Customer
The organisation or person that buys from you — at group, entity and site level, which are three different things that reporting routinely collapses into one.
01 Why it matters
What depends on getting this right
It is the denominator of the entire commercial half of the value tree. Active customers, retention, frequency, lifetime value and share of wallet all divide by it, so duplication makes churn look worse and value per customer look better than either really is.
02 Where it lives
Every system holds a different version
None of these is wrong. Each was built for a purpose and records the part of the entity that purpose needed, which is exactly why resolution is required rather than optional.
| System | What it holds of this entity |
|---|---|
| CRM | the relationship as the commercial team sees it: accounts, contacts, hierarchy |
| ERP | the sold-to, ship-to and bill-to parties, and the credit terms attached |
| E-commerce and POS | guest checkouts and card-present transactions with no account |
| Billing or core platform | who is actually invoiced, which is often a different legal party |
| Loyalty | the identity that ties transactions together across channels |
03 Match keys
What actually matches, and what only looks like it does
| Key | How well it works |
|---|---|
| Registration or tax number | The strongest key for B2B, and absent for most small customers. |
| Normalised name and address | Works well after standardisation; fails on trading names and shared serviced addresses. |
| Email and phone | Strong for B2C, but shared household addresses and role accounts (info@) create false matches. |
| Domain | Effective for B2B when personal-email domains are excluded. |
| Payment instrument | A powerful signal that must be tokenised before it is used as a key. |
04 Survivorship
When two records disagree, which value wins
Survivorship is a business decision, not a technical default. These rules should be agreed with the people who own the data and then applied consistently, because changing them later restates history.
Most recently verified wins for contact data
Addresses and phone numbers decay; recency is a better proxy for truth than source authority.
Source authority wins for legal attributes
Registered name, tax number and legal form should come from the system that is audited, not the one that is convenient.
Longest complete value for names
Truncated names in one system should not overwrite a full one in another.
Never survive a blank over a value
The most common survivorship defect: an empty field from a "newer" record wiping a good one.
05 The cost of not doing it
What stays broken while it is unresolved
Churn measured against a moving base
Duplicates inflate the customer count, so retention percentages describe an entity population that never existed.
Credit exposure understated
The same group trading under four accounts can exceed its limit without any single account breaching.
Cross-sell invisible
Products held by the same customer under different identities look like separate relationships, so share of wallet cannot be computed at all.
Marketing spends to acquire existing customers
Acquisition campaigns target people already in the base, and the cost lands in CAC.
06 Metrics that divide by it
The numbers this entity carries
07 Questions
Frequently asked
Should we resolve to the group or to the site?
Both, in a hierarchy, and report deliberately at each level. Group-level answers the commercial and credit questions; site level answers the service and logistics ones. Collapsing to one loses whichever question you did not think of first.
How good does matching have to be?
Good enough that the business trusts the number, which usually means automated matching for the confident majority and a review queue for the ambiguous remainder. A resolution process with no human step will either over-merge or under-merge, and over-merging is much harder to detect.
See the duplicates in your own data
We will resolve one entity on your systems, live, and show what the duplicates are costing.