Writing an exhibitor retention rate definition two analysts can both reproduce

Two analysts, one warehouse, one show, two editions. One says the exhibitor retention rate is 70 per cent. The other says 76. They spend the reconciliation meeting reading each other's SQL and discover that neither of them made a mistake.

Everything in the gap comes from three choices, none of which was written down anywhere: what goes in the denominator, what counts as one exhibitor, and on what date you look. Settle those three in a paragraph and the definition becomes reproducible. Leave them unsettled and you will have this meeting every year.

The three disagreements

The denominator is the loudest of the three because it is the one people argue about openly. Distinct exhibiting companies present at edition N is the obvious base, and it immediately raises three questions your data may not answer cleanly. Do co-exhibitors sharing somebody else's contract count as exhibiting companies? Do pavilion participants whose space is bought by a national trade body count? Does a company that booked, paid and then failed to occupy the stand count as present?

The entity is quieter and does more damage. A registration and contract file will carry a corporate group under several legal entities, national subsidiaries and sometimes a distributor exhibiting the parent's brand. Counting entities and counting parents give different denominators, different numerators and different rates, and neither is wrong.

The measurement date is the one nobody thinks about until they have been burned. Retention is a comparison between two editions, and edition N plus 1 is a moving target for about nine months. Ask on 1 February and ask again on the show's opening morning and you get materially different answers from the same query.

What should the definition actually say?

Distinct exhibiting companies present at edition N, rolled up to parent, that are present again at edition N plus 1, measured on the opening morning of edition N plus 1, divided by distinct exhibiting companies present at edition N rolled up to parent. That paragraph is the load-bearing part of any renewal intelligence reporting, and it should be pasted at the top of the query rather than held in somebody's head.

Present means holding allocated space on the floorplan under a contract in their own name, or listed as a co-exhibitor on another company's stand. Both count, because both are exhibiting companies with a reason to come back and both can be lost. What matters is applying the same rule to both editions, since counting co-exhibitors at N and excluding them at N plus 1 manufactures churn out of nothing.

Two exclusions from the denominator are defensible. Companies dissolved, in administration, or acquired and absorbed into an existing exhibitor between the two editions had no ability to return, and leaving them in measures your exhibitors' corporate mortality rather than your retention. Whether to also exclude them is a judgement, and either choice is fine as long as you publish which you made and count them the same way every year.

Everything else stays in. The temptation to strip out "one-off" exhibitors, product-launch stands and companies sales had written off is where a definition stops being a measurement and becomes a target.

Running it on one cohort

640 distinct exhibiting companies at edition N. On the opening morning of edition N plus 1, 448 of them are on the floorplan. 448 divided by 640 is 70.0 per cent.

Now hold that cohort fixed and move only the measurement date.

At the contract cut-off of 1 February, 419 of the 640 had signed, which is 65.5 per cent. At the floorplan freeze in April, 436, or 68.1 per cent. On the opening morning, 448, or 70.0 per cent. At the close of edition N plus 1, counting late additions who took released space during build-up, 455, or 71.1 per cent.

Same cohort, same show, same query, four dates, and a range of 5.6 points. If you report the number in February and your predecessor reported it in September, your show appears to have lost four and a half points of retention that it never lost.

This is why I would fix the date at the opening morning. It is unambiguous, it is late enough to capture nearly all of the book, it is early enough that you are not waiting until the show has closed to report on the previous one, and it corresponds to a physical fact somebody could verify by walking the hall.

Rolling up to parent, and what it costs

Take the same file at parent level. The 640 legal entities resolve to 601 parents. Of those 601, 436 have at least one entity exhibiting at edition N plus 1, giving 72.5 per cent against the 70.0 per cent entity-level figure.

The 2.5 point gap is entirely accounted for by groups that returned with fewer of their entities, and this is the number I would govern with, because a parent is the thing that holds a budget and makes a decision. When a group cuts from three brand stands to two, the entity view records a churn and the parent view records a retained relationship that shrank. Both facts are true. The second one is the one a sales director should act on, and the shrinkage belongs in the space and revenue measures rather than in a count of logos.

