These two terms are used interchangeably in most organizations, and the conflation causes a specific, recurring failure.

A system of record is where a piece of data is created and maintained. Payroll is the system of record for salary. The CRM is the system of record for opportunity stage.

A source of truth is where the organization has agreed to look for an answer, for a given question, at a given grain.

They are frequently different, and that is not a defect.

Why they diverge legitimately

Headcount is created in the HRIS. But for a question about cost, the finance system's view — which includes contractors on the payroll and excludes unpaid leave — may be the right source of truth even though the HRIS is the system of record.

Neither is wrong. They answer different questions. The error is assuming that the system where data originates is automatically the place to answer every question involving it.

The failure this produces

Without the distinction, every reconciliation becomes an argument about which system is "right." That argument is unwinnable, because both systems are correctly recording what they were built to record.

Teams then either reconcile endlessly, or pick a winner and quietly make the other view unavailable — which breaks the questions that legitimately needed it.

The rule that resolves it

Source of truth is declared per metric, per grain — not per system.

Illustratively:

MetricSystem of recordSource of truthWhy
Headcount for cost analysisHRISERP payroll viewIncludes contractors carrying cost
Headcount for capacity planningHRISHRISReflects who can actually be assigned
Project revenueERPERPSingle origin
Project scope at saleCRMCRMOnly place the promise is recorded

Note that headcount appears twice with two different sources of truth, deliberately. That is correct, and writing it down is what stops it becoming an argument.

What to do with it

For each certified metric, record both — the system of record and the declared source of truth — and the reason they differ where they do. It takes one extra column and eliminates an entire category of recurring dispute.

Where the distinction earns its keep

During a system migration. The system of record changes; the questions do not. If source of truth was declared per metric, you know exactly which answers are affected by the change and which are not. If it was never declared, every metric is potentially affected and the whole reporting estate has to be re-verified.

When a new system arrives alongside an old one. A new planning tool holds a version of headcount. Without a declaration, three systems now hold headcount and every report becomes a negotiation.

When someone disputes a number. The declaration converts an argument about which system is right into a lookup: for this metric, at this grain, we agreed to look here, and this person decided that.

The common objection

"Surely there should just be one source of truth."

That is the goal for a piece of data, and it is often achievable. It is not the goal for a question, because different questions legitimately need different views of the same underlying data.

Forcing a single source of truth per system means one of the two questions above gets a wrong answer — usually silently, and usually the capacity question, because finance systems tend to win these arguments.

The discipline is not to eliminate the divergence. It is to make it deliberate, documented, and owned.

Declaring source of truth per metric is part of the entity work in a Management Intelligence Build.