Every business has a version of this. Someone opens two windows, reads from one, types into the other, and does it again. Nobody has ever measured it, because it is not a project — it is just Tuesday.
It is usually the most expensive thing in the building.
Why it stays invisible
Rekeying does not appear anywhere. It has no budget line, no owner, no project code. It is distributed across people who each think it is a small part of their week, which is true individually and false in aggregate.
It also does not feel like a problem, because nothing is broken. The data does arrive. It just arrives slowly, through a person, and slightly wrong some fraction of the time.
The moment it becomes visible is when that person is ill, or leaves. Then everyone discovers the process was a person.
What it actually costs
Four things, and only the first gets counted.
The typing. Straightforward: instances times minutes. This is the number people quote.
The switching. Every interruption to rekey something costs more than the rekeying, because getting back into the previous work takes minutes that never appear on any timesheet.
The errors. Transcription has an error rate. It is not zero and it is not stable — it rises with volume, tiredness and time pressure, which is to say it is highest exactly when the stakes are highest.
The delay. Data that moves through a person moves at human speed. If a decision waits on the transfer, the cost is the waiting, and that one can be very large indeed.
The error cost is the one that bites
A typing mistake in an internal spreadsheet is annoying. The same mistake in a payment reference, a customer record, a stock figure or a regulatory return is a different category of event.
And these errors are hard to find, because the system did what it was told. There is no exception, no alert, no failed job — just a number that is wrong and looks exactly like a number that is right. It surfaces weeks later, in a reconciliation, or in a complaint.
What replacing it looks like
Less than people expect. The two systems usually already have a way to talk — an interface, an export, a file drop. Connecting them is a well-understood piece of work, measured in days rather than months, and the specification writes itself because the manual process is the specification.
The parts that need care are the unglamorous ones: what happens when a record does not match, what happens when the source is malformed, and who gets told. Get those wrong and you have replaced a slow reliable human with a fast silent failure, which is worse. Get them right and the transfer becomes a thing that happens rather than a thing someone does.
How to find yours
Ask this at your next team meeting: what do you type into one system that you just read out of another?
You will get more answers than you expect, quickly, and with visible feeling. People know exactly which part of their week this is. Then apply frequency, error cost and hours to what comes back, and the ranking will be obvious.
The one at the top has probably been there for years. It is normally the cheapest significant improvement available to a business of any size — and if you want the ranking done properly with the numbers attached, that is a Reality Check.
This is the part we do — the crossing from a demo to a system that survives production.