Designing customer and master-data identity across CRM and ERP
A case about identity: what “customer” means at each point in the lifecycle, how commercial accounts map to legal entities and ERP customer numbers, which system may issue or change each identity—and what survives a merge.
The case in brief
Current reality
One word—“customer”—stands for a commercial relationship, a corporate group, a legal entity, an ERP customer of one subsidiary and its delivery locations; prospects exist before any of them is validated, and names are being used as keys.
What must become true
Every identity has a clear meaning, one issuer, explicit cardinality, governed relationships, controlled merge behaviour and resolvable historical aliases—so systems agree on identity without pretending that every identity is the same record.
Design question
What does “customer” mean at each point in the lifecycle—and which system is allowed to issue or change each identity?
01The reality · current state
What the platform looks like today.
A multi-subsidiary B2B organization sells to groups with many legal entities and delivery locations, through several of its own subsidiaries, each with an ERP that numbers customers independently. One organization can buy from several of those subsidiaries, and so hold several ERP customer numbers. The CRM keeps prospects and customers in one account list. Duplicates appear, merges break history, integrations match on names, and nobody can say whether an “account” is a group, a legal entity or a ship-to address.
- ‘Account’ means a group for some teams, a legal entity for others and an address for a few
- The same legal entity has several ERP customer numbers
- Duplicates are created because matching happens after creation
- Merges in the CRM leave ERP references and opportunities pointing nowhere
- Integrations map IDs one-to-one and break when the relationship is one-to-many
- Names are used to match records across systems, and break with every abbreviation or translation
- Group-level performance is invisible because hierarchies are incomplete
02The boundary problem
What is mixed today.
“The source of truth” is too coarse when it is applied to a whole record. The CRM and the ERP were each called the master of “the customer”—and both were right about different parts of it.
- Relationship vs legal entityThe organization we sell to and the entity that signs and pays were stored as one record.
- Legal entity vs ERP customerOne legal entity buying from two of our companies needs two ERP customers—and one identity.
- Hierarchy vs identityGroup structures were used to merge records that are legally distinct.
- Prospect vs customerUnvalidated prospect data was treated with the authority of validated master data.
03Key dimension · Identity, authority & lifecycle
Five levels, three identities
Integration becomes fragile when identity is treated as “just map the ID”. Each level has its own key, its own issuer and its own cardinality—and the group is context, never a merge.
- Corporate groupThe group a legal entity belongs to—context for how we sell and report
- Key
- Group reference
- Authority
- Legal structure from master data; selling hierarchy in the CRM
- Cardinality
- One group, many legal entities
- Commercial accountThe relationship we sell to
- Key
- CRM account ID
- Authority
- CRM
- Cardinality
- One per commercial relationship
- Legal entityThe registered organization that signs and pays
- Key
- Canonical party ID
- Authority
- Master data in the ERP
- Cardinality
- One or more per commercial account
- ERP customerThat legal entity as a customer of one of our companies
- Key
- ERP customer number
- Authority
- Each ERP
- Cardinality
- One per legal entity and company
- LocationWhere goods are delivered or invoices are sent
- Key
- Address ID
- Authority
- ERP, after validation
- Cardinality
- Many per ERP customer
- Match before create
Duplicate candidates are checked on natural keys—registration number, tax ID, normalized name and country—before any record exists.
- One issuer per identifier
The canonical party ID has exactly one issuer. Every other system stores it as a reference and never generates it.
- Merge keeps history
The surviving record keeps its ID; opportunities, activities and cross-references move to it; the retired ID stays as an alias.
- Retire, don’t delete
Retired identities stay resolvable, so historical orders, invoices and reports still point somewhere.
- Correlate every message
Each cross-system request carries a correlation ID and the canonical ID, so outcomes are matched without guessing.
- Hierarchy is not identity
Parent–child links describe how we sell and report. They never merge two legal entities.
- Names are never keys
Names are abbreviated, translated and renamed. Systems match on issued identifiers and registration data, and resolve through the cross-reference.
Most integration defects in this area are identity defects in disguise. Authority belongs to business concepts—and sometimes to single attributes—not to whole records.
05Interaction design
What crosses a boundary, when—and why.
