The Myth of the Single Source of Truth

The persistent promise

For as long as enterprise data management has existed, “single source of truth” has been the goal. One authoritative dataset. One definition of “customer.” One version of “revenue.” One canonical record that everyone uses and everyone trusts.

Every data strategy document we’ve reviewed includes some version of this aspiration. And every organisation we’ve studied has failed to achieve it — not because the technology failed, but because the concept itself is architecturally incompatible with how organisations actually work.

Why organisations don’t have single truths

Multiple legitimate definitions

“Customer” means something different to Sales (someone who might buy), Marketing (someone who might engage), Finance (someone who pays), Support (someone who needs help), and Legal (a contractual counterparty). Each definition is legitimate. Each serves a real operational need. And they’re not reconcilable into a single definition without destroying the utility of at least some of them.

A “single source of truth” for customer data requires choosing one definition and declaring the others subordinate. The chosen definition serves the team that owns it. Every other team gets data that doesn’t quite fit their reality. The “truth” is singular only from one perspective.

Temporal multiplicity

“Revenue” at any given moment exists in multiple legitimate states: booked, recognised, billed, collected, projected, adjusted. Finance sees recognised revenue. Sales sees booked revenue. The CEO’s board report shows projected revenue. The auditor sees adjusted revenue.

None of these is wrong. Each is a different view of the same underlying reality, taken at a different point in the revenue lifecycle. A single source of truth that captures one view necessarily misrepresents the others. The organisation doesn’t have one truth about revenue — it has several, and they coexist by design.

Contextual accuracy

Data that’s accurate in one context may be misleading in another. A product margin calculated at the corporate level (averaging across all customers) tells a different story than the same margin calculated at the customer segment level (which reveals that high-value customers subsidise unprofitable ones). Both calculations are correct. They answer different questions.

A single source of truth that presents corporate-level margins is accurate but potentially misleading for segment-level decisions. One that presents segment-level margins is useful for targeting but misleading for aggregate financial reporting. There is no single level of analysis that serves all decision-making contexts.

The aspiration for a single source of truth assumes that truth is singular. In complex organisations, truth is perspectival. The answer depends on who’s asking, why they’re asking, and what they need to do with the answer.

What actually works

The organisations that manage data effectively don’t pursue a single source of truth. They build something more nuanced:

Authoritative sources with transparent context. Each dataset has a clearly identified owner, a documented definition, a stated scope, and explicit limitations. Users know what the data represents and what it doesn’t. There’s no pretence that it’s the only truth — but there’s clarity about whose truth it is.

Reconciliation protocols. When different datasets tell different stories (and they will), there’s a defined process for understanding why. Not to force agreement, but to make the differences visible and interpretable. The marketing team’s customer count is different from the finance team’s because they define “customer” differently. That’s not a data quality problem. It’s a business definition problem. The reconciliation protocol makes the difference explicit.

Shared identifiers, not shared records. Rather than unifying all data into a single record, link data across systems using shared identifiers while allowing each system to maintain its own contextual representation. The customer ID is shared. The customer record is not. Each system holds the representation that serves its operational needs.

Decision-specific views. For specific decisions, curate a view that combines data from multiple sources in the way that best serves that decision. Don’t pretend it’s the universal truth — label it as a decision-specific composition and document the sources, transformations, and assumptions that went into it.

The myth of the single source of truth has persisted for decades because it’s an appealing simplification. But simplification isn’t free. In this case, the cost is either an unachievable goal that consumes resources indefinitely, or a false unity that serves one perspective while distorting all others.

The alternative is messier. Multiple authoritative sources, explicit reconciliation, transparent context. It lacks the elegance of a single source. But it has the considerable advantage of being possible.