The Overbooked Flight Is a Calculation, Not a Mistake

Perishable Capacity Changes the Arithmetic

A seat on a flight a hotel room reserved for a specific night and a table reserved for 7 p.m.m.of a Friday share something that most products don't: once the moment passes the unsold unit cannot be stored marked down or sold tomorrow. Its value drops to zero. Permanently. That's the whole problem in one sentence and almost everything else in this article simply solves its consequences

Here's the part that makes it worse. The flight costs about the same to operate whether the last seat is full or empty. The plane consumes about the same fuel the crew gets paid the same salary and the landing fee at the other end doesn't change. Therefore the contribution margin of the last seat sold is close to the total fare since almost none of that fare is consumed for additional cost

Now add the fact that reservations are canceled or simply abandoned. A significant proportion of booked passengers never show up historically in the several percent range and considerably higher on fully refundable tickets where cancellation costs the passenger nothing. Sell exactly as many seats as you have and you're guaranteed to fly some of them empty. When I first sat down to address it I found this to be the most counterintuitive part of the whole issue: the cautious conservative option selling only what you physically haveIt is not actually the safest. It is the choice that guarantees a loss

The Decision Is a Cost Comparison

Overbooking clearly defined is the deliberate selling of more reservations than physical seats. It sounds reckless until you see how the level is actually set: by comparing two costs against each other seat by seat

The sweet spot is where the expected cost of selling one more reservation is equal to the expected benefit of selling it. This is a marginal comparison not a guess and it is worth considering why it almost always favors the sale of at least some additional seats. The loss of empty seats is certain and total: zero forever. The additional cost by contrast is limited known in advance and only occasionally triggered. Something limited and probabilistic tends to beat something certain and total at least to some extent. The exact place whereWhere that point is located depends on the route the fare mix the day of the week and the no-show history of the specific flight which is exactly the kind of narrow repeatable forecasting problem that airlines are good at

A Worked Example: Where the Extra Seat Stops Paying

Let me flesh out that marginal comparison with illustrative numbers created for this example rather than drawn from any specific carrier's data. Let's say a flight has 100 seats a fare of $300 and a denied boarding compensation cost of $900 per denied passenger. All three are round numbers chosen to make the arithmetic easy to verify not figures reported by any actual airline

An airline doesn't know exactly how many booked passengers will no-show on a given flight but it has flown this route often enough to have a distribution: a probability table of how many will no-show. Suppose the history of this route looks like this

Multiply each no-show count by its probability and add the results treating "7 or more" as 7 for simplicity and the expected number of no-shows comes out to 3.5. On a 100-seat flight sold with exactly 100 reservations that is 3.5 seats are expected to travel empty on a typical day which is equivalent to 3.5 times $300 or $1,050 in lost revenue. That expected loss is the reasonmain reason why overbooking exists

Now ask the marginal question one seat at a time: Is reservation number 101 worth selling? Because the example is a non-refundable fare the airline keeps the $300 regardless of whether that specific passenger shows up or not so the profit from selling one more seat is a certain $300. The cost only appears if the sale of that seat brings the flight to a situation where more passengers appear than seats which happens exactly when the numberactual number of no-shows is less than the number of additional seats sold

Selling the first extra seat only causes a problem if it turns out there are no no-shows which the table shows happens 5 percent of the time. Expected cost: $900 times 5 percent or $45. Compare that to a certain $300 in income and it's an easy yes worth $255 on average. Push harder and the arithmetic quickly gets tighter

The "Probability of Hitting" column does all the real work so look closely. It's the cumulative probability straight from the first table of no filings equal to or less than the count that would leave the newest extra seat without a real seat behind it: zero or less one or less two or less three or less which is 5 percent 15 percent 30 percent and 50 percent. Multiply each by the costof $900 and compare it to the fixed revenue of $300 and the answer will range from yes to no exactly between the third and fourth extra seat. Sell 103 reservations versus 100 seats not 104. The third extra seat still averages $30. The fourth loses $150

That crossover point where marginal cost first exceeds marginal revenue is full optimization. No one at an airline chooses an overbooking level by feel. They're running exactly this kind of table with a real no-show distribution built from years of flight history instead of my own made-up one and they find the seat where the sign changes from positive to negative

