Data Models
Applying statistical analysis to business data.
Workflow
- 01Data Collection
- 02Model Building
- 03Data Analysis
- 04Insight Generation
If you change this
It implies 4 topics across one module.
See the path ahead →Why it matters
Every report, integration and migration depends on the model. It is also the artefact most likely to be produced by technical people alone and then handed to a business that does not recognise itself in it.
- It encodes meaning: If the model says a customer is an account, then a household with three accounts is three customers.
- It constrains reporting: You can only report what the model can express.
- It is expensive to change: Once data exists in a shape, restating it is a migration.
The workflow
- 01
Concept definition
The business entities and what each one means, in business language.
- 02
Relationships
How entities relate, including the awkward ones — hierarchies, many-to-many, effective dating.
- 03
Attributes
What is held about each entity, and which attributes are mandatory.
- 04
Ownership
Who owns each entity and can approve a change.
- 05
Physical design
The technical implementation of the agreed model.
- 06
Change control
How the model evolves without breaking what depends on it.
Model the awkward cases first. Hierarchies, effective dating and many-to-many relationships are where models break, and they are always discovered after the simple entities are agreed.
What you need in place
- Business-language definitions: Recognisable to the people who use the terms.
- An owner per entity: With authority to settle disputes about meaning.
- Effective dating where needed: Structures that change over time.
- A change process: Because the model will change.
- Traceability to reporting: So the impact of a change is knowable.
Questions to ask
01Language
Would the business recognise these definitions?
If the model is in technical language, it was not agreed, only published.
02Hierarchies
How are org, product and customer hierarchies modelled over time?
This is where most models fail.
03Ownership
Who settles a dispute about what a term means?
Without an arbiter, both definitions persist.
04Change
What happens when a definition needs to change?
If the answer is a project, expect workarounds instead.
05Edge cases
Have we modelled the exceptions, or only the clean cases?
Exceptions are where the data actually lives.
Metrics and KPIs to track
Entity ownership coverage
Entities with a named owner.
Model change volume
Changes per period, and time to approve.
Definition disputes
Raised and resolved, a measure of clarity.
Unmodelled workarounds
Data held outside the model to work around it.
Alongside this