Every analytical capability an entity builds — anomaly detection, forecasting, performance reporting, public transparency — rests on the same underlying records. When those records are inconsistent, the failure surfaces at the analytical layer, where it is expensive to diagnose and easy to misattribute to the analysis. Treating data quality as infrastructure means funding and governing it independently of any particular project.
Six dimensions, and the ones that matter in finance
Standard data quality frameworks list completeness, accuracy, consistency, timeliness, validity and uniqueness. In a public finance context three dominate.
Consistency is usually the binding constraint. The same vendor appears under four spellings; the same programme is coded differently by two departments; a fund is renamed mid-year without the historical records being restated. None of these is an error in any individual record, and all of them break aggregation.
Uniqueness follows, mostly in master data: duplicate vendors, duplicate customer accounts, duplicate asset records.
Timeliness determines whether the data supports oversight or only reporting. A general ledger that is reliable ninety days after month end supports an audit and not a decision.
Master data is where to start
Transactional data quality is largely determined by master data quality. Four masters carry most of the risk:
- Chart of accounts. Structure, active/inactive status, and the mapping to reporting categories. Accounts created ad hoc without a documented owner accumulate, and the resulting structure cannot be aggregated reliably.
- Vendor master. Covered in detail elsewhere; the control and analytical arguments coincide.
- Employee and position master. Position control, funding source allocation, and the linkage between HR and payroll records.
- Asset register. Existence, location, condition, and the linkage to the maintenance system.
Assigning each a named steward with authority over additions and changes is the single highest-return governance act available, and it costs nothing beyond the decision.
Measure quality where it is created, not where it is consumed
Publishing a monthly scorecard by originating department — records rejected, exceptions raised, fields incomplete, days to correction — moves accountability to the point of entry. Dashboards that report aggregate quality tell the analytics team what it already knows and give nobody a reason to act.
Validation at entry
Corrections downstream cost a multiple of prevention at entry, and in a finance system many corrections require a journal entry with its own approval trail. Practical controls: mandatory fields enforced by the system rather than by policy; reference lists rather than free text wherever a controlled vocabulary exists; format validation on identifiers; duplicate checking on creation; and cross-field consistency rules. None of these is technically demanding, and most ERP systems support them out of the box while being deployed with them switched off to reduce friction during implementation.
Reconciliation as an ongoing control
Automated reconciliations between systems — payroll to general ledger, sub-ledger to control account, bank to cash, grants system to expenditure detail — should run on a schedule with exceptions routed to a named owner and a resolution deadline. The value is not only in catching errors but in the trend: a reconciliation whose exception count is rising indicates a process change somewhere upstream that nobody reported.
Lineage and documentation
For every published figure, an entity should be able to state its source system, the extraction logic, the transformations applied, and who owns each step. Where that documentation does not exist, a personnel change removes the entity's ability to explain its own reporting. Lineage documentation is unglamorous and is the difference between a reporting capability and a set of spreadsheets that happen to work.
Funding it
Data quality work rarely wins a budget contest against a visible service. The argument that tends to succeed is specific rather than general: the duplicate payments recovered, the staff hours consumed by manual reconciliation, the audit adjustments traced to data issues, the delay in the close. Quantifying two or three of these produces a case that survives review; an appeal to data as an asset does not.
This publication is general information and is not legal, accounting, audit or financial advice. See our Disclaimer. Found an error? Write to [email protected] — we correct in place and note what changed.