Identity travels with every message—and changes of identity are themselves business events.
| Business event | Identity | State | Who stays authoritative | Timing |
|---|---|---|---|---|
| A prospect is legally validatedCrosses: CRM → integration → ERP | CRM account ID; natural keys for matching | Proposed legal data → validated party | The ERP issues the canonical party ID; the CRM stores it as a reference | Asynchronous, with a pending state in the CRM |
| The customer buys from a second companyCrosses: CRM → integration → ERP of that company | Canonical party ID | A new ERP customer number for the same legal entity | That ERP issues its own number; the cross-reference links it | On the first order request for that company |
| Two accounts are mergedCrosses: CRM → integration → ERPs and data platform | Surviving ID and retired ID | Retired ID becomes an alias of the survivor | A person decides the merge; systems propagate it | Event after the decision |
| Legal data changes in the ERPCrosses: ERP → integration → CRM | Canonical party ID | Updated legal name, tax data or address | The ERP is authoritative; the CRM copy is read-only | Event on change |
06Failure & recovery
Rejected, late, uncertain or unavailable.
Identity errors do not fail loudly—they surface months later as wrong totals, orphaned orders and duplicate customers.
| Situation | Designed response | Who repairs it | How it is detected |
|---|---|---|---|
| A match is ambiguous | Creation stops for review instead of guessing; the candidates are shown side by side | Master-data steward | Matching on natural keys before creation |
| A merge is requested for records with open transactions | The merge waits until references are moved; the retired ID stays resolvable | Master-data steward, with the account owner | Pre-merge reference check |
| A message arrives with a retired ID | Resolved through the alias register to the surviving identity | Integration layer | Alias lookup |
| Cross-references diverge | Reconciliation compares IDs across systems; the issuer’s value wins | Integration support | Scheduled reconciliation |
07Candidate architectures
Credible options, judged against these premises.
One account equals one ERP customer
Single-entity customers and one ERP
Cost: Breaks as soon as a customer buys from two of our companiesThe CRM issues the canonical ID
The CRM is the first and only system to see every party
Cost: Prospects and legal entities mixed; the ERP must trust unvalidated dataCommercial account in the CRM, canonical party ID from master data, cross-reference in integration
Groups, several legal entities, several ERPs
Cost: Needs a cross-reference and a governed merge processDesign decision
Separate commercial identity (the CRM account) from legal and transactional identity (party and ERP customer numbers), linked through a canonical party ID issued by one authority—and assign authority per concept and attribute, allowing it to move with the lifecycle.
08Resulting architecture
Every platform, one job—and a boundary.
What each platform owns in this design, and what it deliberately does not. These are the responsibilities for this context, not universal rules.
CRMIntegration layerERPsData platform
CRM
Commercial identity- Commercial account and relationship context
- Selling hierarchy (group → buying units)
- Duplicate checks before creation
- Merge requests for commercial records
- Legal entity data after validation
- ERP customer numbering
Integration layer
Identity correlation- Cross-reference of CRM, party and ERP IDs
- Correlation IDs on every message
- Propagation of merges and retirements
- Resolution of retired IDs through aliases
- Deciding whether two records are the same entity
ERPs
Legal and transactional identity- Legal entity (party) validation
- Customer numbers per company
- Billing, payment and tax attributes
- Prospects
- The selling hierarchy
Data platform
The group-level view- Performance rolled up through the hierarchy
- Duplicate-candidate analytics
- Identity decisions or merges
09Governance & operations
Who owns the boundary over time.
Identity rules only hold if merges, aliases and exceptions have an owner.
| What | Owner | Cadence | Evidence |
|---|---|---|---|
| Matching rules and natural keys | Master-data owner | Reviewed quarterly | Duplicate candidates found after creation |
| Merge decisions | Master-data steward | Per request | Merges with open references |
| Alias and cross-reference register | Integration owner | Continuous, reconciled weekly | Unresolved IDs in messages and reports |
10The second layer
Questions that change the architecture.
Meaning
- What does an account represent—legal entity, relationship, location or group?
- When should several ERP customers map to one account—and when not?
- Is a prospect the same record as the customer it becomes?
Authority
- Who issues the canonical identity?
- Who may merge two accounts—and who decides they are the same?
- How are duplicate candidates detected, and when?
History
- What survives a merge?
- What happens to historical opportunities and orders?
- How is a retired identity still resolved?
11Decisions & outputs
What the work produces.
- 01Identity model
- 02Attribute authority by lifecycle phase
- 03Hierarchy model
- 04Duplicate rules & natural keys
- 05Merge, alias & retire policy
- 06Cross-system key register
- 07Identity quality measures