The computer industry started as a data-entry backlog at a government agency. The 1880 US census was tabulated by hand and took roughly seven to eight years. The results were not complete until 1887. With the 1890 count approaching and the population growing fast, officials faced a genuinely absurd prospect: a census that would not be finished before the next one made it pointless.
The constitutional stakes were real. Apportionment of the House depends on the count. A census that takes a decade to tabulate does not just produce stale statistics, it produces a representative government working from numbers about a country that no longer exists.
A former clerk who had seen the pile
Herman Hollerith had worked on the 1880 census himself. He knew exactly what the bottleneck was, which is the useful thing about people who have done the work: the problem was not counting, it was that every one of tens of millions of records had to be read by a human, sorted into categories by hand, and tallied by another human, once for every cross-tabulation anyone wanted.
His answer was to make the record machine-readable. Data went onto punched cards, one card per person. A press held an array of metal pins over the card; where a hole had been punched, the pin dropped through into a cup of mercury, completed a circuit, and clicked a counter forward. The hard part was never the counting. It was getting the information into a form a machine could act on, which is still the hard part.
The 1888 trial
The Census Bureau ran a competition, testing three systems on the same real data from St. Louis. Two rival methods captured the data in 144.5 and 100.5 hours. Hollerith's took 72.5.
The gap that mattered showed up in the second phase. Sorting that captured data into categories such as age, sex, and race took the rival systems 44.5 and 55.5 hours. Hollerith's system did it in 5.5.
That ratio is the whole lesson. Getting the data in was somewhat faster. Doing anything with it afterward was transformed, because the expensive part of manual processing is not the first pass, it is that every new question requires another pass through the entire pile.
What it actually delivered
On the real 1890 census, the initial population total was announced in about six weeks. The full tabulation ran roughly 63 million punched cards and finished in under three years, against the seven to eight the 1880 count had taken by hand, with an estimated $5 million saved. The output was not just faster, it was richer: 25 volumes of cross-tabulated data, because once the cards existed you could ask new questions of them cheaply.
Hollerith started the Tabulating Machine Company in 1896. In 1911 it merged into the Computing-Tabulating-Recording Company. In 1924, CTR was renamed International Business Machines. The punched card remained the dominant data-entry medium for another fifty years, and the whole architecture of commercial computing grew out of a machine built to clear a filing backlog.
The part that keeps repeating
Notice what nobody proposed in 1888. Not one of the three competing systems was “hire more clerks.” That option was on the table, it was cheap, and it was obviously available. It was rejected because the previous census had proved that adding people to a manual process buys you a linear improvement against a problem growing geometrically.
The same shape shows up in Wall Street's 1968 paperwork crisis and in the 1968 resolution that created the paralegal. One of the three chose to remove the step. Two chose to add people, and are still adding people.
Today's pile looks different and behaves identically. It is the PDF that has to become fields in three systems, the intake form retyped into a matter record, the statement reconciled by hand every month. The reason it has resisted automation is not that the counting is hard. It is that the information arrives in a form no machine could act on, which is precisely the problem Hollerith solved by changing the form of the record rather than the speed of the clerk.
The difference in 2026 is that you no longer have to redesign the record to get there. The work can be captured from how a person already does it and run from that, unattended, on the documents as they actually arrive.
The rest of the series
Four times a profession ran out of people to absorb its own routine work, and what actually ended each one.
- Who Decided Six Minutes?. One managing partner started timesheets around 1920 to run his firm better. Then the measurement became the price.
- The Year the Stock Exchange Closed on Wednesdays. For most of 1968 the busiest market on earth shut one day a week because the back office could not clear the paper.
- The Paralegal Was Invented to Make Legal Services Cheaper. A 1968 ABA resolution created a role to absorb a lawyer's tedious and routine detail. The role became billable.