Reporting on funds isn’t a question of willingness, but of document volume and deadlines.
The offices in charge know this precisely. The rules are known, the formats are defined, the people are competent. What’s missing is the time to verify, within the reporting windows, documentation that arrives from different implementing bodies, at different times and with different levels of completeness.
The result is work concentrated in the final weeks, with the most thorough checks reserved for a sample and the rest verified as much as the deadlines allow.
The critical points
Three, and they’re different in nature from each other.
Document completeness. Every expense requires a defined set of documents: award act, contract, invoices, payment receipts, minutes, photographic documentation where required. Verifying that everything is there, for every intervention and every implementing body, is a formal, repetitive, high-volume check.
Consistency across sources. The same data appears in different documents: the amount on the award act, the amount on the contract, the invoiced amount, the amount on the receipt, the amount declared in the monitoring system. Discrepancies are often explainable, but they have to be identified and justified. Finding them by hand, across thousands of line items, is the main bottleneck.
Traceability for audit. It’s not enough that the check was done: it has to be demonstrable how, when and on which document. Reconstructing this evidence after the fact, when the request comes from the oversight body, is work that adds to the ordinary workload.
On top of this there’s a structural difficulty: documentation is scattered across offices, implementing bodies and different archives, with the consequence that the first activity of every check is often just finding the material.
What can be verified automatically
The distinction has to be made precisely, because it determines the realistic scope of the project.
Automatically verifiable. The presence of the documents required for the type of expense. The legibility and classification of the uploaded files. The extraction of amounts, dates, references to acts and identification codes. The comparison of values extracted from the different documents of the same case, with discrepancies flagged. Consistency between dates, for example an invoice predating the award. Formal verification that mandatory elements are present in the acts, such as references to the funding program.
Requires a caseworker. The eligibility of the expense under the measure’s rules, which involves interpretation. The assessment of reasonableness. Classifying a discrepancy as a clerical error or as an irregularity. Every decision that has effects on the implementing body.
The line is the same one that applies in every administrative process: the system prepares, formally verifies and flags; the assessment and the decision stay with whoever is accountable for them.
It should be added that the system has to be built to over-flag rather than under-flag. In a control context, a false alarm costs the verification of one case; an omission costs an audit finding.
The value of continuous control
This is the point that changes the way of working the most, more than the automation itself.
Today, document control concentrates near deadlines and, for volume reasons, operates largely on a sample basis. Anomalies surface months after the documentation is uploaded, when fixing them means going back to the implementing body, requesting additional material, and waiting.
A system that verifies at the moment of upload moves the check to the start. The anomaly is flagged while the documentation is still recent, the person who produced it still has the context in mind, and the correction takes days instead of weeks.
The secondary effect is perhaps more significant than the primary one: coverage of the check moves from a sample to the entirety of cases, at least for formal checks. Not because the system is more accurate than a caseworker, but because it can examine everything.
The sample remains for substantive checks, which continue to require professional judgment. But the sample is selected better, because the formal anomalies have already been caught and can inform the choice.
Effects on audit preparation
Four indicators, to be captured against the starting state.
Time to prepare documentation for a request from the oversight body. This is the indicator most directly felt by the office.
Number of findings for incomplete or inconsistent documentation. This should be read together with the next one, because the two move in opposite directions in the initial phase.
Number of anomalies caught at upload time. An increase is a positive result, not a negative one: it means they’re being found earlier.
Actual coverage of formal control, meaning the percentage of documentation verified out of the total, instead of the sampled percentage.
One element that can’t be reduced to an indicator: the immediate availability of documentary evidence for every check. It changes the nature of audit preparation, from reconstruction to extraction.
A note on how to read the second and third indicators in the first few months. More systematic control increases the number of anomalies recorded while decreasing the ones that reach the oversight body. If the two figures are communicated separately, the first can be read as a decline in documentation quality, when it actually measures the effectiveness of the check.
What this approach doesn’t solve
It doesn’t change the eligibility rules or the reporting deadlines.
It doesn’t replace the relationship with implementing bodies. A system flags that a document is missing; obtaining it remains a relational activity, and in many cases that’s where the actual time goes.
It doesn’t solve the dispersion of archives. If the documentation resides in disconnected systems, the first job is discovery and integration, and it has to be planned as a separate activity.
And it isn’t suited to low-volume measures or documentation that can’t be standardized. Repetitiveness is the condition that makes the setup and maintenance cost sustainable.
Finally, the general point about input data quality still applies: on illegible scans or documents uploaded in unusable formats, automatic extraction doesn’t produce reliable results. Defining minimum requirements for uploaded files is part of the project.
In closing
The path we recommend starts from a single measure or a high-volume expense type, with standardized documentation and an identified owner, measuring the starting state before launch.
If you’re dealing with this on a specific program, we’re available for a technical conversation: which documents are involved, which checks would be automatable, and which indicators would be realistic to measure by the next reporting window.
