The audit evidence your registration file has to produce on request

The auditor's extract request is one page long and takes three weeks to satisfy. That is the pattern in most first audits, and it happens because the audit evidence your registration file has to produce was never specified while the file was being built. The person who could have specified it is your registration manager, in March, and nobody asked them.

Fieldwork itself is short. What eats the calendar is producing five or six data extracts that agree with each other on which event they describe. Everything below is knowable a year ahead, which makes this one of the few audit problems you can eliminate rather than manage.

What does the auditor ask for, and in what order?

The sequence follows UFI's Auditing Rules for the Statistics of UFI Approved Events, dated June 2021, which set out the audit as a systems appraisal, an exhibition inspection, a database review with a check-back, and the issuing of the certificate. Each stage has its own evidence.

For the systems appraisal, before the show, the auditor wants a description of the systems used to administer registrations or ticket sales, to control entrance to the exhibition, and to manage the database of registrants, plus access to the managers who run them. UFI's rules require no particular system. The test is whether the system in place, manual or automated, can generate data complying with UFI's Calculation Standards and Definitions.

For the exhibition inspection, during the show, the auditor wants the directory listing and the floor plan, and will spot check listed exhibitors against both to confirm they are present and have rented the space claimed.

For the database review, after the show, the auditor wants the registration database or the ticket sales records, and the record of actual visitor entries, and will reconcile those against the figures you are claiming. On top of that sits the check-back, which needs contactable exhibitors and visitors, so the extract has to carry usable telephone numbers and email addresses rather than a hashed identifier.

The extracts, and the field each one is judged on

Six things, in the order they tend to be requested.

The registration export. Every row, including cancelled and duplicate rows, with the state of each row and the timestamp it reached that state. An export that has already been cleaned is worse evidence than a messy one, because the auditor cannot see what was removed.

The entrance or access control log. Every read event with a timestamp, a reader identifier and the badge or ticket identifier. Not a daily summary. The daily summary is the thing being tested.

The complimentary, invitation and hosted lists. Named, with the issuing route. These are legitimate attendees under the standard and their classification is settled with the auditor rather than assumed.

The exhibitor schedule. One row per contracting party, with stand number, contracted square metres and whether the party is a main exhibitor, a co-exhibitor or a represented company. The last column decides who is countable.

The floor plan as built. Not as sold in January.

The prior edition file. The UFI Standard Audit Certificate, in the form UFI publishes in 2026, asks how many times the exhibition has been organised previously, the dates of the most recent previous editions, and whether each of those was audited. That is administrative until you discover nobody wrote it down.

Everything has to cut to the same edition

The reconciliation that fails most often is not arithmetic. It is scope.

UFI's Calculation Standards and Definitions restrict counting to the event's official opening hours, declared by the organiser. So the extracts have to be filtered to the same window, and the window is the one you declared. Halls that stayed open while a contractor finished a stand do not extend it.

Take a mid-size show with 61,204 rows in the entrance log. Filter to the declared opening hours and 1,142 rows fall outside them, which is 1.9 per cent of the file. Those are early trade access, a press preview, and a reader that kept logging during breakdown on the final afternoon. None of them are dishonest and all of them have to come out, or the declared hours have to change and be consistent with what was published.

The same discipline applies to the registration export. In a file of 24,318 rows you will typically find a small number, perhaps 87, carrying the previous edition's identifier because a duplicate record was merged across editions during a data clean. Eighty-seven rows will not move the headline. They will absolutely move a conversation with an auditor who found them before you did.

Why does the scan log always disagree with the registration export?

Because they count different objects, and the standard is explicit about the difference.

UFI's Calculation Standards and Definitions describe an entry or admission as each individual entering the site during hours officially open. A visit is the action of a person entering on an official open day with an access document, and the rule that governs the arithmetic is that a person may only be counted once per day. A visitor is counted once for the entire show regardless of how many times they came.

So one file becomes three figures. From 61,204 in-hours read events, deduplicating to one per person per day might give 38,470 visits. Deduplicating again to one per person across the whole show might give 16,905 unique visitors. The relationship is fixed by the standard: unique visitors plus repeat visits equals total visits, so the repeat visits here are 38,470 minus 16,905, or 21,565.

None of those three numbers is wrong. They answer different questions and the certificate has a specific place for two of them, which is why the visitor and visit distinction has to be settled before the extract is produced rather than after somebody has quoted a figure.

A practical consequence for the extract: hand over the raw read events and let the auditor do the deduplication, with your own version alongside for comparison. Handing over only your deduplicated output asks the auditor to certify your code.

The space schedule and the floor plan

The space figure fails audits quietly because two internal documents disagree and neither is wrong.

Suppose the signed space schedule totals 8,412 net square metres across 396 contracts, while the floor plan as built totals 8,655. The 243 square metre difference turns out to be a themed feature area, built by the organiser, occupied by a group of small producers who paid nothing.

UFI's definition settles it. Total net exhibition space is the floor space indoors and outdoors occupied by exhibitors, also called contracted space, which may include paid and unpaid space, and it includes space allocated to special shows having a direct relation to the theme of the exhibition. The certifiable figure is 8,655, and the schedule is the document that needs the extra line.

Two more details on this figure. The certificate carries indoor and outdoor square metres separately, so a show with an outdoor demonstration area has to be able to split them. And the space figure gets a national and international breakdown on the same page as the exhibitor counts, which means the exhibitor schedule has to carry a country per contracting party, taken from the address on the contract.

Build it once and version it

The retention obligation makes this worth doing properly. UFI's rules make the organiser responsible for retaining records until the successful completion of the subsequent audit, along with physical evidence such as tickets and registration documents where those exist. For a show audited every other edition, that is a two-edition horizon, and it interacts with whatever retention period your privacy notice promises. Reconcile those two before an auditor asks, because deleting the evidence on schedule is a defensible action that still leaves you unable to certify.

The version that works is a dated folder per edition containing each extract as produced, a README naming the query, the cut timestamp and the declared opening hours, and a short reconciliation sheet showing how the raw counts become the certified ones. Produce it at show close rather than on the day the request arrives. The registration database keeps changing afterwards, and the numbers you can reproduce in November will differ from the ones that existed in September.

Doing this once also changes the second audit. UFI's rules let a returning auditor accept a confirmation that systems have not changed, so a clean pack from last edition plus a short change note is most of the follow-up appraisal. That is the practical reason the choice of auditor is worth treating as a multi-edition relationship.

Where this stops

A perfect extract pack does not make a weak count strong. If your access control has 60 per cent coverage because two entrances were unstaffed on the busiest morning, the pack will document that clearly and the certified figure will be lower than the truth. Evidence discipline protects you from inflation and offers nothing against undercounting, and undercounting is the more common failure at shows with generous access.

The pack also cannot settle classification questions on its own. Whether a particular badge type counts, and under which heading, is agreed with the auditor in advance, and an extract that tags each badge with an internal label the auditor has never seen simply moves the argument later. Agree the label set, then build the tags.

This week, ask your registration platform owner for one thing: a raw entrance read log for last edition, unaggregated, with timestamps and reader identifiers. If that file can be produced in an afternoon, you are in better shape than most first-time applicants. If it cannot, you have found the work item that decides whether your first audit takes three weeks or three months, and it is worth fixing before you commit to the edition cadence the label requires or to any of the standards behind the certificate. The stage-by-stage picture of what the auditor then does with it is in what an audit checks.