25.3 Procurement Data Quality, Integrity and Reconciliation

Data quality changes procurement decisions

Spend analysis, supplier evaluation, inventory planning, payment controls, and performance reporting all depend on reliable data. A perfectly calculated total can still be misleading if records are duplicated, units differ, credits are omitted, or the reporting period is wrong. A public buyer should examine the meaning and origin of the data before presenting a dashboard as evidence.

Accuracy asks whether a value reflects the actual transaction. Completeness asks whether required records and fields are present. Consistency asks whether systems and definitions agree. Timeliness asks whether data are current enough for the decision. Validity asks whether the value follows the required format and permitted domain. Uniqueness asks whether the same event is counted more than once. These dimensions overlap, but passing one does not prove the others.

Establish ownership and definitions

Assign owners for vendor, contract, item, order, receipt, invoice, payment, and performance data. Define identifiers and units. A supplier's trading name can differ from its legal entity, and a parent company can have multiple legitimate subsidiaries. Merging every similar name can be as damaging as failing to identify a duplicate. Retain the original record and a documented mapping to the normalized entity.

Create a data dictionary for measures that management uses. For “on-time delivery,” define the promised date, approved revisions, actual event, partial shipments, canceled orders, time zone, and denominator. For “small-business spend,” define certification status, timing, eligible expenditure, refunds, and the program's counting rules. Without common definitions, two departments can report incompatible percentages while both believe they are correct.

Validate at entry and reconcile across systems

Entry controls can require valid contract IDs, current supplier status, approved units of measure, nonnegative quantities where appropriate, proper date order, and mandatory supporting evidence. Validation reduces errors but should not prevent legitimate exceptions such as credits or reversals; use a controlled exception workflow with reasons and review.

Reconcile the contract register to orders, orders to receipts, invoices to acceptance evidence, and payments to approved invoices and bank records. Look for invoices without orders, payments to inactive vendors, duplicated invoice numbers, unreceived items, and changes to vendor banking data. Treat flags as investigation leads, not automatic proof of fraud. Resolve discrepancies with evidence and retain the correction history.

Worked reporting example

Assume a performance report covers 100 completed orders and the disclosed metric counts each order once. Ninety were delivered by the original promised date. Five more met formally approved revised dates, while five were late even against the revisions. The original-date on-time rate is 90%. The approved-date rate is 95%. Neither is inherently the only possible measure; the report must identify which definition the contract uses and disclose revisions so schedule changes do not disappear.

Now assume the spend extract has a $10,000 invoice recorded twice and a $2,000 credit omitted. An uncorrected reported total of $112,000 becomes $100,000 after removing the duplicate and applying the credit. Preserve the source extract, exception findings, approval, and corrected result. Do not quietly change the published figure without documenting why it changed.

Protect integrity and access

Data integrity includes protection from unauthorized alteration, loss, and destruction. Use role-based access, approval controls for sensitive changes, audit logs, backups, and tested recovery appropriate to the system. Separate the ability to create a supplier, change bank details, approve an invoice, and release payment. Where staffing is limited, an independent review can compensate for some combined duties, but the residual risk must be recognized.

A vendor-master change request should be authenticated through a known independent contact path. A genuine invoice can be diverted by fraudulent bank instructions. Check legal identity, authorization, and change history before payment, and restrict unnecessary exposure of account details. Confidential data should not be pasted into an unapproved analytics or AI service to “clean it faster.”

Use analysis without overstating certainty

A chart can reveal patterns but not establish causation on its own. A price rise may reflect specification changes, volume, freight, market conditions, or a data error rather than supplier misconduct. Compare like with like, document adjustments, and test alternative explanations. Include the reporting period, exclusions, unit, source, and extraction date with important measures.

When migrating or integrating systems, reconcile record counts and financial totals, test samples against originals, verify relationships and attachments, and retain a rollback or recovery plan. A successful file transfer is not proof of a successful data migration. The buyer should be able to reproduce the published result from the preserved source and documented transformations.

For CPPB scenarios, ask whether the data actually support the decision. The correct response to inconsistent records is usually to verify, reconcile, and document before drawing a strong conclusion—not to choose the most impressive percentage or ignore an exception because the total looks plausible.