Denied Boarding Is Regulated, Which Is What Makes It Safe To Model

Here's the thing that turns this from a chaotic gamble into a solvable equation: The cost of guessing wrong is mostly set in advance by regulation before the airline puts a price on anything

US regulations divide denied boarding into two categories. Volunteer Boarding is denied when a passenger accepts compensation in cash or vouchers to give up a seat and take a later flight. Involuntary Denying boarding is the airline choosing someone and that compensation is set by rule as a multiple of the passenger's fare scaled based on the length of the resulting delay and subject to limits. European rules work differently in mechanical terms but the same in spirit: a fixed compensation amount linked to the distance of the flight under a regulation that also covers long delays and cancellations more broadly

Because the maximum limit of the involuntary cost is known in advance the airline can value the entire risk in approximately the same way that an insurer values a policy. The auction that sometimes takes place at the gate in which agents increase the offer of the voucher in steps until enough passengers volunteer is a live price discovery mechanism.do nothing and the airline finds the most economical way to cover the shortfall

Case Study: United Express Flight 3411

The clearest real-world lesson in the limits of this model is United Express Flight 3411 a regional flight from Chicago to Louisville in April 2017. It's worth being precise about what really happened because the popular version of the story gets one detail wrong: This was not strictly speaking a classic case of no-show overbooking. The flight was fully boarded when United needed seats four back for its own crew members who had to reposition themselves to operate a flight.different from the next morning. No one had failed to show up. The airline simply needed seats it had already sold

United offered vouchers for volunteers the same auction described above. Not enough passengers took up the offer so the airline moved to an involuntary selection and chose four passengers based on its own criteria. One of them a doctor named David Dao refused to leave the plane. Aviation security officers physically dragged him down the aisle and other passengers filmed it with their phones. The video went viral within hours

The financial mechanics of the incident were exactly what a compensation model would predict: limited known and survivable. It was the reputational cost that broke the model's assumptions. United stock reportedly lost more than $1 billion in market value at its worst point in the following trading session an amount that dwarfs what the airline could save after years of carefully optimizing overbooking on that single route. The airline's CEO initially defended the removal in a statement widely interpreted asdeaf costing the company another news cycle before the company changed course and apologized more fully. United subsequently settled with Dao for an undisclosed amount and along with peers across the industry substantially increased the maximum voluntary compensation it would offer to volunteers reportedly up to $10,000 in United's case and rewrote its own policies on removing already seated passengers

What this case really teaches is more limited and useful than "overbooking is bad." The failure was not the pricing model. The mathematical compensation for four displaced seats was almost certainly a rounding error compared to what that flight would have lost if it had been seated with four empty seats on a bad no-show day. The real failure was assuming that the cost of an involuntary ejection stops at the compensation check when a single phone camera can turn an entry decision into an unlimited cost.some

The Same Math Shows Up Everywhere Capacity Expires

Hotels carry out the same activity and even have their own verb to describe it: walking a guest which means paying for a comparable room at a different property plus the taxi ride and any incidentals that might smooth it out. That cost per incident tends to be higher than an airline voucher so hotels overbook more conservatively than airlines. Same fringe logic just calibrated for a more costly failure mode

Restaurants face the same problem of no-shows with almost no equivalent to a shock auction so most solved it in a completely different way: passing the cost on to the customer up front through a credit card hold or a flat cancellation fee rather than absorbing the risk themselves. Healthcare has been slower to adopt either solution. Part of that is structural and part of that I think is that "we overbooked your appointment" reads very differently when the product is your knee surgery atinstead of your seat to Denver

Overbooking Is Downstream of Who Bought the Ticket

Overbooking never occurs in isolation. It sits within a broader revenue management system and derives its main input from the allocation of fare classes. A fully refundable ticket carries a much higher no-show probability than a deeply discounted non-refundable ticket since cancellation costs the refundable passenger nothing. A flight sold primarily to business travelers with flexible fares can absorb much more aggressive overbooking than the same plane sold primarily to leisure travelers who paid weeks in advance and cannot get a refund.cent

That's the real reason why the same plane on the same route is overbooked differently on a Tuesday morning departure than on a Saturday afternoon departure. The plane hasn't changed. The passenger mix has and the no-show distribution I built in the worked example above changes along with it

