What the cash machine did to bank tellers

The story everyone tells, and the mechanism nobody names

The lesson you reused at the start of this module gave you the cash-machine story in outline: tellers rose for decades after the machines arrived, then fell. This lesson is about why, because the why is a rule you can apply to any trade, and the rule has two halves that the cheerful version and the gloomy version each tell only one of.

The numbers

In the United States the number of bank tellers roughly doubled between 1970 and 2010, from around three hundred thousand to around six hundred thousand, while the number of cash machines went from zero to four hundred thousand. A machine that did the teller’s core task (handing out cash) coincided with twice as many tellers.

An economist who studied this worked out the mechanism. A cash machine made a branch cheaper to run, because it needed fewer tellers per branch: about thirteen fell to about ten. Cheaper branches meant banks opened more of them (over forty per cent more urban branches in the following decades) to compete for customers. More branches, at ten tellers each, outnumbered fewer branches at thirteen. The teller’s job also changed, from counting cash to selling products and handling problems, which is what tellers had time for once the cash was handled by a machine.

Then, from about 2010, online banking made the branch itself unnecessary. Branch numbers fell and tellers with them, by around a third since the peak. The same occupation, the same technology trend, two opposite effects in sequence.

The rule

Whether automating a task grows or shrinks the occupation around it depends on one quantity: how much more of the output people buy when it gets cheaper.

If demand stretches (if halving the cost of a thing means people buy far more than twice as much) then automating part of the work grows the total work, and the occupation grows even as each worker does less of the automated task. If demand does not stretch (if people want roughly the same amount at any price) then the automated share is gone, and the occupation shrinks by that share.

The cash machine’s first phase was the stretching case: cheaper branches, many more branches. Its second phase was the other: once people no longer wanted branches at any price, no amount of cheapness helped.

The spreadsheet

The same rule, in an office. After the electronic spreadsheet arrived in 1979, the number of bookkeeping and accounting clerks in the United States fell by several hundred thousand over the following decades, because the arithmetic they did by hand was now instant. Over the same period the number of accountants, auditors and financial analysts rose by more than that. Analysis became so cheap that firms wanted vastly more of it: scenarios, forecasts, models that had been unthinkable when each one took a clerk a week. Demand for analysis stretched. Demand for arithmetic did not. One occupation absorbed the other’s loss and then some, and they were not the same people.

The loom

The oldest case. In the nineteenth century power looms let a weaver produce many times more cloth per hour. For decades the number of weavers rose, because cloth became cheap enough that ordinary people bought many times more of it. Then, once everyone had all the cloth they wanted, demand stopped stretching, the productivity gains kept coming, and weaving employment collapsed. Growth, saturation, decline: the same three-act shape as the tellers, over a longer run.

Applying it to now

For any trade you care about, ask: if the language-shaped tasks in it became nearly free, would the world want vastly more of the output, or about the same?

Translation is the hard case. If cheap translation means a billion documents that were never translated now are, the occupation grows. If it means the same documents translated by a machine and checked by fewer people, it shrinks. The freelance data in the next lesson suggests the second so far, at the low-value end, and the first has not yet shown up. Software is the hopeful case: there is more software the world wants than has ever been written, and cheaper code may mean far more of it. Customer support is the gloomy one: nobody wants more support tickets at any price.

The limits of the rule

The rule says what happens to the occupation. It says nothing about the person. The clerks who lost their jobs did not become the analysts who gained them; the weavers did not become the mill’s engineers. Over a generation an economy finds its new shape, and the people displaced were the ones who paid for the finding. That is the reused lesson’s point, and the rule sharpens it: growth in an occupation is a statement about a category, and you are not a category. And the rule assumes the past pattern (a task automated, then a decade or two of adjustment) while the honest uncertainty this time is whether the breadth and speed give the adjustment its usual decades.