“Can it read the handwriting in our 1994 archive?”
A real question, and why the honest answer is usually “some of it”.
This gets asked more often than anything about the cloud. Somewhere in the building there is a room, or a corridor of cabinets, and in it are years of paper that nobody has time to type up and nobody dares to throw away.
The honest answer is: some of it. Typed pages from the nineties come through cleanly. Forms with printed labels and handwritten fields come through in pieces, the labels perfect and the answers uneven. A full page of cursive from a tired clerk on a Friday is a coin toss, and the machine will not tell you which side it landed on unless you ask it to.
That last part is the whole job. The tools that read paper are good now, and they are also confident, and those are two different things. The question we help a team answer is not whether it can read the archive. It is what happens to the pages it reads wrong, and who notices.
In practice this means a small pilot on one cabinet, a person checking a sample every week, and a plain rule about what gets trusted and what gets a second look. Ten weeks later the team knows which decade of the archive is safe to search and which still needs a human hand. That is a smaller promise than “digitise everything”. It is also one that holds.