INSIGHTS · PRACTICAL GUIDE · UPDATED OCTOBER 8, 2026

Financial Data Reconciliation: Beyond a Balanced Total

A balanced total is useful evidence, but it is not enough to prove that a financial data flow is correct. Reconciliation should help explain where data went and whether the intended business meaning survived each transformation.

Start with a control matrix

For every stage, record the input population, output population, rejected records and expected transformation. Specify the comparison rule, tolerance, owner and evidence location. Any tolerance should have a business explanation rather than being chosen to make a test pass.

Check several dimensions of correctness

• Completeness: were all expected source records received?
• Uniqueness: were events loaded once according to the agreed business key?
• Amounts: do totals agree by currency and relevant business dimensions?
• Classification: are accounts, worktags and reporting attributes assigned correctly?
• Timing: are events in the intended accounting period?
• Traceability: can a result be connected to the source evidence?

An illustrative example

Suppose two source records contain 100 and -100. Dropping both leaves the grand total unchanged at zero. The amount check passes while the record count and transaction trace fail. This fictional example shows why independent checks matter.

Separate valid transformations from defects

A transformation may legitimately split one source record into several output lines or summarize many rows. Document that relationship before comparing counts. Use an event identifier or mapping table to explain the expected difference.

Review exceptions as part of the population

Track accepted, rejected and pending records together. A rejected record is not reconciled simply because it did not reach the target. Record its disposition, owner and the action needed to close it.

Make evidence reproducible

Keep the source extract identifier, run date, mapping version, comparison logic and reviewed output. Use appropriately restricted storage. A second reviewer should be able to understand the conclusion without reconstructing the project from chat messages.

Acceptance checklist

1. Agree the population and control dimensions.
2. Run representative normal and exception scenarios.
3. Document unexplained differences and assign owners.
4. Retest changes against the original acceptance criteria.
5. Obtain business review of the evidence before go-live.

These controls are project planning practices. Your finance and assurance teams determine the control requirements appropriate to your organization.

← All insights