Fragmentation is a symptom
Enterprises rarely fragment their data deliberately. It accumulates: an acquisition, a regional system kept for regulatory reasons, a workaround that became permanent. Each decision was locally rational. The aggregate is an organization that cannot answer simple questions about itself.
The instinct is to consolidate everything. That instinct produces multi-year platform programs that deliver a warehouse nobody trusts.
Start from the decision, not the source
A more reliable approach works backwards. Identify the decisions the business makes repeatedly and expensively. Trace what information those decisions need. Fix that chain first, completely, including quality and ownership.
This produces something narrow but trusted, which is far more valuable than something comprehensive but disputed.
A number the business argues about is worse than no number at all. It converts every decision into a debate about the data.
Governance is an operating discipline
Data governance fails when it is treated as documentation. Cataloging every field in the estate produces an artifact that ages faster than it can be maintained.
Governance that works is narrower and enforced: a named owner for each critical data domain, an agreed definition for each metric that carries a decision, and a process for resolving disputes that does not require escalation to a committee.
Quality where it matters
Uniform data quality across the estate is neither achievable nor necessary. Quality standards should be proportional to consequence. The field that drives a regulatory submission and the field that populates an internal dashboard do not warrant the same investment.
What the intelligence layer needs
Three properties, in order: it must be trusted, it must be current enough for the decision it serves, and it must be reachable by the people making that decision. Most enterprise data programs optimize for completeness, which is fourth at best.