Data fragmentation is a governance problem before it is a technical one
Every organisation I work with that has a serious data fragmentation problem describes it in the same terms: data living in too many places, systems that do not talk to each other, reports that contradict each other depending on which system you pull from, and integration work that costs ten times what it should because nobody agreed on a canonical data model at the start.
The instinct is to solve this as a technical problem. Build a data platform. Consolidate the warehouses. Build better integrations. And sometimes that is necessary. But almost always, the technical fragmentation is downstream of a governance failure that was in place long before the first duplicate record appeared.
Where data fragmentation starts
Data fragmentation starts when business units are allowed to make data decisions (what to collect, how to structure it, where to store it, how to share it) without a governance framework that enforces consistency across the organisation.
This happens for understandable reasons. Business units move fast. They need data capabilities that the central IT function cannot deliver at the speed they need. They build their own. And because nobody said they could not, nobody said they had to comply with any particular standard. The result is ten years of independently reasonable decisions that produce an organisationally unreasonable outcome.
What governance needs to cover
Data ownership, not just data stewardship. Every significant data entity in your organisation needs a named business owner who is responsible for its definition, quality, and governance, not just a technical steward who is responsible for its storage. Without business ownership, data quality is always someone else’s problem.
A canonical data model for critical entities. You do not need a canonical model for everything. You need it for the data that crosses organisational boundaries: customers, products, transactions, employees. Agree the canonical definition once, enforce it at the boundary, and let downstream systems adapt.
Integration governance at the point of connection. Every new integration between systems is an opportunity to either reinforce or undermine your data model. Governance needs to be present at the integration design stage, not at the data quality audit stage six months later.
The test of whether governance is working
Ask your business what the total number of active customers is. If different systems give meaningfully different answers, you do not have a data technology problem. You have a data governance problem. Fix the governance first. The technology can follow.