From mail contracts to SABRE

The first question anyone asks when they open an airline reservation system is always the same: how hard can it be to keep a seat and a passenger in one table, and why is all of this such a mess?

The answer is a single seat on a mail plane in 1925.

The passenger was a by-product

The 1925 Kelly Act ended the Post Office monopoly and opened airmail to private carriers. The point of the flight was to carry mail; exactly one seat was set aside for a paying passenger. A traveller called the departure city, and if the seat was free, it was booked.

The business logic matters here. The aircraft’s payload was optimised for cargo, so passenger inventory was the residue. In modern terms it is a fixed allotment: the physical space is the limit. And since revenue came from a government mail contract, yield did not exist as a concept. Price was not something you optimised — it was a number written in a contract.

Who knows the inventory

Early inventory control was not centralised. The true state of the seats was known by the station in the city the aircraft departed from. Before making a booking, a sales agent had to call that station and get confirmation: request and reply. The answer went onto a PNR card and moved by teletype.

It was correct, and it was slow. Every sale cost a phone call.

The fix arrived in Boston in 1939: sell and report. Agents sold freely without asking, up to a threshold of fullness; once the threshold was crossed a “stop sale” message went out and the system fell back to the slow, safe method.

Read that in today’s vocabulary and it is obvious what happened: instead of taking a synchronous lock on every sale, they moved to eventual consistency.¹

The efficiency argument is also the modern one: management by exception. You start the message traffic only near the limit, not on every transaction.

The buffer seat: when sync is slow, you hide inventory

Post-war traffic grew while the process stayed old. The Boston Reservisor (1946) was the first machine to replace the card files; the Magnetronic Reservisor (1952) went in at LaGuardia and held ten days of data for a thousand flights, queryable by several people at once.

But the machine solved less than half the problem. The agent and the operator were still on two ends of a telephone, and more importantly: the seat sold and the passenger name record were matched by hand.

The business rule that came out of this will sound familiar. Because inventory and record could not be synchronised in the moment, the last few seats were closed to sale and a buffer was kept. That was how oversales and denied boarding were managed.

A buffer is not free. It means flying with empty seats — the industry calls it spoilage.²

If the fare is fixed, revenue management cannot exist

From 1938 until the late 1970s the Civil Aeronautics Board ran the industry at the micro level: which carrier flew which route, what a ticket on it cost, whether a merger was allowed.

It is odd to find this in a technical chapter, and it is decisive. If the tariff is fixed, dynamic pricing is not a problem you have. Whenever you book, the seat costs the same. Yield management could not exist until deregulation removed the CAB’s pricing control — not because the technology was missing, but because the rule forbade it.

Worth remembering when you look at what a system does not solve: some features are not missing, they are prohibited.

The real idea: bind inventory to the customer

In 1953 C.R. Smith, the CEO of American Airlines, and a young IBM salesman named R. Blair Smith sat next to each other on a flight. The machine Blair Smith described would not just hold availability; it could record the passenger’s name, their itinerary, even their phone number.

Until then, airline computers were calculators for inventory counts. The idea was, at bottom, a relational database idea: bind a numeric inventory entity to an alphanumeric customer entity. The modern PNR was born there.

An IBM that moved from hardware into software purely to serve one airline’s requirement is also the start of the airline IT vendor model. Amadeus, Sabre, Travelport — all descendants of that decision.

The technical foundation came from the military. SAGE, the air defence system built to track aircraft, turned out to be the right infrastructure for tracking seats.

SAGE became IBM’s ACP; ACP became TPF. TPF still runs inside the big GDSs today — a sixty-year-old operating system, still in production.³

A decision made in 1961 is still standing

SABRE’s architecture left five modules behind: schedules, inventory, PNR, ticketing and DCS. Each answers its own question — where and when do we fly, how many seats can be sold, who is the passenger, what was paid, who boarded.

Here is the part to notice: PNR, ticketing and inventory are separate modules. That was decided in 1961 and we are still paying for it. An e-ticket and a PNR can drift out of sync because the architecture gave birth to them apart. IATA’s ONE Order initiative is an attempt to undo exactly that split: collapse the PNR, the ticket and the EMD into a single retail order.

Sixty years later, we are trying to reverse a modularisation decision.

What it leaves us

Three things, and none of them are specific to aviation.

When synchronisation is slow, you hide inventory. The buffer seat is not incompetence, it is the invoice for latency. If you hold safety stock in your own system you are paying the same invoice; the question is not “how do I reduce the stock” but “why can’t I make the sync faster”.

The constraint is not always technical. What was missing in the CAB era was not an algorithm, it was permission. Before you explain what a system fails to do, ask what it was allowed to do.

Boundary decisions made early live longest. Separating inventory from the record was the right call in 1952; undoing it in 2026 takes an international standards initiative. What you split today stays split.

Footnotes

1. Nobody called it that in 1939, but that is what it was: instead of guaranteeing correctness on every transaction, you wait for a message that tells you when it is wrong. The name arrived forty years later. Availability is assumed correct until a message says otherwise. The idea is still in the field — the AVS messages in legacy GDS distribution are the grandchildren of that stop sale. ↩

2. Spoilage and overbooking are two sides of one coin: flying empty versus selling too many. Revenue management is the job of choosing a point between them. The one-sentence definition of modern revenue management comes from here: minimise spoilage while managing overbooking correctly. Both are invoices for the same synchronisation problem. ↩

3. The longest-lived piece of software in this chapter. It was born in an air defence system and it is still selling tickets. ↩