Preparing for a technical or regulatory audit: arrive with a history, not a pile of PDFs

The week before an audit looks much the same from one site to the next: scattered reports are pulled together at the last minute, histories are reconstructed from memory, answers are drafted for questions that are poorly anticipated. Two days of work to recover what already existed, somewhere, without being accessible. The audit then goes more or less well, but the cost is the same either way: an exhausting sprint to prove that you are in control, when control ought to prove itself.

A well-prepared audit does not demand more work beforehand. It demands a history kept throughout the year, one that answers the questions without anyone having to reconstruct them.

What the auditor is really looking for

You have to let go of one idea: the auditor does not count documents. They check three things, and a stack of PDFs proves none of the three.

First, they check that you know the condition of your equipment. Not that you hold reports, but that you know what those reports contain: which items are degrading, which are approaching a threshold, which had a recent finding. A manager who has to open ten reports to answer a question about one item gives exactly the opposite impression of control. Knowing your asset base means being able to speak to it without leafing through it.

Next, they check that your inspections took place at the right intervals, and that those intervals are justified. This is the heart of the matter in a regulated environment: an overdue inspection, or an interval you cannot explain, are findings. The way those intervals are determined is covered in the inspection interval for equipment. It is not enough for the date to be met, you need to be able to say why that date and not another.

Finally, they check that your decisions are traced. Why this item was kept in service, why a given interval was tightened or deferred, and on what basis. A sound but undocumented decision is treated as a decision never taken. What counts is not only having been right, it is being able to show why. In an audit, the reasoning that stayed in someone's head simply does not exist.

The difference is settled across the year, not the week before

The instinct to prepare for the audit in the days beforehand is the symptom of a history that is not being kept. When each report is filed and put to use as it arrives, there is nothing to prepare: the history is already ready.

This assumes the reports are not left sleeping in email attachments or in binders sorted by contractor, but are centralised and made usable as they come in, a subject developed in centralising a plant's inspection reports. A report that has been received, read, and whose values are filed by equipment is a report that will demand no further work on the day of the audit. The effort is spread thin across the year rather than piled into one dreaded week.

This also assumes that decisions are traced at the moment they are taken, not reconstructed afterwards. A one-line note at the time of a judgement call, explaining why an interval was tightened, is worth more than a reconstruction from memory six months later. The trace taken in the moment is credible; the reconstructed trace never quite is. An auditor can tell the difference, and so can anyone who has sat across the table from one.

What automatic reading of reports changes

This is the point where AI has a concrete effect on audit preparation, provided you stay precise about what it actually does.

It does not sit the audit for you. It makes the history queryable. When reports are read and their data extracted as they come in, retrieving every finding on one item, every measurement at one condition monitoring location, every intervention tied to a single cause becomes immediate. The auditor's question finds its answer while it is still being asked, instead of opening a search. This is exactly the groundwork described in moving from PDF reports to usable data.

Nor does it replace judgement. An extracted value must stay verifiable, and that is precisely what an auditor values: being able to trace back from the displayed figure to the original report. A history that shows a wall thickness without letting you find the source document inspires less confidence than a well-kept binder. The data prepared by the machine must always point to its source, for the reasons developed in what AI decides and does not decide. A figure you cannot trace is a claim, not evidence.

Preparation that holds upstream

• Preparing for the audit the week before.This is the symptom of a history that is not kept. A site that is up to date has nothing to prepare in a rush.
• Confusing holding reports with knowing your asset base.A stack of PDFs does not prove control. The auditor checks what you know, not what you store.
• Reconstructing decisions after the fact.A justification taken in the moment is credible; a reconstruction from memory never quite is.
• Displaying data with no link to the source.A value with no document behind it inspires less confidence than a well-kept binder. Tracing back to the source is the proof.
• Discovering overdue intervals in front of the auditor.Reviewing them upstream lets you correct what can be corrected before it becomes a finding.