to entity and hold anything that falls outside it
Shared file stores and data warehousesFolder path, tag or a derived columnCopies and joins blend entities; warehouse tables may predate the entity columnTrace each warehouse table back to its source system key before using it

Shared master data: the records no filter can split

Customer, vendor, product and contact masters are often shared on purpose, and no filter will divide them cleanly. Microsoft's documentation for cross-company data sharing in Dynamics 365 describes replicating reference and group data across companies, says that records created, updated or deleted in any company within a sharing policy are replicated across all of them, and states that the feature cannot share transactional data. Customer and vendor master data sharing is a separate feature with its own limits. In practice that means a customer record can look like it belongs to the carved-out company when it was created by the parent, or the reverse.

The same documentation shows a quieter version of the problem at the table level. In its cross-company behavior article, a customer list view joins a company-specific customer table to a party table that has no company field, so party names and addresses are shared across companies even when the customer account is not.

The workable rule is to export master records only through the transactions that reference them. If the carved-out entity's invoices, orders or tickets point to a customer, export that customer's master record with them, limited to the fields those transactions need. Customers that only the remaining entities transacted with stay out, even though they sit in the same table. Record this rule in writing, because it decides which counterparties appear in the dataset at all.

Who has authority over a record that both entities touched?

Authority follows the transaction documents, and the export team needs that answer before writing filters, not after. For each record class, note which document gives the carved-out entity the right to hold and use it, which document limits that use, and which records both entities could claim, such as a customer served by both units or a ticket handled by a shared service desk.

The UK Information Commissioner's Office guidance on mergers and acquisitions tells organizations to establish what data they are transferring and the purposes it was originally obtained for, and to seek technical advice before sharing where different systems are involved, because of the risk of loss, corruption or degradation of the data. Both points apply here: the rights question decides which records may leave, and the systems question decides whether the ones that leave arrive intact.

Records that both entities touched need an explicit tie-break written down before extraction. Common choices are the entity that created the record, the entity whose contract the record was performed under, or the entity that billed for it. Pick one per record class, apply it in the filter, and keep the contested rows in a held list that counsel and the integration lead resolve.

The source separation and completeness validation checklist

Run every check for every source system in scope, and keep the evidence in a dated folder per system. A check that cannot be run is recorded as open, not skipped. The first five checks prove separation; the rest prove completeness and integrity.

The source separation and completeness validation checklist
CheckMethodPass conditionEvidence kept
Entity key coverageCount records per table where the entity key is blank or holds an unknown valueZero blanks, or every blank assigned by a documented ruleQuery text, counts and the assignment rule
Shared table inventoryList tables without an entity key and the rule applied to each (exclude, export through references, or split)Every shared table has a ruleTable list with rule and owner
Negative leak testRun the export filter inverted: count rows in the export that carry another entity's key, names or identifiersZero rowsInverted query and result
Contested recordsApply the tie-break rule and list records both entities could claimEvery contested record decided or on the held listHeld list with reason and decision owner
Mapping stabilityCompare queue, unit or company mappings at each year boundary in the export periodEvery renamed or merged unit mapped for the whole periodDated mapping table
Control totalsReconcile record counts and amounts per period to ledger or system reports for the carved-out entityDifferences explained line by lineReconciliation workbook
Period continuityCount records per month across the export windowNo unexplained empty or thin monthsMonthly count chart or table
Referential integrityTest that every foreign key in the export resolves to an exported parent recordZero orphans, or orphans listed and explainedTest output
Uniqueness and required fieldsTest primary keys for duplicates and required fields for nullsZero failuresTest output
Re-run matchRun the extraction a second time from scratch and compare counts and checksumsIdentical results, or differences traced to live changesBoth run logs

Proving completeness against numbers the business already trusts

An export is complete when it reconciles to a figure that was produced independently of the export. The strongest anchors are the ledger, period-end system reports and documents the transaction itself relied on, such as the carved-out entity's revenue and customer counts in the deal materials.

Auditors face the same problem with data a company produces, and their standard is a useful model. PCAOB AS 1105 says that when an auditor uses information produced by the company as evidence, the auditor should test the accuracy and completeness of that information, or test the controls over it. For a carve-out export, that translates into reconciling per-period counts and totals to the ledger for the carved-out entity and explaining each difference, rather than accepting a row count because it looks plausible.

Integrity checks catch what reconciliation misses. The dbt documentation describes four built-in data tests, unique, not_null, accepted_values and relationships, where relationships checks that each value in one column exists in another model, which is referential integrity. Each test is a query that selects failing rows and passes when it returns none. Whether you use dbt or plain SQL, write these tests once per table and keep their output, because a buyer's review will ask how you know nothing is missing.

Illustrative worked example: a services division leaving a shared Dynamics 365 estate

This example is invented to show the checklist in use; the counts are illustrative, not observations. A parent company sells its field services division. Both ran one Dynamics 365 finance and operations deployment, with separate legal entities, and one Dataverse environment for service cases with a business unit per division. Customer and vendor master data sharing was turned on years ago.

