You take a numbered ticket from the dispenser. The LED screen shows 347. Your ticket says 363. You sit down and wait. You don’t know it, but the system managing your wait — one queue feeding multiple service windows, customers arriving at random intervals, service times varying unpredictably — is governed by the same mathematical framework that routes data packets through the internet, allocates bandwidth in mobile phone networks, and manages runway slots at Heathrow. The post office queue is not just a line. It is a queueing system, and the mathematics that describes it is one of the most consequential branches of applied probability ever developed.

Erlang and the Birth of Queueing Theory

The mathematics of queues was invented not for post offices but for telephone exchanges. In 1909, Danish mathematician Agner Krarup Erlang, working for the Copenhagen Telephone Company, developed the first mathematical models to predict how many telephone lines a switch needed to handle incoming calls without unacceptable delay. His central insight: both arrival rates and service times are stochastic — they follow probability distributions, not fixed schedules. You cannot predict when the next call will arrive or how long it will last, but you can predict the statistical properties of many calls over time and design capacity accordingly.

Erlang’s models introduced concepts that remain foundational: arrival rate (how many customers enter the system per unit time), service rate (how quickly each customer is served), utilisation (the fraction of time servers are busy), and the probability of waiting. These quantities are linked by relationships that hold regardless of whether the “customers” are telephone callers, data packets, hospital patients, or people at counter five holding a parcel too large for the slot.

Why One Line Is Better Than Many

The post office uses a single-queue, multiple-server configuration: one line feeding all available windows. Most people intuitively feel this is fairer than multiple parallel lines (one per window), and they’re correct — but the advantage is mathematical, not merely psychological. In a multi-queue system, you must choose which line to join, and an unlucky choice (a customer ahead of you with a complex transaction) creates delays that other queues don’t share. In a single-queue system, one slow transaction delays only the next customer, not an entire line.

Queueing theory proves this formally. For a system with N servers and a total arrival rate λ, a single queue produces a lower average wait time and a lower variance in wait time than N parallel queues with arrival rate λ/N each. The improvement comes from pooling: idle servers are immediately available to the next customer in line, regardless of which window just freed up. In parallel queues, an idle server is only available to customers in its own queue, even if other queues are longer.

Supermarkets, which traditionally use parallel queues (one per checkout), have increasingly adopted single-queue systems for self-checkout stations precisely because the mathematics demonstrates their superiority. Customers wait less. Throughput increases. Variance decreases. The only cost is the physical space required for the single queue to form without blocking aisles.

Little’s Law and the Universal Queue

In 1961, mathematician John Little proved a result of remarkable elegance: in any stable queueing system, the average number of customers in the system equals the arrival rate multiplied by the average time each customer spends in the system. L = λW. The formula requires no assumptions about the distribution of arrival times, service times, or the number of servers. It is universally true for any queue, anywhere, operating in steady state.

Little’s Law connects three quantities that seem independent but are in fact locked together by a single equation. A post office manager who knows the average number of people in the building (L = 12) and the average time each person spends there (W = 8 minutes) can immediately calculate the arrival rate (λ = 1.5 customers per minute). Or, knowing the arrival rate and the desired wait time, they can calculate how many people will be in the building at any moment and plan staffing accordingly.

The same law governs internet router buffers. A router handling 10,000 packets per second (arrival rate) with an average processing delay of 2 milliseconds per packet (time in system) will have, on average, 20 packets in its queue at any moment. Replace “packets” with “parcels” and “router” with “post office” and the mathematics is identical.

The Utilisation Trap

The most counterintuitive result in queueing theory concerns utilisation — the percentage of time servers are busy. Intuition suggests that high utilisation is efficient: if clerks are busy 90 percent of the time, the office is running well. Queueing theory reveals the opposite. As utilisation approaches 100 percent, wait times don’t increase linearly. They increase exponentially.

In an M/M/1 queue (the simplest model: random arrivals, random service times, one server), the average wait time is proportional to 1/(1−ρ), where ρ is the utilisation. At 50 percent utilisation, the multiplier is 2. At 80 percent, it’s 5. At 90 percent, it’s 10. At 95 percent, it’s 20. A system running at 95 percent utilisation produces wait times four times longer than one running at 80 percent, even though the utilisation increased by only 15 percentage points.

This is why post offices feel unbearable at peak times. At moderate traffic, two or three clerks handle the flow with short waits. At peak, the same number of clerks hits 90-plus percent utilisation, and wait times explode nonlinearly. Adding a single clerk doesn’t just share the load — it drops utilisation from the explosive zone back into the manageable zone, cutting wait times by far more than the proportional increase in staffing would suggest.

From Post Office to Internet Backbone

Every major digital infrastructure operates under the same principles. Internet service providers provision bandwidth using Erlang models. Cloud computing platforms allocate virtual machines using queueing simulations. Call centres staff operators using formulas derived directly from Erlang’s 1909 telephone exchange work. Hospital emergency departments model patient flow using the same arrival-rate and service-time distributions. Air traffic control manages landing slots using priority queueing algorithms that are, mathematically, siblings of the ticket-number system at your local Royal Mail branch.

The mathematics doesn’t care what’s being queued. It cares about rates, capacities, and probabilities. A data packet waiting in a router buffer and a pensioner waiting at window three are subject to the same equations, the same trade-offs between utilisation and delay, and the same fundamental truth: the system that feels frustratingly slow to the individual is, more often than not, performing close to its mathematical optimum for the collective. The queue is not disorganised. It is the organisation — governed by a branch of mathematics that was designed for telephone calls and turned out to govern everything that waits.