Where the Model Breaks

All of the above treats overbooking as a clean optimization problem and on the financial side it really is. But the model is optimizing a spreadsheet. It's not about optimizing customer trust and those aren't the same goal although the spreadsheet silently behaves as if they were close enough

The United Express case is the clearest example not because the compensation calculations were wrong but because the model had no line item for a video that went viral and the story ran for a week straight. A denied boarding event that is left inside the gate resolved with an upset passenger and a voucher is exactly as cheap as the regulations imply. A denied boarding event that becomes national news is not a larger version of the same cost. It is a different kind of cost onewith no natural ceiling manifested in falling stock prices in the attention of Congress in a brand association that lasts years longer than the news cycle

The awkward part for anyone who likes this model as much as I do is that you can't fully value tail risk up front the same way you can value regulated compensation. You can raise your voluntary bid cap after the fact which the industry did. You can't know before it happens what specific gate decision on what specific afternoon goes viral. The math says overbooking to the third seat and stopping. Mathematicians have never had to sit on a plane and watch a strangerbeing dragged down the hallway with someone's camera phone. A model that only prices the parts of the world covered by regulation will be quietly wrong about the parts that aren't and may seem correct for years until the moment it isn't

There is a more subtle version of the same problem that never appears in the news. Every voluntary bump every delay at the gate every passenger who saw someone ahead of them in line denied boarding spends a small amount of goodwill that never shows up in any income statement. The spreadsheet treats each increase as a bounded independent event. Frequent fliers experience it as a pattern and a pattern is exactly what leads someone to quietly hire a competitor next time for reasons never explained.They trace back to the revenue management department that actually caused them

How I'd Actually Use This Framework

My read is that overbooking is one of the clearest real-world examples of expected value reasoning you'll find in any industry and that's exactly why I like to use it to teach myself the concept. This is not investment advice it's just a statement about what businesses are fun to model. If you want to test whether you really understand marginal cost versus marginal revenue try building the table from the example above with your own made-up numbers before looking at mine. If you can't get the crossover point to move withgood sense when the compensation cost increases you still don't understand the model. I didn't either the first two times I tried it

I think this framework holds its own not in predicting whether any specific airline's stock will rise next quarter. It's a lens for detecting perishable inventory problems everywhere not just in travel. Concert tickets cruise cabins rental cars in a specific lot on a specific day even seats in a popular college class that always has a few students who sign up and never show up. Anywhere a company has a tough deadline and a predictable no-show rate the same marginal comparison applies whetherWhether anyone involved calls it overbooking or not

What you would really look at if you were analyzing an airline as a business rather than as a statistical exercise is not the overbooking rate itself. Carriers don't report that clearly and I wouldn't trust a single quarter even if they did. You'd look at complaints of involuntary denied boarding per passenger for several years since regulators publish that figure because an upward trend there is the first sign that a revenue management team has started to lean more heavily than before on the airline side.acceptable model losses. That's a cultural signal disguised as a statistic and I'd give it more weight than most people

I'll also say clearly where I might be wrong. It's possible that the reputational risk I spent the last section on is less than I'm making it seem and that United's stock rally in a few months is the best evidence not the initial drop. Markets are quick to forgive often faster than the news cycle suggests they should. If that's right the spreadsheet model is closer to being correct on its own terms than my counterargument believes and the viral incident is aflat tax rather than proof that the whole approach is fragile. I really go back and forth on which of those two readings is the more honest

The Bottom Line

Overbooking is a rational response to perishable capacity and unreliable support and it works because a compensation framework puts a known limited price on failure. Run the marginal comparison as the example above does and the optimal overbooking level comes out of arithmetic rather than a guess. It is defensible when the compensation is genuine and volunteering remains optional. It stops being defensible the moment the airline chooses someone which is exactly the line drawn by the regulationand exactly the line that United Express Flight 3411 crossed in the worst possible way. The most profound lesson generalizes to airlines of the past to any company that sells something that expires: the empty unit is a total and certain loss and almost any limited cost is worth accepting to avoid. What the model still cannot clearly price is the unlimited cost the one that appears when a single phone camera turns a routine input decision into the reason people remember the brand. Get the spreadsheetcorrect and you still won't have fully understood the whole deal