The cost of the parent view is that it depends on identity resolution, and identity resolution is a project. If your file has one company under Acme Retail Group Inc, Acme Retail and Acme Retail Group LLC, your parent count is wrong and your retention rate inherits the error. A partial fix that takes a week is to resolve only your top 200 accounts by revenue by hand, publish the parent-level rate for those, and report the entity-level rate for the tail while the matching work proceeds.

Whichever level you choose, use the same level on both sides of the fraction. Parent denominators with entity numerators inflate the rate, and it is a mistake that survives review because both halves look correct in isolation.

The exhibition-specific problem with churn

There is a definitional issue underneath all of this that subscription businesses do not have.

An exhibitor who skips one edition and returns for the next has not necessarily churned. Companies rotate shows on a two-year cycle, sit out a year while a product is in development, or lose a budget for one cycle and get it back. A binary retained-or-lost flag computed across consecutive editions counts every one of those as a loss, and then counts them again as a new logo the following year, which flatters your acquisition figures and damages your retention figures simultaneously.

Ascarza, Neslin, Netzer and colleagues make the general version of this point in Customer Needs and Solutions in 2018, where they argue for a definition of retention that extends beyond the traditional binary retain or not retain view, and treat the measurement question as unsettled rather than as a detail. An exhibition sits squarely in that awkward middle: there is a contract, so the relationship is partly contractual, and there is no cancellation event when somebody simply does not come back.

The practical answer is to report the consecutive-edition rate as your headline, because it is simple and comparable, and to report a two-edition version alongside it: the share of edition N exhibitors present at either N plus 1 or N plus 2. On our cohort that second figure came to 487 of 640, or 76.1 per cent. The 39 accounts in the gap between 448 and 487 are your skippers, and they need a different treatment from the ones who are genuinely gone.

Why is this argument worth having?

A definitional argument feels like overhead until you attach money to a point of retention.

Reichheld and Sasser made the case in Harvard Business Review in 1990, in the article that introduced zero defections as a management idea. Cutting the defection rate by 5 per cent generated 85 per cent more profit in one bank's branch system, 50 per cent more in an insurance brokerage and 30 per cent more in an auto-service chain, and they were explicit that the magnitude varies by company and by industry even though the direction does not. The mechanism is that the acquisition cost is spent once and the margin recurs, which describes an exhibitor relationship well.

The measurement point follows directly. If a few points of retention move profit that much, then a definition that wobbles by 5.6 points depending on when you run it is not fit to steer with, and a portfolio comparing shows on inconsistent definitions is allocating attention on noise.

There is a related caution about external figures. The CEIR Index, reported through IAEE in 2025, tracks exhibitors as one of four components alongside attendees, net square feet and real revenues, and in the second quarter of 2025 those components sat between 3.7 and 15.6 per cent below their 2019 levels. Even the industry's own index needs four numbers and a stated baseline to describe one quarter. A single retention percentage quoted without its definition is not something to benchmark against.

Where this stops

A definition makes a number reproducible. It does not make it meaningful, and a perfectly specified retention rate can still be the wrong thing to look at.

Counting companies treats a 400 square metre anchor and a 9 square metre startup pod as one unit each. A show can hold 70 per cent of its logos and lose a quarter of its floor, and the retention rate will not move. The count-based measure earns its place because it is stable, easy to audit and hard to game. It should always sit next to a value weighted measure.

The other limit is that retention is a lagging measurement by construction. On the opening morning of edition N plus 1 you learn what happened over the previous year, and every decision that produced it was made months ago. The early read is the onsite rebooking rate, which is available on the Friday and answers a narrower question about one week. The definition work is still worth doing, because a lagging number you can trust is the only thing a leading indicator can be validated against.

Write the definition as one paragraph this week, covering the denominator, the entity level, the measurement date and the exclusion rule, and paste it at the top of the query. Then have a second analyst rebuild the number from the paragraph alone. If they land more than a percentage point away, the paragraph is missing a decision.