Counting service tickets as renewal risk without punishing your largest exhibitors

The operations debrief has a slide called top twenty by ticket volume. At the top is your second-largest exhibitor, 240 square metres, eighteen tickets. Somebody says they must be furious. The account manager, who has spoken to them four times since the show, says they are fine and are asking about the island position for next year.

Both statements are true, and that slide is where counting service tickets as renewal risk goes wrong. Eighteen tickets from a 240 metre island with a two-storey build, three power drops and a demo kitchen is an exhibitor using the service desk the way it is designed to be used. The list is sorted by how much stand there is, and everyone in the room is reading it as a list of unhappy customers.

Service data is genuinely predictive of renewal. Raw counts are close to useless for it, and the fix takes about an hour.

Why does ticket volume rank your biggest exhibitors first?

The reason the league table ranks by size is arithmetic. Tickets get raised per problem, problems arise per installed thing, and installed things scale with square metres and build complexity. A 9 square metre shell scheme has one carpet, one fascia and one socket. There are only so many things that can go wrong.

So the count is a size proxy wearing an emotional label. Building a renewal risk input on it produces a risk list that correlates with revenue, and any process fed by that list will spend its effort on the accounts that were always going to get attention while missing the 18 metre exhibitor on their second edition who raised one ticket, never got an answer, and is not coming back.

There is a second and quieter problem. Ticket volume is partly a measure of how easy you have made it to raise a ticket. An exhibitor with a dedicated account manager who fixes things by text message generates no tickets at all. If you run a manned service desk in one hall and a portal-only process in another, your ticket rates are measuring your own operating model.

Normalise by exposure, then deal with what that breaks

The obvious correction is tickets per 100 square metres. It works, and on its own it creates a mirror-image problem at the other end of the size distribution.

An 18 square metre exhibitor who raises a single ticket about a missing chair scores 5.6 tickets per 100 square metres. That is four times the typical rate on most shows, produced by one chair. Small exhibitors have a small denominator, so their normalised rate is dominated by whether a single event happened.

The standard treatment is to shrink each account's rate toward the show median, weighting by how much exposure the account has. Adjusted rate equals tickets plus a smoothing constant times the show median, all divided by exposure in hundreds of square metres plus the same smoothing constant. Setting the constant to 1.0 means an account gets treated as though it also carried 100 square metres of average-behaving stand.

Take a show whose median is 1.3 tickets per 100 square metres.

The 240 metre exhibitor has 2.4 units of exposure and 18 tickets. Raw rate 7.50. Adjusted, that is 18 plus 1.3, over 2.4 plus 1.0, which is 19.3 over 3.4, giving 5.68.

An 18 metre exhibitor with one ticket has 0.18 exposure. Raw rate 5.56, which looked almost as bad. Adjusted, 1 plus 1.3 over 0.18 plus 1.0, which is 2.3 over 1.18, giving 1.95. Essentially the show median, which is the correct reading of one chair.

A 54 metre exhibitor with four tickets has 0.54 exposure. Raw 7.41. Adjusted, 5.3 over 1.54, giving 3.44.

Raw rates put all three within two points of each other. Adjusted rates separate them into 5.68, 3.44 and 1.95, which matches what anyone who worked that hall would tell you.

Which measures beat the count?

Volume, even normalised, is the weakest of the four things your service system knows. The others are about how the tickets resolved.

Share reopened. A ticket that gets closed and comes back was closed without being fixed. This is the cleanest available measure of service failure, it is independent of size, and most helpdesk systems already record it.

Median days to close. Compute per account and compare against the show median, using working days if your desk is not staffed continuously. During show week the unit should be hours.

Count escalated past the account manager. An exhibitor who contacts your operations director, your show director or anybody with a title has stopped believing the normal route works. This is rare, it is recorded inconsistently, and it is worth the effort of recording it properly.

Run those across the same three exhibitors. The 240 metre account has 18 tickets, 5 of them reopened, a median close time of 6.5 days against a show median of 1.5, and 3 escalations. The 54 metre account has 4 tickets, none reopened, a median close of 1.0 days and no escalations. Same normalised volume band, opposite situations. The first is an account where your delivery failed repeatedly. The second is an account that knows how to use a service desk.

That is the whole argument for looking past the count. Reopen share and time to close measure your failure. Ticket count measures their engagement, and engagement is closer to a positive signal than a negative one.

For a single number, rank every account on each of the four measures, sum the four ranks, and use the sum. Rank-summing avoids arguing about units and stops one wild value on one measure dominating the composite, which happens immediately if you standardise and add. The resulting service score is one input to a renewal intelligence picture and it becomes a work queue only once it is merged with everything else, which is how an at risk call list gets built (G22).

What to do with a high scorer, and why the obvious move is wrong

Here is where I would push back on standard practice. The reflex when an account lights up on service risk is to send a discount, or a free upgrade on next year's space, or a sponsorship sweetener.

Smith, Bolton and Wagner tested what actually recovers a damaged service relationship in the Journal of Marketing Research in 1999, across restaurant and hotel settings. Their finding is that customers prefer recovery resources that match the type of failure they experienced, in amounts commensurate with the magnitude of that failure. A service failure wants a service remedy. Compensating a botched power installation with money is a mismatch, and it reads as a transaction where an apology and a guarantee were wanted.

The exhibition translation is fairly direct. An account with five reopened tickets about the same stand problem wants a named person owning their build next edition and a written commitment about response times. Offering ten per cent off the space is answering a different question, and it also teaches the account that complaints produce discounts, which is a lesson you will regret having taught.

The second correction is about who to contact at all. Ascarza combined two field experiments with machine learning in the Journal of Marketing Research in 2018 and found that the customers carrying the highest risk of churning are not necessarily the best targets for a proactive retention programme. Risk of leaving and responsiveness to an offer are separate properties of an account, and her recommendation is to model how much customers vary in their response to the intervention and target on that, whatever their risk. Her framing of the failure is worth keeping: a retention programme can be futile because the targeting rule is wrong and not because the incentive is.

The account with three escalations and a six-week-old unresolved dispute may already have decided. The account with two reopened tickets and no escalations, who has said nothing, is the one where a phone call changes the outcome. Working out which is which requires holding a control group out and measuring the difference, and uplift modelling for renewal offers has its own treatment (G24).

Where this stops

The whole approach assumes your ticket data is a record of exhibitor experience. On most shows it is a record of what got typed into a system by people under pressure during build.

Three specific distortions are worth checking before you present any of this. Tickets raised on behalf of an exhibitor by a stand contractor may be logged against the contractor. Tickets raised in person at a manned desk during show week often never get logged at all, so your busiest hall can look like your happiest, and fixing that is part of the wider problem of capturing what happens while the hall is open (G15). Tickets closed in bulk at the end of show week to clear the queue produce artificially fast close times and hide reopens, because nobody reopens a ticket after the hall has been struck.

Each of those biases the measure in a direction that flatters somebody, and none of them is visible in the output. The only reliable check is to take twenty accounts, read their full ticket history, and ask the floor manager who covered that hall whether it matches what happened.

There is also a limit on what any of this can predict. Service quality is one input to a renewal decision that also involves the exhibitor's budget, their category strategy, their view of your audience and whether a competitor launched. An account can have a flawless service record and leave because their marketing director was replaced. Service risk earns a place as a component in a broader score, not as the score.

Pull your last edition's ticket export, join it to the exhibitor account and space file, and compute two columns: reopen share and median days to close per account. Sort by reopen share descending and read the top thirty names against your lapsed list from the same edition. If the overlap is meaningful on your show, the rest of this is worth building.