Illustrative worked example: a services division leaving a shared Dynamics 365 estate
StepWhat the team found (illustrative)DecisionCheck closed
Entity key coverage on sales invoicesEvery invoice line carries the division's dataAreaIdFilter by dataAreaIdEntity key coverage
Customer master3,100 customers in the shared master; 1,900 have division invoicesExport only the 1,900, limited to fields the invoices useShared table inventory
Service cases in DataverseAbout 4% of division cases are owned by business units of the parent after a reorganizationMap old units to the division for the dated period and recheckMapping stability
Shared help desk mailboxMixed requests from both divisions with no entity fieldHold the mailbox; export only threads linked to a division case numberContested records
Ledger reconciliationInvoice totals per year match the division ledger except one month with a reversed batchDocument the reversal and keep both entriesControl totals
Inverted filter12 cases with a parent customer slipped in through a shared contactRemove them and rerunNegative leak test

The exception: sources with no entity key at all

Some of the most useful operational history has no structured link to any entity. Shared mailboxes, team chat channels, file shares and recorded calls were often used by both units, and nothing in them says which company a message belongs to.

Do not assign these by inference alone. Export only items that link to a record already in scope, such as an email that cites a case number, an order number or a job number from the carved-out entity's systems, and hold the rest. A classifier or keyword rule can help find candidates, but a person should review a sample of what it assigns, and the held remainder should stay out of any licensed dataset until someone with authority decides it. This is the place where the cost of a precise export usually rises fastest, so estimate it from a sample before committing to it.

When does a separation gap mean pausing the export?

Pause the export when a gap could put the wrong entity's records into the dataset or leave the remaining entity's customers exposed, and bring in the integration lead and counsel. The usual signals are listed below.

  • A shared table that the carved-out entity's transactions depend on has no agreed rule.
  • The negative leak test returns rows that cannot be explained and removed.
  • Control totals differ by more than the team can explain line by line.
  • Records were collected under privacy terms that differ between the two entities, or the separation agreement does not mention historical data.
  • Access to the source system runs under a transition services agreement that allows operations but is silent on bulk extraction.

How does a separated export become a licensable dataset?

Once the export is separated and reconciled, the carved-out company holds evidence that most rights reviews ask for first: what is in the dataset, what was excluded and why, and how completeness was proved. SourceX can take that separated history through rights review, contracting, preparation and delivery as a licensed dataset, with the carved-out company approving the scope and terms and signing before any data is delivered. Big exports do not need to be hosted by SourceX; they can remain in the company's storage or ship on encrypted drives.

Whether the separated history could qualify is the subject of a short company assessment, and no data has to be uploaded for it.

Frequently asked questions

What is a carve-out data export?

It is the extraction of the records that belong to a business being separated from its parent, out of systems the two shared before the deal. It differs from a normal export because the filter has to exclude the remaining entity's records as carefully as it includes the carved-out entity's.

Can we just filter a shared ERP by company code?

For transactional tables that carry a company key, usually yes. Master data such as customers, vendors, products and contacts is often shared across companies, so filtering by company code alone either misses those records or pulls in records the other entity owns.

How do we prove a carve-out export is complete?

Reconcile record counts and totals per period to figures produced independently of the export, such as the ledger or period-end reports for the carved-out entity, and explain each difference. Add integrity tests for duplicates, missing required fields and orphaned references, and keep their output.

What happens to shared mailboxes and chat channels?

Export only items that link to a record already in scope, such as a message that cites a case or order number from the carved-out entity, and hold the rest until someone with authority decides. Inferring ownership from content alone is not enough without human review.

Does a transition services agreement let us extract full history?

Not necessarily. Many agreements grant operational access during the transition without addressing bulk extraction or later use of the history, so counsel should read the agreement before the export is planned.

Sources

  • Microsoft documents that Dynamics 365 cross-company data sharing replicates reference and group data among companies in a policy, and cannot share transactional data. Source
  • Microsoft documents that tables with SaveDataPerCompany set to Yes carry a dataAreaId, tables set to No are shared, and cross-company views can return records for multiple companies. Source
  • Microsoft documents that each Dataverse record has an Owning Business Unit column and that, with record ownership across business units on, users can own records in other business units. Source
  • Odoo documents that products and contacts are shared by default across the parent company and its branches unless restricted through the Company field. Source
  • The dbt documentation describes the unique, not_null, accepted_values and relationships data tests, with relationships checking referential integrity. Source
  • PCAOB AS 1105 requires auditors using information produced by the company to test its accuracy and completeness, or the controls over it. Source
  • The ICO says organizations should establish what data they transfer and its original purposes, and seek technical advice where different systems are involved because of the risk of loss, corruption or degradation. Source

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies