What revenue management for exhibitions borrows from airlines and what it cannot

The hall filled in September. In the second week of November your largest account calls and wants to go from 400 square feet to 900, on the main cross aisle, and will pay whatever it costs. There is nothing there. The last three islands went in August at the standard rate to accounts who would have taken anything.

Somebody in the room says the word yield, and somebody else says that is an airline thing, and both of them are half right. Revenue management for exhibitions does borrow most of its machinery from airlines. One of its assumptions does not transfer, and knowing which one saves you from building a model that produces confident nonsense about booth 1130.

What conditions does the method need?

Talluri and van Ryzin set out the field in The Theory and Practice of Revenue Management, published by Springer in 2004 as volume 68 of the International Series in Operations Research and Management Science. Their treatment splits into quantity based control, which covers single resource capacity control, network capacity control and overbooking, and price based control, which covers dynamic pricing and auctions.

The conditions under which any of it earns its keep are the ones an exhibition satisfies almost perfectly.

Capacity is fixed and known well in advance. You have the hall contract. The number of square feet does not respond to demand, and taking more hall is a separate decision made long before any of this arithmetic applies.

Inventory is perishable on a hard date. A booth unsold at show open is worth zero forever, and there is no next quarter to sell it into.

Demand arrives stochastically over a long horizon, and the willingness to pay of a buyer arriving in November is systematically different from that of a buyer arriving in February. The exhibitor who calls late is usually late because something changed in their year, and something changing in their year is exactly what makes them less price sensitive.

Segments can be separated by a fence that buyers accept. In airlines the fence is a Saturday night stay. In exhibitions it is the booking deadline, the zone line and the priority points standing, all of which your exhibitors already understand and mostly regard as fair. How much the early booking discount should be worth is the question that sets the height of the first of those fences.

Given that CEIR's Index was forecast in May 2026 to grow 2.1 per cent across 2026, measured on net square feet, professional attendance, exhibiting companies and gross revenue over fourteen sectors, the volume growth available to most organisers next year is thin. Yield is where the money is.

The condition that fails

Seats in economy are interchangeable. Any one of them satisfies any passenger holding an economy ticket, which is why an airline can talk about the number of seats to protect for late booking business travellers without caring which seats they are.

Booth 412 and booth 1130 are not interchangeable and no amount of modelling makes them so. An exhibitor who wants a corner on the main cross aisle near the theatre will not take an inline at the back of hall B at any price, and an exhibitor who wants a cheap 100 square feet to qualify for the exhibitor list will not pay for the corner. Your inventory is a set of distinguishable objects with attributes, and demand attaches to the attributes.

This breaks the standard network formulation in a specific way. Network capacity control works by treating an itinerary as a bundle of legs and asking whether the fare covers the displacement cost of each leg. There is no equivalent decomposition for a hall, because the thing being consumed is a named position and its value depends on what is placed next to it, which you also control.

Space as an assignment problem with attributes

The formulation that does work is to treat every sellable position as a row with attributes, and every demand class as a set of preferences over those attributes.

Attributes worth carrying, at minimum: net square feet, number of open sides, distance to the nearest main entrance, distance to the nearest feature or catering area, aisle traffic class, whether the position carries a column or a low ceiling, and which zone it falls in. Every one of those is knowable from the plan before sales opens.

Once positions are rows rather than a single pooled quantity, the yield question becomes tractable again, but at the level of a zone rather than the hall. Within a zone, positions genuinely are close to interchangeable, which is the whole justification for drawing zone lines in the first place. Across zones they are not, and you should stop trying.

So the model is: pool within zone, protect across time, never pool across zones. That is a smaller and more honest object than a hall wide optimisation and it can be built from a floorplan and three years of contract dates.

Protection levels, worked on one zone

Protection is the decision of how much premium stock you refuse to sell early, held back for the buyer who has not called yet.

The rule for the simplest case is old and easy to work by hand. Consider the last unit of premium stock. If you sell it now to an early buyer you receive the early rate for certain. If you hold it, you receive the late rate, but only if late demand reaches that far. Hold the unit while the probability that late demand reaches it, multiplied by the late rate, exceeds the early rate.

Put numbers on a premium corner zone. Your early rate is $36 per square foot. Your realised late rate on the same positions, once the deadline discount has gone and the buyer is in a hurry, is $47. Corner positions run 200 square feet.

The ratio you need is 36 divided by 47, which is 0.766. So you protect corners up to the point where the probability of late demand reaching that many corners is still 0.766 or better.

Now the demand history. Over the last three editions you have taken 24, 18 and 24 late premium corner requests after the early deadline, so call the mean 22 with a standard deviation of about 7. You want the quantity y such that the chance of late demand exceeding y is 0.766. On a normal with that mean and spread, that is roughly 0.73 standard deviations below the mean, so y is 22 minus 0.73 times 7, which is 22 minus 5.1, so 17.

Protect 17 corners. Sell the rest early.

Check the sense of it in both directions. If you protect 25 instead, you are holding 8 more than expected demand, and each corner that goes unsold cost you 200 times $36, which is $7,200 of certain early revenue, to chase an uplift of 200 times $11, which is $2,200. The asymmetry is why the protection level sits below the mean rather than above it. If you protect nothing, which is what most halls do by default, you have sold your entire premium inventory to whoever asked first, and the November call gets an apology.

The published behaviour of large organisers already looks like this even where nobody calls it revenue management. IAAPA runs its Expo space allocation in separate dated phases, with premium space allocated between 26 January and 6 February 2026, an inflatable pavilion allocation on 10 and 11 February, and priority space allocation running from 2 to 20 March. Premium stock is fenced off in its own window with its own rules. That is a protection level implemented as a calendar.

Protection is a quantity control, which is one half of the field. Moving the rate itself as the season runs is the price based half, and it is a different mechanism with different failure modes.

Why does overbooking not port from airlines?

The third leg of quantity based control is overbooking, and it is the piece organisers reach for last and should probably not reach for at all.

Airlines overbook because a no show returns an identical unit to inventory at the moment of departure, and because the compensation for bumping a passenger is a known price. Neither holds here. An exhibitor who cancels in January returns a specific position, not a generic one, and the position they return is rarely the one your waiting exhibitor wanted. A bumped exhibitor is not compensated with a voucher, they call your competitor.

There is a defensible version, which is holding a small contingency of unassigned square feet rather than double assigning positions. That is a different mechanism with a different name and it should be budgeted as a service reserve.

Where this stops

The honest limit is that most of the value here is in the protection level and almost none of it is in the sophistication of the forecast.

Getting from no protection to a defensible protection level per zone is worth real money and takes a spreadsheet. Getting from a normal approximation to a properly fitted arrival process is worth a rounding error, because your zone level demand counts are in the tens, and at that sample size the difference between distributions is smaller than the year to year variation in your category.

The bigger limit is relational. Airline yield management works because no passenger has a fourteen year relationship with seat 14C. Your top accounts have priority points, a history and a view about where they belong, and a protection level that quietly moves a long standing exhibitor off the corner they have held since 2015 will cost you more than the uplift. Revenue management here operates inside a constraint set written by your points system, and the constraint set wins.

There is also a data limit worth naming. All of this needs contract dates as well as contract values, and a surprising number of organisers store the date the contract was countersigned rather than the date the space was committed. If your only date is the finance date, your arrival curve is your invoicing calendar. A commitment date held against each position is the input exhibitor analytics needs before any of this can be computed.

This week, pull the last three editions of contracts for one zone, split them at the early booking deadline, and count how many premium positions were sold before it and what rate each side realised. If the two sides come out at the same rate per square foot, you have no fence, and the protection level is not the first thing to fix.