People are not jobs#
SLOP2710 · Week 11
Three things jobs never do#
Balk — see the queue, don’t join
Renege — join, then leave
Game — find the rule, optimise against it
A queue with heavy balking
looks exactly like a queue with low demand.
And reneging is what stabilises an overloaded queue#
Week 2’s table has no row for ρ > 1. Real overloaded systems still reach an equilibrium.
The length of an overloaded queue measures its customers’ patience, not its capacity.
You install an accurate live wait estimate.#
What happens to its accuracy?#
- W answered
- —
- announced on joining
- —
- hung up
- —
- answered an hour
- —
- W if nothing is said
- —
60% improvement on the dashboard#
38% failed to get help, before and after#
Counting the people who left#
Erlang’s formulas assume callers wait as long as it takes.
Erlang A adds one parameter — patience — and changes what every other number means.
Which gives you a trap#
More agents → wait down, abandonment down
Better hold experience → abandonment down, measured wait up
A system that makes waiting feel shorter
has removed its own pressure
to make waiting be shorter.
What should the announcement say?#
Nothing — maximises the most expensive feature of a wait
A position — honest, verifiable, uninterpretable
A time — most useful, and it eats itself
So announce conservatively.#
Credibility is the asset.#
Gaming, properly#
Scalping is a queueing attack — FCFS defeated by being better at coming first
The resale price measures what the original rule left on the table
Three responses#
Harden — an arms race the defender loses on cost
Randomise — gives up arrival-order fairness, buys immunity to speed
Price it — week 8; capture the value instead of burning it
Prefer rules robust when gamed#
over rules optimal when obeyed#
Three questions for your brief#
- What does somebody gain by misreporting, and how would you know?
- Who can see your queue, and what do they do when they see it?
- Who leaves, and would you find out?