For a recent project, Telegenisys was asked to take tax records from counties all over the United States and enter these into a spreadsheet. Naturally each county has their own formats and codes. The one thing we were certain that we would not do is to have human agents read the data from the PDFs and type them into XLS.
The problem with having humans reenter data is that humans remember things. They try to apply the last county’s formats to the current data, they repeat the last row which partially matched to the current row, they transpose digits, and so on. Now we could apply a second level of agents to check the work of the first, but they’d also make mistakes.
Machines don’t remember what you don’t tell them to remember, but this also means that computers have no common sense and will output garbage without thinking about it, because they do not think.
Our answer is to have humans define patterns and then let the computers apply the patterns for data conversion. While human hands never touch the rows of data, human eyes examine the inputs for patterns, define the patterns, check the error logs, and spot check the output. The result is higher efficiency and fewer errors, because the agents do not reenter the data. Thus data entry needs human eyes, not human hands