Leading indicators of exhibitor churn ranked by how early they appear

Every lapsed-exhibitor post-mortem I have sat in runs the same way. The account manager says they saw it coming. The show director asks when, and the answer is somewhere between "for a while" and "since they cut back to a 24 metre corner". Then somebody points out that the final invoice went thirty days past due and everyone agrees that was the moment, and the meeting moves on.

The final invoice is a receipt. Every one of the leading indicators of exhibitor churn fired long before it, because by the time an exhibitor is sitting on an unpaid invoice for a show that has already closed, the internal decision was made months earlier, in a budget meeting you were not invited to.

The useful way to sort those signals is by how much warning each one buys, because warning is the only thing that converts a signal into an action. A signal that arrives ten days before the decision and a signal that arrives four hundred days before it are different products, and most organisers rank them by how obvious they are instead.

Lead time is the property nobody writes down

Ask a renewal team to list the things that tell them an exhibitor is at risk and you get eight or ten items in about a minute. Space cut. Stopped answering emails, which is a measurable silence with its own method behind it (G17). Complained about the location. Sent junior staff. Paid late, and the link between payment behaviour and lapsing is deep enough to have its own post (G37). Did not book the sponsorship they normally book.

Ask the same team how many days before the renewal decision each of those becomes observable and the conversation stops. The information exists in the data, nobody has ever computed it, and it is the property that decides what you can do about each signal.

Lead time is not a soft judgement. For each candidate signal you have a date it becomes true in your systems and you have a date the renewal outcome was decided, which for most shows can be approximated by the contract signature date or the rebooking deadline. Subtract one from the other, take the median across accounts, and you have the number.

How do you measure lead time on your own show?

The method is deliberately simple, because the answer is show-specific and a complicated method will not get run twice.

Pick your candidate signals and define each as a binary flag with an explicit observation date. Space reduced at the previous renewal, observable the day that renewal closed. Median days from invoice to payment up by more than twenty days, observable when the final payment lands. Fewer than half the booth staff badges registered against entitlement, observable at the registration cutoff. Decision-maker contact not registered four weeks out.

For each signal, fit a single-variable logistic regression of the churn outcome on the flag, using only accounts where the flag could have been evaluated. One regression per signal. You are not building a model here, you are measuring each signal's standalone value, and putting them in one regression together would confound them.

Then record five numbers per signal: the median lead time in days, the number of accounts the flag fires on, the churn rate among those accounts, the share of all churners the flag catches, and the base churn rate for comparison.

The table, worked

Take a show with 640 exhibitors and a churn rate of 28 per cent, which is 179 accounts lost.

Read the precision column first. Every signal lands between 42 and 56 per cent against a base rate of 28 per cent, so each one roughly doubles your prior. The best of them, the unpaid invoice at 55.8 per cent, is barely four points better than the space reduction at 52.3 per cent.

Now read the lead column. The space reduction gives you 380 days and the unpaid invoice gives you 10. They carry almost identical evidential weight and one of them arrives in time to do something and the other does not.

The cumulative picture is the part worth taking to a show director. At 380 days out you can see 46 of the eventual 179 churners, which is 26 per cent. By 190 days out the first two signals together identify 62 of them, 35 per cent. By 30 days out the union of all five reaches 118, which is 66 per cent.

Two thirds of your churn is visible a month before the decision and a quarter of it is visible a year before. What you can do at a year is a floor plan conversation, a category strategy, a pricing change or a genuine attempt to fix whatever made them cut space. What you can do at ten days is offer a discount to somebody who has already told their board they are not coming back.

Which signals belong in the table?

The candidate list should come from somewhere other than a brainstorm, and the exhibition literature has done some of the work. Anything generated while the hall is open is a separate family with a capture problem attached, and collecting it before teardown is G15's subject.

Liu and colleagues published a meta-analysis in Sustainability in 2020 covering 26 empirical studies of exhibitor satisfaction and loyalty in the Chinese exhibition market. Their finding worth acting on is that the drivers of satisfaction and the drivers of loyalty separate. Booth management, service personnel and the exhibition environment came out strongest for satisfaction, while exhibition brand was the strongest factor for loyalty. Service quality mattered more for satisfaction, perceived value mattered more for loyalty.

That separation has a direct consequence for a leading-indicator table. Signals that measure satisfaction, such as complaint volume or a poor post-show survey score, are measuring the thing that is a weaker driver of the behaviour you actually care about. Signals that proxy perceived value, such as space held, spend per square metre and how much of their own staff time an exhibitor commits, sit closer to the loyalty side. Build both, expect the value proxies to carry more weight, and let your own table settle the argument.

The commercial context also matters for how you read a signal. The UFI Global Exhibition Barometer published in January 2026, based on 378 companies across 57 countries, found 47 per cent of respondents reporting activity up by more than five per cent in their country in 2025 and 31 per cent reporting operating profit growth above ten per cent. In a market where most operators are growing, an exhibitor cutting space is doing something that goes against the market, which makes the signal stronger. In a contracting market the same signal is much weaker evidence, because everyone is doing it.

A warning and a receipt are different objects

I would formalise the distinction in the way the signals get routed, because sales teams treat everything on a risk list as the same kind of item and then work them in the order they appear.

Signals with lead times over 180 days belong to account planning. They should be reviewed once a year, in a session about categories and floor plans, with the commercial director in the room. Nobody should be phoning an exhibitor because they cut 12 square metres eleven months ago. Routing by lead time is the piece of renewal intelligence design that decides whether a signal ever produces an action.

Signals with lead times between 20 and 60 days belong to the renewal campaign. This is where a call changes the outcome, because the exhibitor's internal decision is being formed and has usually not been communicated.

Signals with lead times under 15 days belong to finance and to the post-mortem. They are worth recording, because they improve next year's model, and treating them as a call trigger produces the most demoralising conversation in the renewal cycle.

Where this stops

The lead times in your table are medians, and the distribution behind them is wide. A signal with a median lead of 30 days will have accounts where it fires at 90 and accounts where it fires at 4. Any routing rule built on the median will misroute the tails, and there is no clean fix beyond reporting the interquartile range next to the median and accepting the misrouting.

The bigger limit is that lead time is a property of when your systems learn something, not of when it happened. Booth staff registrations falling thirty days out does not mean the exhibitor decided thirty days out. They decided in a budget meeting in March and the registration behaviour is the first trace that reached you. Every number in the table measures the latency of your own instrumentation as much as anything about the exhibitor.

That is genuinely useful to know, because it points at where to add instrumentation. A signal you currently observe at 30 days that the exhibitor's own behaviour generated at 120 days is an argument for capturing something earlier, which is usually a survey question or a conversation logged properly, and not a modelling problem at all.

Start with one signal and one show. Take the accounts that cut space at your last renewal, follow them to the renewal after that, and compute the churn rate among them against the churn rate among everyone else. Two numbers, one afternoon, and you will know whether the earliest signal available to you is worth building the rest of the table around.