When Finance says 16% and Delivery says 18%, the instinct is to look for an error. That is usually the wrong place to look, and it wastes the first week.

In practice there are four causes, and they are worth checking in this order because they run from most to least common.

1. Different definitions

The most common cause by a wide margin. Both calculations are correct; they are calculating different things.

Fully loaded cost may or may not include benefits load, bench time, unbilled overtime, or allocated overhead. Each inclusion is a defensible choice. Two teams making different defensible choices produce different correct answers.

How to spot it: ask each team to write the definition in one sentence, separately, without conferring. Compare the sentences before comparing the numbers.

2. Different grain

Both use the same definition, at different levels of aggregation, and the aggregation is not commutative.

Margin averaged across projects is not the same as total revenue over total cost. Weighting differs. A firm with one large low-margin project and nine small high-margin ones gets substantially different answers depending on which way it is computed — and both are legitimate.

How to spot it: ask what the denominator is.

3. Different period boundaries

Finance closes on a calendar month. Delivery may work in four-week cycles. Payroll may run semi-monthly. Revenue may be recognized on a different basis from when the work occurred.

Three teams reporting "last month" may be describing three overlapping but non-identical windows.

How to spot it: ask for the exact start and end dates, not the period name.

4. Actual data error

It happens, and it is the least common of the four. Duplicated records, a failed load, an orphaned key after a system change, a mapping that was never updated after a reorganization.

How to spot it: it is the residual after the first three are excluded — which is why checking it first wastes time.

Why this keeps recurring

Each individual instance gets resolved in a meeting. The resolution is not written down, or it is written in an email that nobody can find in April. The next time the question arises, a different analyst reconstructs a different answer, and the discrepancy returns.

The recurrence, not the discrepancy, is the actual problem. An organization that resolves the same definitional dispute three times a year is paying for the same decision repeatedly and getting no compounding benefit from it.

The structural fix

A written definition, a stated grain, an explicit period basis, and a named owner who decides. Once those exist, the dispute takes four minutes and produces a citation rather than a debate.

Practical check

Keep the first comparison deliberately small. Put the two definitions, grains, periods, and exclusion rules side by side before examining any records. Most disputes become visible in that single page. Data-level investigation should begin only after those four rows match and the numbers still differ.

Certified metrics with named owners are how a Management Intelligence Build makes that permanent.