For every day spent in the field, we spend two with our heads in the documentation.
As an auditor and compliance manager, my priority has always been to make sure operations are genuinely safe. But there is one reality every professional in this sector runs into: verifying documentary compliance absorbs an enormous amount of energy.
Take a concrete example: auditing flight and duty times. We analyse the planned roster and the hours actually flown to check that FTL rules are being respected. But before getting there, you have to sit through a tedious exercise: take each requirement from AIR-OPS ORO.FTL or EU-OPS Subpart Q, open Part A, Chapter 7 of the Operations Manual, and check point by point that everything matches. Sometimes there are procedures and forms to work through on top of that.
This work is essential. But is that really where an expert adds value?
What AI does very well
Artificial intelligence has one quality that is particularly useful for document review: it never tires of comparing texts. You can ask it to analyse a large number of regulatory requirements and to search a body of documentation for the passages likely to address them. It can also structure the results, identify the associated references, flag an apparent gap in coverage and draft a justification.
As I said above, for a compliance audit this represents a considerable share of the time spent in preparation. It also demands enormous concentration, even though it is largely repetitive.
Automating that first reading does not necessarily take work away from the auditor. What it takes away is the part of the work that consumes their time without fully engaging their expertise.
What AI cannot see
A regulatory text never exists in a vacuum. Two organisations subject to the same requirement may meet it differently depending on their activity, their size, the scope of their approval, their structure or the tasks they outsource.
A procedure can look like a perfect answer to a requirement when you read it. In the field, the reality can be quite different. Conversely, an imperfect documentary match can sometimes be explained by context, or by other elements of the management system.
That is where the auditor’s real job begins. They ask the questions the document cannot settle:
- Is this applicable here?
- Is this procedure actually used?
- Is the risk adequately controlled?
- Is the evidence reliable?
- Does this situation reveal an isolated case or a systemic problem?
An AI can supply elements towards an answer. It has neither the field experience nor the professional responsibility that would allow those elements to be turned automatically into an audit conclusion.
From document search to judgement
The distinction matters. In a traditional audit, a significant share of the expert’s time goes into collecting the information their judgement requires. The aim of automation is not to remove that judgement. It is to shorten the path to it.
Picture a review covering several hundred requirements. Instead of starting from a blank page, the auditor has a first matrix setting out, for each requirement:
- the regulatory passage concerned
- the corresponding internal document
- the reference identified
- a proposed compliance status
- the justification behind that proposal
Their work changes. They no longer have to search systematically for every match; they have to examine, challenge and validate the ones put in front of them. This is no less demanding intellectually. It is, in fact, precisely where their expertise becomes most useful.
An AI must also be able to say « I don’t know »
In a regulated environment, a very convincing but incorrect answer is worth less than a clearly flagged doubt. This is why using AI in compliance cannot be conceived like a general-purpose chatbot you simply ask: « Are we compliant? »
A useful analysis has to be traceable. You must be able to go back to the original requirement, to the passage of documentation used, and to the justification for the status proposed. And when the evidence is insufficient or ambiguous, the system has to surface the doubt rather than hide it.
The value is not in artificial certainty. It is in the ability to prepare a human decision on verifiable evidence.
The paradox: more AI can mean more human
Automation is often presented as a way of replacing certain human tasks. In document-based auditing, we think it can be looked at differently.
If auditors spend less time hunting for references across several documents, they can spend more of it understanding the organisation. Questioning the teams. Analysing the root cause of a finding. Assessing risk. Checking that the procedures described are genuinely applied. Supporting corrective actions.
In other words: auditing.
The thinking behind RegUp
RegUp was not designed to declare, in the auditor’s place, that an organisation is compliant. It was built to automate part of the documentary work that comes before that conclusion. The platform checks regulatory requirements against manuals and procedures and prepares a compliance matrix with statuses and their justifications. The auditor keeps control of the analysis.
That is an important choice, particularly in sectors where a conclusion cannot be separated from its context or from the responsibility of the person who states it.
Artificial intelligence can read faster. It can compare more data. It can retrieve in moments a piece of information that would have taken a long manual search. But understanding what that information means for an organisation remains a profession.
And if AI lets auditors devote more of their time to that part, it does not take their place. It may simply give them back the one they should always have had.
