Running a registration approval workflow that does not stall the campaign

An exhibitor rings the week before doors. She invited her biggest account, they registered, and they have heard nothing. You look up the record and it is sitting in the vetting queue, submitted eighteen days ago, still pending, behind 1,400 others.

Nobody made a mistake. A registration approval workflow has a review rate, the review rate is lower than the arrival rate, and the arithmetic has been running against you since about week nine. What makes it hard to see is that the queue is normally reported as a single number, the approval rate, and an approval rate looks perfectly healthy right up until the moment it is being measured on decisions that arrive too late to matter.

A vetting queue is a funnel

Treat it the way you would treat any other funnel, with stages, volumes and elapsed time between them.

A registration on a vetted show passes through submitted, in review, decided, and communicated. Then it either becomes a live registration or it does not. Every one of those transitions has a timestamp, and almost nobody stores them, which is why the queue is usually invisible until an exhibitor escalates.

Store four timestamps per record: submitted, first opened by a reviewer, decided, and decision sent. From those you get every number worth having, including the one that matters most, which is the gap between submitted and decision sent.

Three numbers that describe the queue

Time to decision. Report the median and the 90th percentile, never the mean, because the mean is dragged around by a handful of records that sat for two months and it will make a broken queue look fine.

Approval rate. Approvals divided by decisions, not by submissions. The difference matters, because records that never reached a decision are not rejections and folding them in makes both numbers wrong.

Decisions arriving after they mattered. The count of approvals issued more than some number of days after submission, and separately the count still pending at show open. This is the number nobody publishes and the only one that connects the queue to lost attendance.

The vetted shows that do this well publish the timeframe up front, which is a discipline worth copying whether or not you hit it. NY NOW's buyer qualification documentation for its 2026 Summer Market tells applicants to allow ten business days from uploading their documentation for credentials to be reviewed and a confirmation with a barcode to be sent. It also names 30 July 2026 as the last day to submit credentials, after which anyone registering online has to bring them for on site entry. Two weeks of calendar time, published as a commitment, with a dated fallback for people who arrive too late for the queue to serve them. Both halves of that are the design.

Working one edition's queue

Take a show that put 9,400 registrations through vetting across the campaign.

Of those, 8,900 reached a decision before show open and 500 did not. Of the 8,900 decisions, 6,319 were approvals, which is an approval rate of 71.0 per cent, and 2,581 were rejections. The median time to decision was 4.2 days and the 90th percentile was 11 days.

Read as three numbers, that is a functioning queue. A 71 per cent approval rate on a vetted show is unremarkable, and a 4.2 day median sounds responsive.

Now split the approvals by how long they took. Of the 6,319 approvals, 3,410 were decided inside 48 hours, 1,729 took between two and seven days, and 1,180 took seven days or more.

Turnout across those three groups was 76 per cent, 68 per cent and 59 per cent. The 17 point spread between the fastest and slowest group, applied to the 1,180 slow approvals, is 201 people who held a valid badge and did not walk in.

Add the 500 who never received a decision. Some of those would not have attended anyway, and all of them were people who filled in a form and got nothing back.

What does a slow decision actually cost?

The turnout gap above is a correlation and it needs saying plainly, because it is the number most likely to be over-read. Records that take longer to decide are not a random sample. They are disproportionately the ones with unclear credentials, unfamiliar company types and thin documentation, and those registrants might have turned out at a lower rate regardless of how fast you answered.

What you can say without any causal claim is the second piece of arithmetic, and it is the one that changes the design. Maritz's Registration Insights Report 2024, built from more than 360,000 attendee registration records across 30 trade shows over three years, put 22 per cent of registrations in the final week before the show and nine per cent on site.

Apply that shape to the 9,400 submissions. Twenty-two per cent is 2,068 registrations arriving inside the last seven days. Against a published ten business day review window, every one of those has a scheduled decision date after the show has closed. Against this show's actual 90th percentile of 11 days, roughly a tenth of them are in the same position.

That is not a service level problem. It is a structural collision between when people register and when the queue can answer, and no amount of reviewer effort fixes it, because the reviewers are already working. The queue has to be designed around the arrival curve or the last fortnight of the campaign has to be handled by something other than the queue.

Designing the queue around when people arrive

Three changes, in the order I would make them.

Auto-approve the population you would approve anyway. On most vetted shows a large share of submissions are trivially fine: a returning attendee from a company already in your approved list, at a job function already on the criteria you sold exhibitors on, with a corporate domain matching that company. Write that as a rule, apply it at submission, and the record is decided in seconds. If that rule covers half the file, the human queue halves and its median collapses. Sample the auto-approved population weekly and review a hundred by hand to check the rule has not drifted.

Move reviewer capacity to match the curve. Most teams staff the queue evenly and then find the last three weeks unmanageable. Look at your own arrival curve, work out where the volume actually lands, and put the capacity there. This costs nothing if the total hours are the same.

Publish a fallback for the final window, and make it a real one. After a stated cutoff, stop promising a pre-show decision and route registrants to on site credential checking instead. That converts an unanswerable queue into a staffing decision at the door, which is a problem your operations team already knows how to solve. It also stops you failing quietly, which is the worst of the available outcomes because the registrant does not know they need to do anything.

There is a fourth change worth making at the same time, and it costs a dropdown. Record a reason code on every rejection, from a short closed list written before the campaign: credentials not supplied, credentials insufficient, company type excluded, job function excluded, duplicate of an existing record, suspected fabricated record. Six codes, no free text, mandatory on every rejection. That last code should be fed by the automated screen for fabricated records instead of a reviewer's judgement, since the evidence for it is machine readable.

The reason to bother is that a rejection rate without reasons is unimprovable. If 2,581 rejections turn out to be 1,700 credentials not supplied, the problem is your instructions rather than your audience, and the fix is a clearer upload page rather than a looser rule. If they are mostly company type excluded, the campaign is buying the wrong people and the media plan is the thing to change. Those are opposite conclusions drawn from the same headline number, and without the codes you will pick between them by argument.

What should you tell the exhibitor?

The exhibitor in the opening paragraph does not want your median. She wants to know whether her account is coming.

Give exhibitors a view of the guest registrations they generated with the state on each one: pending, approved, rejected, and how long it has been sitting. Nothing else in this post will do as much for the relationship, because it converts an invisible internal process into something they can chase, and it moves the escalation from a phone call in show week to a nudge in week six.

It also produces a useful side effect. Exhibitors will start telling you which rejections were wrong, in specific terms, with the account name attached. That feedback is the cheapest evaluation data your criteria will ever get.

Where the measurement misleads

The approval rate is not a quality measure and it will be read as one. A rate of 71 per cent tells you how many applicants matched your criteria. Whether the criteria were right is a different question, and the queue cannot answer it because it never sees who was deterred from applying.

The queue is also blind to the people who abandoned partway. Somebody who reached the credential upload step, discovered they needed two recent invoices and a lease, and closed the tab never enters the funnel at all. Your submitted count is measured after that filter, so a stricter credential list improves your approval rate and shrinks your audience at the same time, and both effects show up as good news in this report. Measuring that loss means instrumenting the questions the form asks before submission, which sits upstream of everything here and belongs in the same attendee analytics pack.

The last limit is that time to decision is easy to game and will be gamed the moment it becomes a target. A reviewer under pressure clears the ambiguous ones as rejections because rejecting is faster than investigating. The rate holds, the median improves, and the show quietly stops admitting exactly the buyers whose paperwork is unusual, which on most shows means the smaller and newer firms.

This week, pull the submitted and decided timestamps for the last completed edition and plot two lines on one chart: submissions per week and decisions per week. The point where those lines cross, and the area between them after that point, is the backlog that produced the phone call. If you cannot draw the chart because the timestamps are not stored, that is the thing to fix before the next campaign opens.